AI-Enabled Battery Storage and Electric Vehicle Flexibility for Power-Quality-Constrained Photovoltaic Integration: A Critical Review

Abstract

High photovoltaic (PV) penetration in distribution networks can be limited by voltage rise, unbalance, harmonics, thermal loading, and converter-current constraints before the available solar resource is fully utilized. Battery energy storage systems (BESS) and electric vehicle (EV) flexibility can relieve some of these limits, but their contribution depends on battery state, degradation, converter headroom, mobility, communications, and point of common coupling requirements. This critical review asks when artificial intelligence (AI) improves the use of stationary and mobile battery flexibility under those constraints. The synthesis compares these engineering dimensions qualitatively and includes a source-level comparison of 30 selected sources, together with focused comparisons of battery technologies and AI constraint-enforcement methods. Across the reviewed evidence, fast reactive and waveform-support services are often constrained by converter current more directly than by battery energy, whereas energy shifting and vehicle-to-grid operation are strongly affected by state of charge, availability, efficiency, and aging. AI is most defensible as a forecasting, estimation, and supervisory layer; it does not replace electrical feasibility or battery safety limits. Usable PV energy, meaning energy delivered after curtailment, conversion losses, and battery constraints, is not the same quantity as a larger nominal hosting-capacity rating. The review therefore recommends reporting net recovered PV energy together with compliance margin, converter utilization, battery throughput, aging assumptions, uncertainty, fallback behavior, and validation maturity.

1. Introduction

Distribution systems are becoming active renewable-generation platforms rather than passive delivery networks. In photovoltaic (PV)-rich feeders, the practically relevant quantity is not installed PV nameplate capacity alone but the portion of available solar power that can be exported or locally consumed without violating voltage, thermal, unbalance, harmonic, protection, or converter constraints. Classical hosting-capacity studies established that feeder topology, impedance, and voltage-control design make distributed-generation penetration strongly location-dependent [1,2]; later work showed that harmonic distortion, background waveform distortion, and frequency-dependent network impedance can also become binding constraints [3,4,5]. These mechanisms connect power quality (PQ) directly to renewable-energy utilization because each binding constraint can translate into curtailment, additional storage cycling, flexible-demand rescheduling, or network reinforcement.

Battery energy storage systems (BESS) and electric vehicles (EVs) make the PV operating boundary time-dependent. Stationary BESS can absorb excess solar energy, discharge later, and provide fast active or reactive support, while controllable EV charging and vehicle-to-grid (V2G) operation can align demand with midday PV production, alter feeder peaks, and contribute converter-based voltage support [6,7,8,9,10]. The electrical capability of these resources is not equivalent to continuously available flexibility, because state of charge (SOC), state of health (SOH), battery temperature, degradation, converter current, round-trip losses, charger connection, departure-energy requirements, communication availability, and user participation all restrict the service that can be delivered.

Artificial intelligence (AI) can improve forecasts, battery-state estimates, and sequential dispatch under uncertainty. Its outputs remain subordinate to converter ratings, feeder limits, SOC/SOH bounds, protection, and fallback control. Battery-management reviews document progress in learning and digital twins [11]; recent studies extend learning to battery–grid coordination and physically constrained charging [12,13]. The question is whether these methods improve usable PV energy under the same physical and lifetime constraints as conventional control.

The central research question is whether stationary and mobile batteries increase usable PV energy after the applicable point-of-common-coupling (PCC), feeder, converter, battery, mobility, and deployment constraints are enforced. Four subsidiary questions make that comparison explicit: (i) which electrical or PQ mechanisms become binding as PV penetration increases; (ii) which BESS or EV services can relax each mechanism, and at what cost in battery duty and converter headroom; (iii) when AI-enabled forecasting, estimation, or control add value over deterministic alternatives under the same physical constraints; and (iv) what validation evidence is required before a reported gain can be transferred from simulation to deployment. This distinction matters because a strategy that raises installed hosting capacity while increasing annual curtailment, conversion loss, cycling, or battery wear may deliver less renewable energy than a smaller nominal hosting increase evaluated over representative operating conditions.

Existing reviews cover EV-related voltage, harmonic, and loading effects [14,15,16]; PV–storage architectures, voltage regulation, and energy management [17,18,19,20]; and broad AI-enabled grid-storage applications [21,22]. Table 1 compares their organizing questions with the present synthesis. Here, reported gains are compared through the disturbance, actuator, battery duty, compliance boundary, and validation level. Battery-aging and V2G evidence [23,24] constrains the transferable claim: a larger hosting rating does not establish greater PV delivery after losses, cycling, and electrical limits are accounted for.

Table 1. Scope of the present review relative to recent reviews. The final column identifies what each review does not treat as a primary organising variable; it does not assert that the topic is absent.

Reference	Primary Focus	Key Complementary Strength	Not Treated as a Primary Organising Variable
[14]	EV charging, grid interaction, PQ, and stability	Detailed treatment of charger-related electrical effects and mitigation	PV-energy utilization and stationary battery duty are not the main organizing outcomes
[15]	Voltage regulation in low-voltage networks with renewables and EVs	Strong voltage-regulation and coordinated-control perspective	Harmonic, thermal, battery-lifetime, and multi-timescale storage constraints require joint treatment
[16]	Broad EV impacts and mitigation in distribution networks	Broad EV integration and operational coverage	Renewable-energy utilization and battery-service cost are not the primary synthesis variables
[17]	PV–energy-storage voltage regulation	Direct PV/BESS voltage-control coverage	EV availability, V2G, harmonic constraints, and deployment evidence lie outside the principal boundary
[18]	Hybrid energy storage integrated with renewable generation	Battery-focused review of hybrid energy storage system (HESS) configurations, control, optimization, applications, and PQ-related renewable-integration motivations	Does not center PV–EV/V2G–BESS distribution feeders, simultaneous PCC constraints, usable PV energy, or deployment-evidence maturity
[19]	PV–battery grid-forming configurations and coordinated control	Direct treatment of PV/BESS architecture and grid-forming control	EV/V2G availability, multi-PQ constraints, usable PV energy, and deployment maturity are outside its principal review boundary
[20]	BESS optimization in buildings and energy communities	Battery-focused synthesis of applications, optimization, SOC, and degradation	Distribution-feeder PQ, PCC harmonic constraints, EV mobility, and PV hosting are not the main synthesis axes
[21]	AI-driven optimization of grid-scale BESS	Systematic treatment of AI methods, techno-economic performance, environmental effects, and policy/regulatory issues for grid-scale storage	Does not center distribution-feeder PCC constraints, EV/V2G availability, simultaneous PQ mechanisms, or usable PV energy as the principal outcome
[22]	Artificial intelligence (AI)-enabled BESS for renewable energy	Broad synthesis of machine learning (ML), deep learning, reinforcement learning (RL), battery management, degradation, PV/wind, microgrids, and smart grids	The synthesis narrows AI to PQ-constrained usable-PV recovery with stationary/mobile battery flexibility, hard feasibility, and validation boundaries
[11]	AI and digital twins for lithium-ion battery management	Detailed SOC/SOH/remaining useful life (RUL), lifecycle, cloud–edge, and intelligent battery management system (BMS) synthesis	Does not organize the evidence around PCC power-quality limits, EV/V2G availability, and usable PV energy in distribution feeders
[23]	V2G applications and battery cycle aging	Direct battery-lifetime perspective for mobile grid storage	Does not jointly organize stationary BESS, PV curtailment, multi-PQ constraints, and validation maturity
[24]	Rechargeable batteries for grid-scale energy storage	Broad 2026 assessment of battery technologies, performance, safety, and grid applications	Distribution-feeder PQ, EV flexibility, and usable PV energy are not the central system-level synthesis axis

Filip et al. quantify the effect of EV/BESS deployment on LV PV hosting [25], while Khunkitti et al. optimize BESS/PV placement and capacity in an EV-integrated feeder [26]. These studies establish the combined system boundary. The present review examines how concurrent PQ limits, battery condition, conversion losses, and validation affect interpretation of such gains. Usable PV energy is the system outcome; battery energy, current capability, SOC/SOH, temperature, degradation, and availability determine the flexibility that can support it.

The review evaluates BESS and EV technologies by their effects on PV hosting, curtailment, compliance margin, net usable solar energy, battery throughput, and lifecycle limitations. A reported improvement in voltage, harmonics, cost, or hosting is therefore interpreted only after the feeder boundary, regulated quantity, converter headroom, battery duty, vehicle availability, control assumptions, and validation platform are identified. The four subsidiary questions stated above provide the structure for Section 3, Section 4, Section 5 and Section 6, so the discussion moves from the physical constraint to the available flexibility, then to control and validation rather than presenting technologies as independent catalogues.

Table 2 compares selected system studies, battery experiments, and reviews used in the synthesis. The entries illustrate distinct engineering claims and validation boundaries; they are not a representative sample of the bibliography.

Table 2. Selected sources informing the critical synthesis.

Source	System and Objective	Battery and Control Representation	Validation Level	Reported Findings and Scope Limitations
[25]	Low-voltage (LV) distribution network; effect of EV and BESS deployment on PV hosting	Stationary and mobile storage flexibility in a feeder-level hosting study	System-level numerical study	Establishes that EV/BESS deployment changes PV hosting; does not by itself provide a cross-study accounting of PQ mechanisms, battery aging, and net delivered PV energy.
[26]	EV-integrated distribution feeder; coordinated BESS/PV placement and sizing	Optimization of location and capacity	Optimization/feeder simulation	Establishes that placement and sizing materially affect the combined PV–EV–BESS boundary; transfer depends on feeder model, objectives, and battery assumptions.
[20]	Buildings and energy communities; BESS optimization and grid services	SOC- and degradation-aware storage optimization	Review synthesis	Provides a strong battery-operation and degradation perspective; distribution-feeder PQ and PCC harmonic constraints are outside its main comparison boundary.
[21]	Grid-scale BESS; AI-enabled techno-economic and environmental optimization	Broad AI-method and grid-storage synthesis	Systematic review	Establishes the breadth of intelligent grid-storage optimization; it does not use distribution-feeder PQ constraints or usable PV energy as its principal comparison outcome.
[22]	Renewable-energy BESS applications across PV, wind, microgrids, and smart grids	ML, deep learning, RL, battery management, and degradation	Comprehensive review	Establishes a broad 2026 AI–BESS taxonomy. The present review therefore does not claim novelty from AI coverage itself; its narrower contribution is the PQ-constrained energy-accounting and validation boundary.
[27]	Distribution-network Volt–Var control	Model-augmented reinforcement learning with a safety layer	Feeder simulation	Illustrates constraint correction of learned dispatch; battery lifetime and full PCC compliance are not demonstrated.
[28]	V2G cycling using real-world usage profiles	Cell-level lifetime response to V2G duty	Experimental battery study	Supplies direct evidence that V2G duty and cycling assumptions matter for lifetime claims; feeder-support value still depends on charger availability and network constraints.
[29]	PV charge allocation in multi-battery energy systems	Physics-informed neural-network control	Intelligent-control study	Supports physically constrained battery allocation; it is battery/control evidence rather than direct proof of distribution-feeder PQ compliance.
[30]	PV–wind laboratory microgrid with battery–supercapacitor storage	Hybrid storage transient buffering	Laboratory experiment	Demonstrates experimental transient-buffering behavior and links storage dynamics to AC-side performance; laboratory scale does not by itself establish utility deployment maturity.
[31]	MW-class zinc–iron flow battery source–grid–load–storage coordination	SOC estimation and coordinated power allocation	Large-system model	Extends the evidence beyond lithium-ion systems and shows the importance of internal battery state; chemistry, service objectives, and feeder context limit direct transfer to PV-rich LV networks.

2. Review Scope and Analytical Framework

This narrative critical review synthesizes sources selected for their engineering relevance to PQ-constrained PV integration with stationary and mobile batteries. Heterogeneous feeder models, battery duties, compliance boundaries, and validation methods preclude a statistical meta-analysis or a universal ranking of controllers. The bibliography is an expert-selected evidence base; it is not the included-study cohort of a systematic review.

2.1. Source Selection and Scope

Sources were selected for their contribution to one of four questions: the electrical mechanism limiting PV utilization, the battery or EV service available to relieve it, the control constraints governing that service, and the evidence supporting transfer to operation. Selection considered methodological detail, relevance to the distribution/PCC boundary, and the distinction between modeled, laboratory, and field outcomes. Standards and original project reports were used for compliance definitions and deployment evidence; battery experiments were used for aging and ripple claims. Reviews supplied context rather than independent validation of a controller.

The evidence base includes foundational studies predating 2015 and targeted recent publications beyond June 2026. These sources are retained when they establish a physical mechanism, measurement definition, experimental battery response, or directly relevant system comparison. Publication dates therefore do not follow the narrower window of the supplementary OpenAlex check. Historical database query logs were not retained, so the cited literature cannot be assigned a retrospective database-specific screening flow. IEEE Xplore, Scopus, Web of Science, and ScienceDirect are consequently not reported as a reproducibly searched database set. Pure materials studies without a grid-interface implication and unrelated applications are outside the review scope.

2.2. Source Appraisal and Supplementary Coverage Check

Supplementary Tables S1 and S5 record the system boundary, reported outcome, validation level, and limits of use of selected sources discussed in the main comparisons. These purposive assessments support interpretation of specific claims; their proportions do not estimate the validation distribution of the literature. Supplementary Table S3 profiles the references actually cited in the current manuscript.

An independent OpenAlex coverage check was executed on 2 October 2026 using eleven title-field queries. The archived API requests retrieved publications dated 1 January 2015–16 September 2026; the reported comparison subset retains records dated through 30 June 2026. This publication cutoff is distinct from the execution date. A rule-based Python 3.13.13 script applied metadata filters and keyword expressions to available titles and abstracts, assigning one primary classification code per record. It classified the 10,487 records remaining after metadata triage automatically; they were not independently assessed by human reviewers. The terms, rules, stage counts, and record-level outputs are supplied in Supplementary Table S4.

The narrow title queries and automated rules provide a supplementary check of topical coverage, not the selection process for the bibliography. Their small overlap with cited sources cannot establish completeness or representativeness, and an automated pass is not full-text eligibility or inclusion. The coverage-check counts are therefore kept in the supplement and are not used to infer how many studies support the review’s conclusions. Selected-source appraisal and the illustrative comparisons remain separate from this automated register.

2.3. Analytical Synthesis Framework and Evidence Hierarchy

Table 3 defines the dimensions used for qualitative comparison. Table 4 applies them to ten illustrative sources, recording the binding constraint, battery or EV duty, validation, and outcome used in this review. The source assessments in Supplementary Tables S1 and S5 provide further claim-specific detail.

Table 3. Analytical dimensions used for qualitative evidence synthesis.

Dimension	Extracted Attributes	Engineering Significance for Renewable Integration	Representative Anchors
System and PCC boundary	Voltage level, phases, grid strength, location, and monitoring point	Device-level improvement does not establish feeder hosting; impedance and phase coupling govern transfer	[1,2,4]
Binding PQ constraint	Disturbance, metric, bandwidth, and assessment interval	The binding mechanism determines whether P/Q control, phase balancing, filtering, storage, or reinforcement is relevant	[3,5,32,33]
Battery duty	Power, SOC/SOH, temperature, throughput, ripple, and reserve	Energy shifting changes throughput; fast support also depends on converter headroom and DC-side buffering	[6,28,34,35]
EV availability	Connection, bidirectionality, departure energy, SOC, and compensation	Charger capability and user constraints determine whether scheduled flexibility is actually available	[7,8,23,36,37]
Control architecture	Local, centralized, or distributed control; update horizon	Comparisons require matched resource limits and adequate measurement/communication speed	[38,39,40,41,42]
AI/data-driven role	Forecasting, state estimation, scheduling, and action correction	Prediction/reward gains require a demonstrated effect on feasible dispatch, PV delivery, or battery duty	[11,12,13,43]
Converter and resource limits	P/Q/current envelope, harmonics, thermal derating, efficiency, and shared reserves	Nominal ratings overstate flexibility when concurrent services share current and energy headroom	[40,44,45,46,47]
Validation maturity	Simulation, real-time test, laboratory, pilot, or field operation	Planning feasibility, fast dynamics, interoperability, and operational readiness need different evidence	[3,48,49,50]
Compliance basis	Applicable document, point, metric, and aggregation method	Equipment emissions, installation allocation, and supplied-voltage characteristics are distinct boundaries	[32,51,52,53,54]
Reported renewable outcome	Hosting, curtailment, PV-origin delivery, losses, and battery throughput	An instantaneous voltage or capacity gain does not establish net delivered PV energy	[2,6,9,10,55]
Transferability boundary	Feeder impedance, climate, chemistry, charger, market, and connection code	Results remain conditional on their electrical, resource, and institutional setting	[50,56,57,58]

Table 4. Application of the analytical dimensions to ten illustrative sources.

Source	Binding Boundary	Battery or EV Duty	Validation Used	Renewable Outcome Used
[25]	LV feeder hosting	EV and BESS flexibility	Feeder simulation	Hosting change, not net annual PV energy
[26]	EV-integrated feeder	Placement and capacity	Optimization study	Siting effect, not a universal energy gain
[20]	Building and community BESS	SOC and degradation	Review	Battery operation, not feeder PQ
[21]	Grid-scale BESS	AI optimization	Systematic review	Techno-economic scope, not usable PV energy
[22]	Broad AI–BESS	Management and degradation	Comprehensive review	Method coverage, not PQ compliance
[27]	Distribution voltage	Safe reactive dispatch	Feeder simulation	Constraint satisfaction, not battery lifetime
[28]	Cell lifetime	V2G cycling	Cell experiment	Lifetime response, not feeder energy
[29]	Multi-battery allocation	Physics-informed control	Intelligent-control study	Constrained allocation, not feeder compliance
[30]	Laboratory microgrid	Hybrid storage buffering	Laboratory experiment	Transient behavior, not field hosting
[31]	MW flow-battery system	SOC and power allocation	Large-system model	Coordination result, not LV transfer

The evidence hierarchy is claim-dependent, not a universal score. Time-series studies can support annual energy and planning claims when feeder data and uncertainty are credible, while fast voltage, harmonic, resonance, protection, and converter claims may require electromagnetic-transient (EMT), controller-hardware-in-the-loop (HIL), power-hardware-in-the-loop, or prototype evidence. Pilot and field demonstrations add interoperability, operating-procedure, communication, customer, and institutional evidence that simulations cannot supply. This hierarchy is used to moderate the strength of conclusions rather than to rank technologies simply by the highest validation label reported in a paper.

3. System Boundary and Resource Roles in PV-Rich Networks

Figure 1 adopts a renewable-generation-centered boundary rather than treating PV, EVs, and BESS as parallel topics. PV is the generation resource whose usable output is to be maximized; EVs and BESS are time-varying electrical interfaces that can consume, shift, inject, or condition power; and the feeder/PCC imposes simultaneous compliance constraints. EV and storage technologies are included only to the extent that they change the amount, timing, and quality of PV electricity that can be accommodated by the distribution network.

Figure 1. Renewable-generation-centered framework showing PV, BESS, and EV/V2G resource interactions, AI supervisory functions, and the binding electrical, battery, and mobility constraints. Abbreviations: AI, artificial intelligence; BESS, battery energy storage system; DRL, deep reinforcement learning; EV, electric vehicle; MPC, model predictive control; OPF, optimal power flow; PCC, point of common coupling; PV, photovoltaic; RL, reinforcement learning; RUL, remaining useful life; SOC, state of charge; SOH, state of health; V2G, vehicle-to-grid.

3.1. PV Generation and Feeder Hosting Constraints

Distribution-level PV is predominantly interfaced through power-electronic inverters, so the usable renewable-energy output depends on both the available irradiance and the admissible converter/network operating region. At low penetration, a grid-connected PV inverter is often represented as a controlled current source operating close to unity power factor. That simplification becomes inadequate when reverse power flow, voltage rise, weak-grid synchronization, converter-current saturation, or harmonic interaction becomes binding. Reported impacts also vary with phase connection and feeder topology, particularly where single-phase PV changes local voltage and phase balance [59,60]. Classical hosting-capacity studies already showed that feeder location and network voltage control can change the amount of generation that can be connected [1]; annual-duration approaches further demonstrated that installed PV capacity and annual usable PV energy are not equivalent when curtailment is required to respect voltage limits [2]. For a renewable-energy assessment, the relevant quantity is the solar energy delivered within the feasible electrical envelope rather than nameplate penetration alone.

Most existing distribution PV operates in a grid-following mode. The inverter synchronizes to the PCC voltage, commonly through a phase-locked loop (PLL), while maximum-power-point tracking and DC-link control determine the available active-power command. In strong feeders this architecture is well established, but in weak feeders the interaction among synchronization dynamics, current limits, feeder impedance, and protection can become important. Grid-forming (GFM) control is consequently receiving attention because it can provide a voltage-source-like response and improve voltage/frequency support, yet weak-grid stability cannot be inferred from a single benchmark or from an unconstrained converter model [61]. A GFM inverter remains bounded by semiconductor current, DC-side energy, thermal limits, filter design, and ride-through/protection requirements. Virtual-synchronous-generator studies provide an additional converter-control baseline for damping and grid-support behavior [62], while PV converter research shows that DC-side power-decoupling design also shapes the realizable operating envelope [63]. These limits are central when the objective is to preserve PV generation during stressed operating conditions.

Voltage-management research has progressed from fixed power factor toward autonomous Volt–Var/Volt–Watt characteristics and coordinated active/reactive control. Local reactive-power control can reduce overvoltage without continuous communication, and feeder studies demonstrate the sensitivity of its benefit to network impedance and inverter capability [64,65]. Coordinated curtailment, PV–storage energy management, and high-penetration PV feeder models further show that the selected control boundary changes the apparent hosting benefit [66,67,68,69,70]. Hierarchical and coordinated schemes can allocate reactive power, curtailment, tap actions, and storage more selectively across a feeder [38,41]. Their principal advantage is system awareness; their principal weakness is dependence on network models, measurements, communications, and a defined fallback state. Reactive support is also not free. Near maximum PV output, converter apparent-power limits reduce the available Q headroom unless the inverter is oversized or active power is curtailed, and additional reactive current can increase feeder loss and converter temperature. These constraints explain why coordinated storage and inverter studies must be interpreted with their assumed tap control, feeder model, and device rating rather than through voltage deviation alone [6,9,10].

The PV hosting boundary is not purely a fundamental-frequency voltage problem. Experimentally validated work shows that pre-existing harmonic voltage distortion can reduce hosting capacity [3]; studies on high-PV urban and distorted distribution networks further show that harmonic propagation depends on phase allocation, background distortion, converter spectra, and frequency-dependent feeder impedance [4,5]. A PV installation can satisfy a local device test while the aggregated feeder still approaches a PCC distortion limit. Source-attribution methods become relevant when customer injections and upstream background distortion coexist [71]. These findings justify treating harmonic headroom, resonance, and measurement bandwidth as renewable-integration constraints rather than as post-design PQ checks.

Taken together, recent PV control developments reveal a recurring trade-off. Autonomous inverter functions are scalable and fast but have limited knowledge of system-wide constraints; coordinated controls can use network sensitivity more efficiently but introduce communication and model dependence; storage can recover curtailed energy but consumes finite SOC and lifetime; and active filtering can improve waveform quality while competing for converter current. Each method is evaluated by the additional usable PV energy it can enable after these physical limits are represented, not by voltage improvement alone.

3.2. Electric-Vehicle Demand and Mobile Battery Flexibility

EV integration changes the PV operating envelope through demand timing, converter emissions, phase loading, and bidirectional flexibility. The first effect is temporal. Daytime workplace or destination charging can absorb local PV generation and reduce reverse power flow, whereas synchronized evening charging can increase feeder peak demand, voltage drop, transformer loading, and losses. Load models, smart-charging studies, and office-charging analyses show that charging coincidence, connection location, and user behavior must be represented explicitly when estimating these impacts [72,73,74,75]. Early PV–V2G studies showed that a parking facility with rooftop PV and bidirectional vehicles can enhance distribution capacity or reduce peak demand when the charging window overlaps the renewable-generation profile [7,36]. These benefits are conditional on vehicle presence, charger rating, driver energy requirements, and the location of the feeder constraint.

The second effect is converter-related. EV chargers are nonlinear power-electronic loads whose current spectrum is not fixed across all charger technologies or operating points. Modern active-front-end chargers can achieve substantially better power factor and lower low-order distortion than passive rectifier interfaces, but feeder-level PQ remains a function of aggregation, grid strength, background voltage distortion, and charging state [76,77]. Measurements demonstrate that harmonic current distortion can change with battery SOC during charging [46]; probabilistic feeder studies likewise show that EV and nonlinear-load uncertainty changes the distribution of PQ outcomes rather than producing one deterministic distortion value [33]. For residential networks, single-phase charger allocation also makes phase unbalance and neutral loading relevant even when three-phase aggregate power appears acceptable.

Smart charging uses this demand flexibility without requiring energy export from vehicle batteries. By shifting charging away from a constrained interval or toward local solar production, scheduling can reduce coincident demand and improve voltage conditions [73]. V2G extends the controllable range by permitting active-power export and, where the charger permits, reactive-power support. Aggregator formulations coordinate many vehicles to provide grid services while respecting charging-cost or availability objectives [8,78,79]. Real-time voltage-control studies use active and reactive EV flexibility to reduce voltage deviations [80,81,82,83,84], while distributed MPC exploits reactive V2G to coordinate multiple resources without relying on one central optimization step [39]. Disturbance-observer MPC and Volt/Var characteristic design continue this trend toward faster converter-aware support [42,85].

The electrical capability, however, is not equivalent to firm flexibility. A vehicle may be disconnected, the user-required departure SOC may be binding, a bidirectional charger may not be available, or additional battery throughput may be unacceptable without compensation. Distribution-system impact reviews and orderly-charging studies identify connection uncertainty, charging coordination, and network operating conditions as essential modeling inputs [86,87,88,89,90]. Planning and scheduling studies further show that network reconfiguration, stochastic demand, multiobjective dispatch, and coordinated charging/storage decisions alter the feasible service window [91,92,93,94,95,96]. Infrastructure-planning and solar-charging studies expose the conversion and siting trade-offs of mobile storage [97,98]. Lifecycle and second-life studies add chemistry, repurposing, and degradation constraints [99,100,101,102], while dispatch studies show how service schedules determine battery throughput and delivered grid value [103,104,105]. The renewable-energy measure is the additional PV energy absorbed or delivered, or curtailment avoided, after accounting for charging/discharging loss, battery throughput, mobility constraints, and the timing of the network bottleneck. This distinction prevents V2G from being credited with renewable-hosting value when it only shifts energy at high lifecycle cost.

Experimental evidence supports the technical feasibility of converter-based EV support but also illustrates the validation gap. A three-phase off-board charger prototype has demonstrated practical grid-voltage regulation [48], and active harmonic-rejection control demonstrates that charger converters can participate in waveform support [40]. Field projects now extend the evidence beyond single-device laboratory work: the Australian Realising Electric Vehicle-to-Grid Services (REVS) trial involved 51 EVs connected through bidirectional chargers [50], and Project Sciurus deployed more than 300 domestic V2G units in the United Kingdom (UK) [106]. These demonstrations establish that aggregated V2G is no longer purely a simulation concept, while also exposing interoperability, customer participation, metering, communication, and jurisdiction-specific market constraints. For PV-rich feeders, EV flexibility is best represented as a bounded, probabilistic resource, not as guaranteed storage.

3.3. Stationary Battery Storage: Energy Shifting and Grid Support

BESS is the most direct bridge between renewable-energy time shifting and fast converter control. It can absorb PV energy during export-constrained periods, discharge later, provide reactive power, reduce voltage excursions, and coordinate with PV inverters and feeder regulators [6,9,10]. Storage technologies differ materially in response time, power density, usable energy, cycling behavior, and balance-of-plant requirements; these differences determine whether a resource is suited to fast PQ support, energy shifting, or neither [107,108,109]. Battery storage is the storage class emphasized because it is explicitly coupled to PV and EV operation in the target system.

At the feeder level, a BESS can support at least three different time scales. Fast converter control can provide reactive-power or waveform support without large energy exchange. Minute-scale active-power control can smooth ramps and relieve a temporary voltage or thermal constraint. Hour-scale scheduling can shift midday solar energy into later demand periods. These services compete for the same converter rating and, for active-power services, the same SOC window. Reserving apparent-power headroom for reactive or harmonic support can reduce simultaneous charging/discharging capability; reserving SOC for contingency support can reduce the amount of midday PV that can be stored; and aggressive active-power regulation can increase battery throughput and degradation.

This coupling is often hidden when studies report only an increase in hosting capacity. A storage controller may permit a larger installed PV nameplate but still curtail a substantial fraction of annual solar energy, or it may recover that energy at the cost of frequent cycling and conversion losses. A stronger evaluation reports the MW/MWh rating, SOC bounds, converter apparent-power limit, charging/discharging efficiency, degradation representation, priority among P/Q/PQ services, and the fraction of otherwise curtailed PV energy that is actually recovered. Sizing and dispatch studies often optimize losses, revenue, or a single service; their results cannot be treated as PQ-support evidence unless the converter and network constraints are also represented [110,111,112]. Distributed storage studies show that coordinated operation with transformer or feeder voltage control can mitigate PV-driven voltage rise [9,10], while coupled PV/BESS control can coordinate reactive support and energy storage [6]. Broader PQ-oriented storage control has also been used to address voltage, power-factor, and renewable/EV-related network issues [113].

The broader evidence separates three sources of storage value. EV and grid-integration reviews distinguish mobile and stationary storage by infrastructure, service, and interoperability requirements [114,115,116,117,118]. Renewable and hybrid-storage reviews instead emphasize technology selection and the power/energy requirements of the target service [119,120,121,122,123]. Forecasting and renewable-smoothing studies show that dispatch value changes with forecast quality and hybrid-resource configuration [124,125,126]. Planning and optimization studies make the result conditional on the objective, uncertainty model, and network constraints [110,111,112]. Recent battery-system evidence further supports realistic SOC windows, cycling depth, C-rate, temperature, and aging instead of treating BESS power as a cost-free control input [47].

A useful renewable-energy indicator is incremental usable PV energy per unit of storage throughput together with the avoided network constraint. Active-power cycling changes stored energy and battery duty. Reactive or waveform support uses converter current and can also impose DC-side ripple, depending on phase configuration and power decoupling. Its battery impact therefore requires the interface model described in Section 3.4.

3.4. Battery Chemistry, Degradation, and Interface Architecture

Battery chemistry affects the sustainable power and energy envelope, but fast PQ response is delivered by the converter and its controller. Comparing chemistries by response time alone confounds electrochemical capability with inverter bandwidth. Table 5 therefore distinguishes repeated active-power duty from reactive or harmonic support, which can require substantial AC current with relatively little net battery-energy exchange. The comparisons are qualitative: cell design, temperature, SOC window, pack sizing, and manufacturer limits prevent a universal numerical ranking [24,34].

Table 5. Relative suitability of battery technologies for converter-mediated PQ support and energy shifting.

Technology	Fast PQ and Active-Power Duty	Energy Shifting	Principal Qualification
Lithium iron phosphate (LFP)	Suitable for repeated short active-power duties when current and thermal limits are respected; reactive/harmonic service remains inverter limited	Established stationary-storage option where cycling and thermal robustness are priorities	Lower energy density than nickel-rich cells can increase pack size; flat open-circuit voltage complicates SOC estimation; temperature and charging limits still apply [11,24]
Lithium nickel manganese cobalt oxide (NMC)	High energy density supports compact EV packs; sustained high-current duties require chemistry-specific thermal and aging assessment	Suitable for mobile storage and space-constrained installations	EV energy density does not establish economical stationary cycling; high SOC, temperature, and user reserve restrict V2G duty [24,28]
Lithium titanate (LTO)	Power-oriented cells can suit repeated active-power pulses and intensive cycling	Lower energy density can make bulk energy shifting less attractive where space and cost dominate	LTO denotes an anode material rather than a unique cathode chemistry; evaluate the complete cell, rated power, and pack cost [18,34]
Redox flow batteries	Fast electrical support requires an appropriately rated converter; repeated active support also depends on stack and electrolyte operating conditions	Independent stack-power and tank-energy sizing favors longer-duration stationary services	Pumps, auxiliary demand, electrolyte condition, and stack efficiency must enter net-energy accounting; not a mobile V2G resource [24,31]
Lead–acid	Short support is possible, but repeated deep cycling and high-rate duty require careful derating	Applicable where mature technology and initial cost outweigh weight and lifetime limitations	Depth of discharge, partial-SOC operation, temperature, and replacement frequency can dominate lifecycle cost [34]

Degradation models enter the comparison at three levels. Throughput or equivalent-cycle models assign a marginal wear cost to energy exchange and are convenient for dispatch but cannot distinguish two schedules with the same throughput and different SOC, temperature, or dwell time. Semi-empirical calendar-plus-cycle models capture these dependencies using fitted stress factors and, where appropriate, cycle-depth counting; their coefficients are cell- and duty-specific. Electrochemical or reduced-order physical models can represent internal limits and mechanisms more directly, at a greater identification and computation burden. Data-driven SOH models support online updating but inherit the coverage and uncertainty of their training data. A grid-service study should state its model, calibration domain, calendar-aging treatment, capacity-fade and resistance-growth measures, and how those measures change allowable current and usable energy [11,20,23].

The distinction has measurable consequences. Darikas et al. tested commercial 4.5 Ah lithium-ion cells using charger-derived V2G profiles at 5, 25, and 40 °C. Their comparison estimated approximately 0.62 full equivalent cycles per day for shallow 30–50% SOC cycling at a capacity-fade rate comparable to prolonged high-SOC storage. This is a cell-specific comparison with a high-SOC calendar-aging baseline, not a general wear-free V2G allowance [28]. In a separate simulation, Chen et al. reported up to 32% lower transient overshoot, 25–40% shorter settling time, and 12–23% lower peak battery-current stress at 60% SOH relative to non-SOH-aware control. These are modeled control-performance gains, not measured lifetime extensions [35]. Such results support including battery condition in control; they do not justify a universal degradation penalty or guaranteed service life.

Table 6 compares four experimental studies of battery aging under V2G or superimposed ripple duty; NCA denotes nickel cobalt aluminum oxide. The control simulation by Chen et al. is discussed above as a control-performance result and is excluded from this experimental table.

Table 6. Experimental battery-aging evidence relevant to V2G and converter ripple.

Source	Reported Conditions	Reported Aging Observation	Limit of Use
Darikas et al. [28]	Commercial 4.5 Ah cells; charger-derived V2G profiles; 5, 25, and 40 °C; shallow 30–50% SOC cycling	About 0.62 full equivalent cycles per day, with a fade rate comparable to prolonged high-SOC storage	Cell-level comparison against a high-SOC calendar baseline; not a feeder lifetime guarantee
Dubarry et al. [127]	Commercial graphite/NCA cells; driving, charging, and additional grid-discharge schedules	Additional grid-discharge cycling accelerated degradation relative to the tested driving/charging schedules	Schedule- and chemistry-specific; neither a universal V2G penalty nor evidence about reactive-only service
Uddin et al. [128]	Fifteen cells; 1200 cycles; DC duty with superimposed AC at five frequencies	Frequency-dependent capacity/impedance degradation and greater cell-to-cell variation were reported	Ripple spectrum and test duty govern transfer; not a generic loss factor for every inverter
Bessman et al. [129]	28 Ah NMC/graphite cells; 21 A peak ripple at 1 Hz, 100 Hz, or 1 kHz; about 2000 cycles in one year	About 80% capacity remained; no additional capacity or power fade attributable to the tested ripple was detected	The measured capacity loss is total aging, not ripple-induced loss; results are confined to the tested cells and amplitudes

The battery management system (BMS) places an additional boundary on grid commands. Pack-level SOC and SOH estimates may conceal cell imbalance, local temperature, sensor bias, and unequal resistance. A dispatch optimizer must therefore receive the available charge/discharge limit, rather than assume the nameplate rating remains available. SOC-estimation uncertainty is particularly relevant for LFP voltage plateaus, while balancing, thermal derating, contactor protection, and charging restrictions can make an otherwise feasible feeder action unavailable. Learned state estimation complements these protections; it does not authorize overriding them [11].

Temperature closes part of the same practical gap in low-voltage feeders. Ambient heat, enclosure design, and pack thermal resistance can force the BMS to reduce charge or discharge current before the electrical SOC window is exhausted. A controller that treats the nominal SOC limits as always available can therefore overstate both usable energy and fast PQ support during hot operating periods. The appropriate report is the thermally derated power together with the SOC window, not a single nameplate rating.

PV–BESS topology also changes the available actuator. In an AC-coupled installation, PV and storage typically have separate grid interfaces; their apparent-power limits and losses must be represented separately. In a DC-coupled installation, PV and storage exchange power through a common DC network, often with a shared grid inverter. This can permit DC-side PV charging without an intermediate AC conversion, but grid export, reactive support, and harmonic compensation compete within the shared inverter rating. DC–DC converter limits and DC-bus control remain relevant. Consequently, identical battery capacity does not imply identical curtailment recovery or PQ capability across AC- and DC-coupled systems [18,19]. Comparisons should identify the topology, conversion path, shared ratings, and auxiliaries before attributing a gain to chemistry or AI.

Phase configuration is material even when mean active power is zero. For sinusoidal single-phase voltage and current, with the voltage defining the cosine reference, instantaneous AC power is

$$p\left(t\right)=P+P\cos \left(2\omega t\right)+Q\sin \left(2\omega t\right),$$

(1)

where P and Q are average active and reactive power and $\omega$ is the fundamental angular frequency. The $2\omega$ component must be buffered by the DC link, a decoupling circuit, or the battery interface; reactive-only operation can therefore impose battery ripple despite negligible mean energy transfer. Balanced sinusoidal three-phase power cancels this component across phases, although unbalance, distortion, and switching can reintroduce ripple. An LFP storage experiment examined twice-grid-frequency ripple through temperature, balancing, and round-trip efficiency [130]; power-decoupling designs change how much pulsating power reaches the DC source [63].

Battery ripple cannot be assigned a universal aging penalty. The contrasting experiments in Table 6 show that chemistry, ripple amplitude and frequency, DC bias, SOC, temperature, and interface filtering matter [128,129]. An additional heating estimate is $P_{\mathrm{ripple},\mathrm{loss}}\approx {\sum }_f{I}_{b,f,\mathrm{rms}}^2\mathrm{Re}\left\{Z_b\left(f\right)\right\}$, where $Z_b\left(f\right)$ is battery impedance at frequency f. This estimates electrical dissipation rather than capacity fade. PQ-support studies should report the battery-side spectrum and thermal response, and validate aging under representative duty instead of inferring it from AC-side reactive energy alone.

3.5. AI-Enabled Battery and EV Coordination

AI has a defensible role in this system boundary when it improves information, prediction, or sequential decision making that would otherwise be represented by fixed forecasts, static battery parameters, or inflexible scheduling rules. It does not enlarge converter current, feeder thermal capacity, or battery energy by itself. An AI-enabled architecture is technically meaningful only when the learned component operates inside explicit electrical and battery constraints and when a deterministic fallback remains available if data, communications, or the learned model become unreliable.

Recent reviews already cover AI-driven grid-scale BESS optimization and broad AI-enabled BESS functions across battery management, state estimation, fault diagnosis, predictive maintenance, thermal management, renewable integration, microgrids, and smart grids [21,22]. Replicating those broad taxonomies would not address the narrower grid-facing question considered here: AI is considered only when it changes the scheduling, observability, or robustness of stationary and mobile battery flexibility under an explicit PQ-constrained PV-integration objective. This distinction also separates the grid-facing synthesis from battery-management surveys centered on SOC/SOH/RUL prediction and cloud–edge digital twins [11].

Forecasting and battery-state learning can improve the information supplied to a constrained scheduler. Reinforcement learning instead selects sequential actions, as illustrated by mobile-storage routing [43]. Learning-assisted system scheduling and physics-informed battery allocation operate at different boundaries [12,29]; electrochemical charging models add battery-safety constraints [13].

For the present application, the distinction between AI-enabled and AI-controlled operation is material. AI can rank candidate actions, update forecasts, estimate hidden battery states, or adapt controller parameters while a projection, MPC/OPF, or rule-based layer enforces network voltage, converter current, SOC/SOH, mobility, protection, and interconnection limits. Comparative studies need to disclose the training and test domain, input observability, uncertainty treatment, inference or update time, hard action constraints, battery-aging model, communication assumptions, fallback action, and out-of-distribution tests. The relevant outcome is not prediction accuracy or cumulative reward in isolation, but the additional usable PV energy or compliance margin obtained after losses, battery throughput, degradation, and violation risk are accounted for.

In supervisory operation, learned forecasts or state estimates supply a constrained scheduler. In direct learned control, the policy proposes actions and a projection or safety layer corrects them before execution, as illustrated by distribution-voltage studies [27,131]. The comparison must identify which component enforces feasibility and how it responds when no admissible correction exists.

Method Taxonomy and Constraint Enforcement

AI functions differ in both their decision role and their safety burden. Supervised forecasting estimates PV, load, and vehicle availability for a separate scheduler. Battery-state learning updates SOC, SOH, or degradation parameters. Reinforcement learning (RL) maps operating states to sequential decisions, while hybrid methods couple a learned policy or forecast to model predictive control (MPC) or optimal power flow (OPF). Physics-informed learning penalizes inconsistency with physical equations during training. These functions require different evidence: forecast accuracy must be connected to dispatch benefit, and dispatch benefit must be evaluated under the same battery, converter, and network limits as the baseline.

Constraint enforcement cannot be inferred from the word “safe” or from a reward penalty. An action projection solves for an admissible command closest to the proposed action. A safety layer corrects actions using a modeled or learned constraint response, whose validity depends on model error and update delay [27,132]. A constrained Markov decision process (CMDP) instead places limits on accumulated or expected costs during policy optimization [133]; satisfaction of those limits does not generally guarantee every instantaneous voltage, current, or temperature limit. A physics-informed loss likewise encourages consistency without necessarily enforcing all inequalities at execution. Table 7 separates these approaches and the evidence needed for PQ-sensitive operation; Table 8 compares representative applications and their validation limits.

Table 7. AI decision functions and approaches to constraint enforcement.

Approach	Role	Constraint Mechanism	Qualification for PQ Operation
Forecasting or state estimation	Inputs to MPC/OPF or rule-based dispatch	Feasibility resides in the downstream controller	Report forecast/state uncertainty and resulting compliance, energy, and battery-duty effects [11]
Action projection	Correct a learned command before execution	State-dependent feasible set; potentially coupled network and device inequalities	Validate the set against nonlinear/unbalanced operation, stale measurements, and infeasibility; simple power clipping is insufficient
Safety layer	Correct actions using estimated constraint response	Analytical or quadratic-programming correction	Safety is conditional on constraint-model accuracy and remaining feasible action; retain deterministic fallback [27,131,132]
CMDP/constrained policy optimization	Learn under explicit cost budgets	Expected or cumulative constraint costs	Report per-step violations separately; expected-cost satisfaction is not a conformance test [133]
Hybrid RL–MPC	Combine adaptive decisions and predictive feasibility	Explicit controller constraints plus learned objectives/actions	Disclose coupling, horizon, solve time, aging assumptions, and fallback; feasibility depends on the predictive model and constraint set
Physics-informed learning	Improve physical consistency	Conservation or state-equation residuals in training	Residual accuracy does not certify hard voltage/current or BMS limits at execution [29]

Table 8. Representative AI studies: reported function, validation, and limits of the claim.

Source	Method and Boundary	Reported Evidence	What Is Not Established
Kou et al. [131]	Deep deterministic policy gradient with safety layer; distribution voltage control	Numerical studies on modified IEEE systems; voltage constraints and power-loss reduction	Field operation, battery-aging benefit, and harmonic/PCC compliance
Gao and Yu [27]	Model-augmented RL with quadratic-programming safety layer; Volt–Var control	Simulated IEEE feeders; improved constraint satisfaction during training	Safety under arbitrary topology changes or complete PQ compliance
Ding et al. [43]	DRL route and charge/discharge decisions for mobile storage	Numerical network/transport optimization	Mobile storage routing is not equivalent to EV-user availability or measured feeder PQ
Tong et al. [12]	Learning-assisted multi-energy and battery–grid scheduling	Hourly system case study of energy balancing and switching activity	Hourly dispatch does not validate fast PQ response or electrochemical health
Al Qwaid et al. [13]	Attention-enhanced proximal policy optimization (PPO); battery charging	Single-particle electrochemical simulation; mean plating overpotential +0.023 V versus −0.015 V for baseline PPO	Modeled plating margin, not measured capacity retention or feeder compliance
Akinwola and Alkuhayli [29]	Physics-informed multi-battery PV charge allocation	MATLAB/Simulink R2023a comparison: final SOC variance 0.08% versus 0.65% for MPC; settling time 8.2 versus 14.3 s	SOC equalization and transient settling do not establish field/PCC harmonic compliance

The audited AI examples establish algorithmic and simulated control capability, with uneven treatment of battery duty and PQ metrics. No audited source demonstrates the complete combination of AI-controlled PV–BESS–EV operation in an actual distribution system, independently assessed PQ compliance, battery-health limits, and long-duration net-PV recovery. Field deployment of some grid-support functions is well established in Section 6.3, but those projects do not document that complete AI claim. This is an evidence gap within the audited set, not proof that no such deployment exists elsewhere.

3.6. Battery and EV Flexibility Across Operating Time Scales

Battery flexibility depends on service duration. Sub-cycle to second-scale PQ support is constrained by converter bandwidth, current, thermal margin, and DC-side ripple. Minute-to-hour dispatch additionally depends on usable energy, SOC, conversion efficiency, forecasts, and accumulated wear. V2G adds connection and departure-energy constraints. Table 9 relates each service to these resource limits.

Table 9. Battery and EV services for PQ-constrained PV integration: resource limits and representative evidence.

Service	Primary Electrical Action	Target Battery and Converter Duty	Primary Modeling Risk	Evidence
PV energy shifting	Charge during excess PV and discharge when the feeder can accept power	SOC window, usable energy, charge/discharge efficiency, throughput, temperature, and degradation	Hosting gain can be obtained at the cost of heavy cycling, losses, or insufficient energy capacity	[6,34,55]
Reactive voltage support	Inject or absorb reactive power to alter feeder voltage	Converter current and thermal headroom; battery energy exchange may be small but auxiliary losses remain	Apparent-power saturation can reduce simultaneous active-power charging or discharge capability	[6,9,10]
Ramp and active-power smoothing	Charge or discharge over seconds to minutes to moderate net power ramps	Power rating, SOC reserve, throughput, efficiency, control bandwidth, and repeated cycling duty	Fast smoothing may consume battery life without recovering substantial curtailed PV energy	[34,47,108]
V2G flexibility	Shift charging or export active/reactive power from connected EVs	Vehicle connection, bidirectional charger rating, departure energy, SOC/SOH, wear, and user participation	Treating an uncertain mobile resource as firm storage overstates available feeder support	[7,8,23,28]
Harmonic or multifunctional support	Allocate converter current to waveform compensation while maintaining the fundamental service	Harmonic-current share, fundamental current, thermal margin, filter design, and controller bandwidth	Independent attribution of full converter capacity to active, reactive, and harmonic services violates current sharing	[40,45,134]

Temperature, SOC, and cycle depth affect V2G aging [23,28]; compensation and participation affect its economics [37]; and parking duration and demand patterns affect PV–BESS sizing at charging facilities [55]. These dependencies should be modeled explicitly rather than absorbed into a generic flexibility factor.

For a time-series assessment, the renewable-energy consequence should be expressed as an integrated quantity rather than inferred from one favorable snapshot or an installed-capacity limit. Two quantities are distinguished. Gross accepted PV energy is

$$E_{\mathrm{PV},\mathrm{gross}}=\sum _t\left[P_{\mathrm{PV},\mathrm{avail}}\left(t\right)-P_{\mathrm{curt}}\left(t\right)\right]\Delta t,$$

(2)

where $P_{\mathrm{PV},\mathrm{avail}}\left(t\right)$ is available PV power and $P_{\mathrm{curt}}\left(t\right)$ is curtailed PV power. Gross acceptance is useful for identifying curtailed energy, but it should not be reported as delivered renewable energy when storage losses are material. A complementary boundary is therefore

$$E_{\mathrm{PV},\mathrm{net}}=\sum _t\left[P_{\mathrm{PV},\mathrm{direct}}\left(t\right)+P_{\mathrm{PV},\mathrm{dis}}\left(t\right)-P_{\mathrm{aux},\mathrm{PV}}\left(t\right)\right]\Delta t,$$

(3)

where $P_{\mathrm{PV},\mathrm{direct}}$ is PV power delivered directly to local load or the grid, $P_{\mathrm{PV},\mathrm{dis}}$ is PV-origin energy subsequently delivered from storage after discharge conversion losses, and $P_{\mathrm{aux},\mathrm{PV}}$ is auxiliary consumption assigned to that PV–storage service. PV energy remaining in storage at the end of the assessment horizon is reported separately unless the study adopts and states a terminal-energy accounting convention. Terminal SOC should not be given the same value as energy delivered to load or the grid. If a study values residual stored energy, it should state the valuation convention, report the terminal SOC and residual energy explicitly, and must not also count that residual as delivered PV energy. This separation prevents a storage strategy from receiving credit for energy that entered the storage interface but was lost or remained unavailable for useful delivery.

The balance point for Equation (3) is the common AC bus at which local load receives energy and grid export is metered. Direct PV and battery discharge must be referred to that same bus, with applicable conversion and feeder losses included; gross PV capture uses its separately declared PV measurement plane. When the battery also charges from the grid, an explicit attribution convention is needed. Under proportional mixing, separate stored-energy inventories track PV-origin and grid-origin charging after charging losses. For positive stored energy $E_{b,t}$, the PV share ${\alpha }_t=E_{b,\mathrm{PV},t}/E_{b,t}$ gives $P_{\mathrm{PV},\mathrm{dis},t}={\alpha }_t{P}_{d,t}$; discharge from an empty store is prohibited. Discharge and storage losses are allocated in the same proportion. Initial stored energy of unknown origin receives no PV credit, grid charging never becomes PV-origin energy, and auxiliary consumption is deducted only if it has not already been netted from the metered flows. This is a declared accounting convention, not physical identification of electrons.

For an accounting illustration, suppose 10 kWh of PV reaches that bus: 6 kWh serves load directly and 4 kWh charges an initially empty idealized store; another 2 kWh is charged from the grid. With 95% efficiency in each direction and no auxiliaries, the PV and grid inventories become 3.8 and 1.9 kWh. Full discharge delivers 3.61 kWh of PV-origin energy and 1.805 kWh of grid-origin energy, returning the store to its initial inventory. Delivered PV energy is $6+3.61=9.61$ kWh, rather than 11.415 kWh of mixed-source delivery or 10 kWh of gross capture. Empty initial/final inventory simplifies the arithmetic; it is not a recommended battery SOC range or a reported project result.

A consistent comparison reports the balance point, gross PV capture and curtailment, net PV-origin delivery, grid charging, conversion/auxiliary losses, initial/final SOC and provenance, battery throughput, and the limiting PQ metric with its assessment interval. Baseline and controlled cases should use the same horizon, weather/load data, and terminal-energy rule.

Battery state must be carried between time steps so that energy shifted in one interval cannot be reused without accounting for charging and discharging. An energy-state representation avoids changing the state variable merely because usable capacity changes with state of health or temperature [6,9,10,34]:

$$E_{b,t+1}=E_{b,t}+{\eta }_c{P}_{c,t}\Delta t-{\displaystyle \frac{P_{d,t}\Delta t}{{\eta }_d}},{\mathrm{SOC}}_t={\displaystyle \frac{E_{b,t}}{E_{\mathrm{cap},t}}},{\mathrm{SOC}}_{\min }\le {\mathrm{SOC}}_t\le {\mathrm{SOC}}_{\max },$$

(4)

where $E_{b,t}$ is stored battery energy, $E_{\mathrm{cap},t}$ is the available capacity used for SOC normalization, $P_{c,t}$ and $P_{d,t}$ are charging and discharging power, and ${\eta }_c$ and ${\eta }_d$ are the corresponding efficiencies. A study that allows $E_{\mathrm{cap},t}$ to vary with SOH or temperature should state that model separately from the energy balance. When lifetime claims are made, throughput should also be defined explicitly, for example, as charge-plus-discharge energy or equivalent full cycles, and the degradation model should identify chemistry, temperature, C-rate, depth of discharge, and calendar-aging assumptions [23,28,47].

For a multifunctional converter, active-power transfer, reactive-power support, and harmonic compensation compete for finite semiconductor current and thermal capability. A simplified root-mean-square current-sharing envelope, consistent with converter-current allocation studies [40,45,134], is

$$I_1^2+{\displaystyle \sum _{h=2}^{h_{\max }}}{I}_h^2\le I_{\max }^2,$$

(5)

where $I_1$ is the fundamental current, $I_h$ is harmonic current at order h, and $I_{\max }$ is the applicable current limit. This is a screening envelope, not a semiconductor thermal model. Experimental and control studies show that harmonic compensation and active grid support are constrained by converter design and available current headroom [40,45,134]; manufacturer thermal, modulation, filter, and protection limits can produce a smaller feasible region. Figure 2 places these electrical and battery limits in the energy-accounting path from available solar resource to usable PV energy.

Figure 2. Constraint chain from available solar resource to usable PV energy. Electrical limits, battery limits, and EV/deployment conditions must be evaluated jointly so that recovered PV energy is not credited independently of battery throughput, losses, availability, or compliance margin. Abbreviations: BESS, battery energy storage system; EV, electric vehicle; PCC, point of common coupling; PV, photovoltaic; SOC, state of charge; SOH, state of health.

4. Power-Quality Mechanisms Limiting Usable PV Generation

Usable PV generation can be limited by several coupled electrical mechanisms, including voltage excursions, unbalance, waveform distortion, resonance, converter-current saturation, and thermal loading. The following subsections relate each mechanism to PV curtailment and to the specific BESS or EV flexibility that can relieve it.

4.1. Voltage Rise, Voltage Drop, Rapid Voltage Change, and Flicker

For a simple radial equivalent with resistance R, reactance X, net active-power injection P, reactive-power injection Q, and voltage magnitude V, a first-order steady-state voltage-change approximation is as follows [1,64]:

$$\Delta V\approx {\displaystyle \frac{RP+XQ}{V}}.$$

(6)

Here, positive P and Q denote net injection toward the feeder. Equation (6) explains why active-power export can create significant voltage rise in resistive low-voltage (LV) feeders and why the effectiveness of Q control changes with feeder resistance-to-reactance ($R/X$) ratio. Feeder studies identify phase shifting, reverse power flow, and location-dependent voltage sensitivity as distinct causes of the observed voltage rise [135,136,137,138]. Local and combined active/reactive inverter controls exploit that sensitivity with different information requirements [64,65,139]. Hierarchical and coordinated schemes extend the control boundary to storage and multiple feeder devices [6,38,140]. Forecast-aware curtailment and overvoltage-prevention methods address a different operating trade-off: they exchange active energy for voltage margin [141,142,143,144]. The approximation should not be used as a substitute for an unbalanced nonlinear power flow when phase coupling, tap action, voltage-dependent load, or converter limits are material.

PV ramps, cloud transients, EV charging steps, and abrupt BESS dispatch can also create rapid voltage changes. Repeated voltage fluctuations may be evaluated through flicker indices when the disturbance falls within the applicable measurement framework. IEC 61000-4-15 specifies flickermeter behavior, whereas IEC 61000-4-30 defines measurement methods for PQ parameters including flicker and rapid voltage changes [52,145]. Studies of PV and charging networks also show that the outcome depends on feeder strength, charging events, generation variability, and the temporal aggregation used for assessment [56,57,146,147]. For twelve consecutive short-term severity values, the conventional long-term flicker index is

$$P_{\mathrm{lt}}={\left({\displaystyle \frac{1}{12}}{\displaystyle \sum _{i=1}^{12}}{P}_{\mathrm{st},i}^3\right)}^{1/3}.$$

(7)

Here, $P_{\mathrm{st},i}$ is the ith short-term flicker severity and $P_{\mathrm{lt}}$ is the long-term flicker severity. Equation (7) follows the standardized aggregation framework in IEC 61000-4-15 [145]. It is a measurement aggregation rule rather than a universal PV ramp limit. A renewable-integration study should not infer compliance from an arbitrary voltage-change threshold when the applicable requirement is based on a standardized statistic and assessment interval. Conversely, not every PV ramp should be described as perceptible flicker; the terminology and metric should follow the relevant measurement definition.

EV charging changes the sign and timing of net P. Daytime charging can reduce PV export and overvoltage, whereas uncontrolled evening charging can deepen voltage drop and transformer stress. Coordinated charging and V2G can reshape this envelope [7,39,73]. The control objective is bidirectional: release PV export capacity during high generation without creating undervoltage or charging-service failures during periods of high demand.

4.2. Voltage Unbalance, Phase Allocation, and Neutral Current

Residential PV and EV chargers are frequently single-phase, so a balanced three-phase model can conceal the actual hosting constraint. Phase allocation, grounding, and the local mix of PV and charging loads determine whether the negative- and zero-sequence components become material [60,148,149,150]. A standard sequence-component indicator is the voltage-unbalance factor (VUF) [52]:

$$\mathrm{VUF}={\displaystyle \frac{|V_2|}{|V_1|}}\times 100\%,$$

(8)

where $V_1$ and $V_2$ are the positive- and negative-sequence voltage magnitudes. Equation (8) follows the sequence-component definition used for standardized PQ assessment [52]. Unbalance influences customer voltage, rotating equipment, converter control, and harmonic propagation because sequence impedances are different. Phase-aware connection planning, phase-aware smart charging, and coordinated reactive support can create hosting value that does not appear in a balanced feeder study.

Nonlinear single-phase converters also inject triplen harmonic components that can add in the neutral conductor instead of canceling. The resulting neutral current depends on the charger/inverter spectrum, diversity, conductor configuration, and phase allocation; it should not be generalized from one device measurement. Stochastic and probabilistic feeder studies are useful because charging coincidence and connection location determine the aggregate result [33,72,77]. For PV/EV-rich residential feeders, phase allocation and neutral loading should be included whenever conclusions are made about thermal headroom or PQ compliance [59,60,148,149].

4.3. Harmonics, Interharmonics, Supraharmonics, Resonance, and Source Attribution

Harmonic compliance is a network interaction problem. IEEE 519 frames harmonic control at the PCC [32], while IEC 61000-4-7 defines instrumentation and grouping for harmonics/interharmonics and IEC 61000-4-30 defines broader PQ measurement methods [52,53]. A device that satisfies an equipment-level emission test does not guarantee feeder-level compliance because the resulting waveform depends on aggregate converter spectra, background voltage distortion, frequency-dependent network impedance, resonance, and operating state [151,152,153,154,155,156].

Voltage and current total harmonic distortion (THD) are defined over the stated assessment bandwidth as follows [52,53]:

$$\begin{array}{cc}\hfill {\mathrm{THD}}_V& ={\displaystyle \frac{\sqrt{{\sum }_{h=2}^{h_{\max }}{U}_h^2}}{U_1}}\times 100\%,\hfill \end{array}$$

(9)

$$\begin{array}{cc}\hfill {\mathrm{THD}}_I& ={\displaystyle \frac{\sqrt{{\sum }_{h=2}^{h_{\max }}{I}_h^2}}{I_1}}\times 100\%,\hfill \end{array}$$

(10)

where $U_h$ and $I_h$ are the rms voltage and current components at harmonic order h, $U_1$ and $I_1$ are the fundamental rms voltage and current, and $h_{\max }$ is the highest harmonic order included in the assessment. For network-current assessment, IEEE 519 uses total demand distortion (TDD), which can be expressed as follows [32]:

$$\mathrm{TDD}={\displaystyle \frac{\sqrt{{\sum }_{h=2}^{h_{\max }}{I}_h^2}}{I_L}}\times 100\%,$$

(11)

where $I_L$ is the maximum demand load current at the PCC under the applicable assessment convention. Unlike instantaneous current THD, TDD uses a demand-current basis and therefore avoids inflating the ratio solely because the fundamental current is small.

IEEE 1547-2018 instead specifies total rated-current distortion (TRD) at the reference point of applicability (RPA) for DER current-distortion assessment [51]. Using the prescribed measurement bandwidth,

$$\mathrm{TRD}={\displaystyle \frac{\sqrt{I_{\mathrm{rms}}^2-I_1^2}}{I_{\mathrm{rated}}}}\times 100\%,$$

(12)

where $I_{\mathrm{rated}}$ is DER rated current referred to the RPA and $I_{\mathrm{rms}}$ includes harmonic and interharmonic components. TRD differs from both current THD and Equation (11) in normalization and, for the integer-harmonic sum shown there, spectral content. IEEE 1547 excludes distortion attributable to pre-existing area-system voltage distortion under its assessment rules. Its TRD limit is 5%, subject to the standard’s provisions, including agreed active-filtering operation. A TDD value at a customer PCC cannot simply be relabeled as DER TRD.

A compact Thevenin-type harmonic-network representation used in impedance-aware assessment [3,4,157,158] is

$$V_h=Z_h{I}_h+V_{h,\mathrm{bg}},$$

(13)

where $V_h$ is the harmonic voltage, $I_h$ is the injected harmonic current, $Z_h$ is the feeder harmonic impedance, and $V_{h,\mathrm{bg}}$ is the background harmonic voltage. This simplified relation makes the location and operating-state dependence explicit: identical current injection can produce different harmonic voltages. Experimentally validated PV work confirms that background harmonic voltage can reduce hosting capacity [3], and feeder studies show that high PV penetration and passive-filter tuning can alter harmonic hosting and resonance behavior [4,5]. Related work on measured distortion, PV harmonic propagation, and nonlinear-load interactions likewise supports an impedance- and boundary-aware interpretation [157,158,159,160,161]. Source-attribution methods are needed when utility background distortion and customer converter emissions coexist [71].

Resonance is especially important when shunt capacitance, filters, cables, and grid inductance create a high-impedance region near a converter emission. A simplified parallel-resonance estimate used for initial impedance screening [5,158] is

$${\omega }_{\mathrm{res}}\approx {\displaystyle \frac{1}{\sqrt{L_{\mathrm{grid}}{C}_{\mathrm{shunt}}}}},$$

(14)

where ${\omega }_{\mathrm{res}}$ is the estimated resonant angular frequency, $L_{\mathrm{grid}}$ is the equivalent grid inductance, and $C_{\mathrm{shunt}}$ is the shunt capacitance. The actual resonance requires the complete frequency-dependent network and converter model. The practical implication is that passive filtering can increase harmonic-constrained PV hosting in one feeder [5] but can be ineffective or adverse after the network impedance changes.

The THD expressions above apply to integer harmonics and do not aggregate interharmonics. The reviewed spectrum also extends beyond conventional low-order harmonics. Interharmonics occur at noninteger multiples of the fundamental, and conducted emissions in the 2–150 kHz range are commonly discussed as supraharmonics. IEC 61000-4-30:2025 includes measurement treatment for emissions across the 2–150 kHz range, while IEC 61000-4-7 remains central to harmonic/interharmonic grouping at lower frequencies [52,53]. Charger measurements show that the spectrum changes with topology, charging power, state of charge, and the local voltage waveform [46,162,163]. Harmonic-control studies add the effects of filtering, compensation bandwidth, and charger operating state [134,164,165,166]. The assessment must state its bandwidth; low-order THD cannot summarize the full converter spectrum.

4.4. DC Injection, Protection Interaction, and Converter-Current/Thermal Coupling

Transformerless converters can produce a small DC component because of sensor offset, modulation asymmetry, or control bias. Persistent DC injection can bias magnetic components and increase thermal stress. The admissible level and test method must be taken from the applicable interconnection/product requirement; no universal percentage is assigned across standards or jurisdictions. IEEE 1547 and its 2020 amendment are used here for distributed energy resource (DER) interconnection behavior, not as a universal rule for EV chargers or all customer equipment [51,167]. Protection and ride-through evidence must be interpreted with feeder fault level, protection settings, inverter current limiting, and the active-power reduction strategy applied during voltage sags [168,169,170,171,172,173]. Distribution-network studies quantify how PV and EV integration changes line loss, voltage drift, and feeder loading [174,175,176,177]. Peak-demand and integrated PV/EV studies add the dependence on charging demand, storage, and demand response [178,179], while PQ indices and distorted-network assessments address waveform-related loading [180,181]. Thermal headroom therefore requires direct assessment of conductor and transformer ampacity, neutral loading, harmonic losses, and converter derating.

PQ support can also shift the binding constraint from voltage to converter current. With an apparent-power rating $S_{\max }$, a standard fundamental-frequency converter-capability boundary [6,45] is

$$P^2+Q^2\le S_{\max }^2.$$

(15)

Equation (15) is a simplified fundamental-frequency converter-capability boundary. It shows that operation near maximum PV active power leaves less reactive-power headroom unless the inverter is oversized or active power is curtailed. Harmonic compensation also consumes current capability and can increase switching and thermal stress [40,45,134]. For BESS and V2G, the feasible region is further reduced by SOC, energy reserve, mobility, and charging constraints. A controller should report its feasible P/Q/harmonic-current region under saturation rather than an unconstrained ancillary-service capability.

Table 10 summarizes how these mechanisms translate into renewable-generation consequences and identifies the associated battery or converter burden. The organization is intentionally mechanism-based because the same mitigation can release one constraint while tightening another through current saturation, additional cycling, altered harmonic impedance, or reduced energy reserve.

Table 10. Coupled PQ mechanisms limiting usable PV generation and associated evidence.

PQ Mechanism	Impact on Renewable Generation	EV and BESS Coupling	Core Mitigation Strategy	Standards and Evidence Basis
Steady-state voltage rise/drop	PV curtailment or reduced hosting when voltage limits bind	charging may absorb export; BESS/V2G can shift or inject P/Q	local or coordinated P/Q control, scheduling, storage	[1,2,39,64]
Rapid voltage change/flicker	limits acceptable ramping or switching behavior	charger/BESS steps can add or counter voltage changes	ramp-rate control, scheduling, fast inverter support	[52,145]
Unbalance/neutral loading	phase-specific voltage or thermal limit can bind before three-phase capacity	single-phase charging and V2G alter phase loading	phase allocation, phase-aware scheduling and control	[16,33]
Harmonic distortion/resonance	harmonic headroom can reduce PV hosting despite acceptable fundamental voltage	charger/BESS spectra add to or compensate distortion	impedance-aware assessment, active/passive filtering, source attribution	[3,4,5,71]
Converter-current/thermal limit	Q or harmonic support can force active-power curtailment	BESS/V2G services compete for the same converter rating	explicit current, thermal, SOC and service-priority constraints	[6,34,40]

5. Control Under Coupled Network and Battery Constraints

Table 11 provides a source-referenced comparison without treating heterogeneous case-study results as a universal performance ranking. The comparison focuses on the renewable/PQ benefit, binding implementation constraints, and failure modes that can invalidate an otherwise favorable result. More fundamentally, a controller should be judged by the constraint it releases and the cost of maintaining that release: a voltage reduction that increases feeder current, a harmonic intervention that exhausts converter headroom, or a BESS schedule that recovers curtailed PV only through excessive cycling is not a complete renewable-integration benefit. Comparisons require an explicit baseline, time horizon, feeder and PCC boundary, device saturation model, fallback behavior, and accounting of curtailed/recovered PV energy, losses, and battery throughput. This criterion prevents nominal control performance from being mistaken for transferable operating value.

Table 11. Control strategies for expanding the feasible PV operating region.

Control Strategy	Renewable and PQ Benefits	Binding Implementation Constraints	Potential Operational Failure Modes	Evidence
Local Volt–Var/droop	Fast autonomous voltage support with low communication burden	Inverter current and reactive-power headroom; local sensing; interaction among many devices	Local actions can increase losses or move voltage problems elsewhere in the feeder	[1,49,64]
Coordinated Volt–Var/Volt–Watt	Allocates reactive power and curtailment system-wide and can increase PV hosting	Telemetry, model or sensitivity accuracy, communication delay, and curtailment allocation	An optimizer solution may become infeasible or ineffective when telemetry is delayed or the network model is inaccurate	[6,38,41,142]
Centralized OPF/MPC	Explicitly coordinates multiple network, converter, and battery constraints over an operating horizon	State estimation, communications, computation time, forecast quality, and credible device models	A nominal optimum can fail when mobility, SOC, converter headroom, or feeder conditions differ from the assumed state	[41,81,112,182]
Distributed MPC	Reduces dependence on one central optimizer and can coordinate geographically dispersed EV flexibility	Neighbor or agent communications, convergence tolerances, asynchronous updates, and data availability	Communication degradation or poor convergence can fragment the feasible region and invalidate the coordinated service	[7,39,42,183]
BESS dispatch	Absorbs solar export and provides active or reactive support across different operating time scales	SOC, SOH, degradation, converter rating, losses, thermal limits, reserve requirements, and efficiency	A reported hosting gain may be purchased through excessive cycling, battery-health loss, or low net renewable-energy recovery	[6,9,35,47,55]
AI-enabled supervisory coordination	Forecasts, state estimates, and adaptive BESS/EV scheduling	Training domain, uncertainty, telemetry, inference time, hard constraints, and fallback	Apparent gains can depend on unseen violations, weak baselines, or omitted degradation	[11,12,13,43]
V2G aggregation	Adds mobile active and reactive power flexibility and can align charging demand with PV generation	Vehicle availability, user SOC/SOH, battery wear, charger interoperability, compensation, mobility reserve, and market access	Scheduled service becomes unavailable when vehicles disconnect, user energy requirements bind, or lifetime and compensation constraints are ignored	system/control: [7,8,36]; deployment: [50,106]; availability and lifetime: [23,28,37]
Active filtering/converter PQ support	Provides fast harmonic compensation and can improve waveform quality at the PCC	Converter current margin, switching and thermal stress, sensing bandwidth, filter design, and simultaneous fundamental-frequency services	Fundamental active/reactive service and harmonic compensation compete for finite semiconductor current and thermal capability	[40,134,164,184]

5.1. Autonomous Local Control

Local Volt–Var and Volt–Watt functions retain operation during loss of upstream communication. However, correcting local voltage can increase current and losses elsewhere, and the first site to reach a curtailment threshold need not have the greatest sensitivity to the feeder constraint. Settings therefore require feeder-aware operating envelopes [1,64].

5.2. Coordinated and Optimization-Based Control

Hierarchical control, OPF, and MPC can coordinate voltage, thermal loading, curtailment, SOC, and EV availability [38,41,81]. Evaluation should disclose model and forecast errors, update interval, solve time, infeasibility handling, and fallback when telemetry is stale. Benchmark optimization alone does not validate the information and communication chain [185,186,187].

5.3. AI-Enabled Supervisory Control

The AI functions and safety mechanisms in Section 3.5 must be evaluated against a conventional controller with matching physical constraints. Mobile-storage routing and physically informed battery allocation demonstrate useful decision roles [29,43]; battery-state surveys identify uncertainty and transferability limits [11]. A convincing system comparison reports usable PV delivery and violations under forecast error, communication loss, and states outside the training domain.

Distributed coordination can reduce dependence on a single optimizer and exploit geographically dispersed EV reactive-power capability through distributed MPC [39,42]. It does not, however, eliminate communication requirements. Neighbor-to-neighbor exchange, convergence tolerances, asynchronous updates, and packet loss can determine whether the resulting solution remains feasible. Its value therefore lies in resilience and scalability under a specified communication model, not in an assumed reduction in data exchange.

Storage and V2G availability require power/energy ratings, efficiency, and cycling duty to be specified [34,108,109,188,189]. Active support uses stored energy; reactive support can also impose battery ripple unless the interface buffers the pulsating power. EV availability, user requirements, and charging coordination further restrict the service [8,39,75,190,191,192]. The energy-attribution convention in Section 3.6 applies when grid and PV charging coexist.

5.4. Harmonic Compensation and Hardware Support

Active filtering and multifunctional chargers can recover harmonic headroom within the shared current envelope of Equation (5). Studies address charger models, rectifier suppression, combined storage/filtering, and bidirectional architectures [193,194,195,196,197,198]; static synchronous compensator (STATCOM), distribution STATCOM (D-STATCOM), and adaptive control studies supply related voltage/waveform evidence [199,200,201,202]. Passive filters require impedance checks because a network change can create a new resonance [5,40]. Selection should follow the binding disturbance, available hardware, and validated operating range.

6. Compliance, Validation, and Deployment

6.1. Compliance Boundary and Jurisdictional Scope

Table 12 distinguishes interconnection requirements, installation-level disturbance allocation, equipment emission limits, and measurement methods. IEEE 1547.9 applies IEEE 1547 to storage DER, including relevant bidirectional interfaces [203]. EN 50549-1 and -2 address generating-plant connection to low- and medium-voltage (LV/MV) networks, subject to national implementation [204,205]. IEC TR 61000-3-6 covers MV, high-voltage (HV), and extra-high-voltage (EHV) installation emission allocation; IEC TR 61000-3-14 addresses LV installations. IEC TR 61000-3-15 assesses LV dispersed-generation electromagnetic compatibility (EMC) requirements [206,207,208]. The latter is cataloged by IEC as a Technical Report (TR). These installation documents complement equipment limits and do not replace them. Battery safety and EV communication are separate boundaries under IEC 62619 and ISO 15118-20 [209,210,211]. Applicability requires the local connection code, adopted edition, operating mode, and measurement point to be stated.

Table 12. Principal standards referenced in the review and their assessment boundaries.

Standard/Code	Assessment Boundary	Required Conformance Evidence	Implications for Grid Integration
IEEE Std 519-2022 [32]	PCC harmonic environment	voltage/current spectra, demand-current basis, aggregation/operating condition	equipment compliance alone does not establish feeder/PCC compliance
IEEE Std 1547-2018 [51]	DER–area electric power system (EPS) interface at the RPA	performance category, settings, ride-through, and current distortion including TRD	rated-current normalization differs from PCC TDD; converter limits still apply
IEEE Std 1547.9-2022 [203]	Storage DER interconnection; applicable bidirectional modes	application of IEEE 1547 to storage operating modes, export capability, and connection arrangements	a storage application guide, not a substitute for equipment testing
EN 50549-1/-2:2019 + A1:2023 [204,205]	Generating plants on LV/MV networks, respectively	applicable plant category, protection and operational capabilities, and national implementation	European connection requirements must be mapped to the local network code
IEC TR 61000-3-6:2008 [206]	Distorting installations on MV/HV/EHV systems	planning levels, background distortion, and allocated emission contribution	installation allocation differs from an individual converter emission test
IEC TR 61000-3-14:2011 [207]	Disturbing installations on LV systems	harmonics/interharmonics, voltage fluctuations and unbalance, with system absorption capacity	supports site-specific allocation; does not prescribe a mitigation device
IEC TR 61000-3-15:2011 [208]	Low-frequency EMC of dispersed generation on LV networks	emission/immunity assessment and test context for single- and multi-phase converters	technical guidance rather than a universal interconnection limit table
IEEE Std 1547.1-2020 [212]	DER interconnection conformance testing	prescribed test configuration, procedures, and measured response for claimed IEEE 1547 functions	controller simulation does not replace equipment-level conformance evidence
IEC 61000-4-7 [53]	harmonic and interharmonic measurement	stated bandwidth, grouping and instrument method	comparisons require consistent spectral measurement definitions
IEC 61000-4-15/IEC 61000-4-30 [52,145]	flicker and broader PQ measurement	instrument class/method, aggregation interval, rapid-voltage-change (RVC) and flicker definitions	PV ramps or charger steps should be evaluated using the applicable standardized metric
IEC 61000-3-2/IEC 61000-3-12 [213,214]	equipment harmonic-current emission limits	rated input current, equipment category, test conditions, and applicable connection boundary	equipment-level emission compliance does not by itself establish feeder/PCC harmonic compliance
IEC 61000-2-2/IEC 61000-2-12 [215,216]	Compatibility levels on LV/MV public systems	network environment, disturbance quantities, and applicable compatibility levels	compatibility levels are distinct from equipment limits and allocated site emissions
IEC 62619:2022 [209]	safety of industrial secondary lithium cells and batteries, including stationary applications	applicable safety tests, battery design, and declared operating conditions	grid-service claims remain bounded by battery safety and operating limits
ISO 15118-20:2022 + Amd 1:2026 [210,211]	EV–charging-station communication for bidirectional power transfer and DER services	interoperable communication functions, supported services, and security provisions	electrical V2G capability is not deployable without a compatible communication and service interface
EN 50160 [54]	characteristics of supplied voltage	applicable national adoption and monitoring interval	supply-voltage characteristics are not a substitute for DER equipment requirements

Market remuneration, flexibility procurement, stakeholder liability, tariff design, export rights, and cost allocation are more jurisdiction-specific than the underlying electrical equations. The Australian and UK projects discussed below show that dynamic export and V2G services can be implemented, but they do not establish universal payment rules or a globally harmonized grid-forming requirement. Any deployment must map the engineering control to the local interconnection code, metering and settlement arrangement, consumer agreement, responsible network actor, and allocation of curtailed-energy, degradation, and communication costs. A technically effective controller can still be unattractive when it secures voltage compliance through excessive PV curtailment, costly communication infrastructure, or uncompensated battery wear; conversely, a BESS or V2G action with a high direct cost can be justified when it recovers renewable energy, defers reinforcement, and delivers multiple verifiable services over its lifecycle. The electrical feasibility region must consequently be evaluated together with the commercial mechanism that makes the assumed flexibility available, while avoiding unsupported extension of IEEE/IEC/EN technical evidence into global policy claims.

6.2. Validation Evidence and Transferability

To avoid equating the volume of simulation literature with deployment maturity, the review attaches validation level to the claim being made rather than converting evidence into an unsupported numerical score. Figure 3 summarizes the claim-specific progression from planning studies to field deployment while preserving the distinction between feasibility, real-time implementation, interoperability, and operational transferability.

Figure 3. Claim-specific evidence maturity ladder used in the review. Higher levels add implementation and deployment evidence, but the appropriate endpoint depends on the claim and the ladder is not a numerical ranking of controller families. Abbreviations: EMT, electromagnetic transient; HIL, hardware-in-the-loop.

Validation evidence is discussed explicitly but is not converted into a numerical distribution unless every counted record and classification is supplied in a traceable audit trail. A validation label is used only when the study boundary and platform are explicit; title metadata alone are not used to infer maturity. Recent laboratory work provides a useful example: a PV–wind microgrid with battery–supercapacitor storage was evaluated experimentally against a battery-only baseline, linking transient DC-bus buffering to AC-side voltage behavior [30]. This distinction separates evidence for engineering feasibility from evidence for operational readiness. Benchmark and real-feeder simulation remain useful for planning claims, whereas fast converter, harmonic, protection, and communication claims may require electromagnetic-transient (EMT), controller- or power-hardware-in-the-loop (HIL), laboratory, pilot, or repeated field evidence. The appropriate endpoint depends on the claim: a planning-screening method does not require the same evidence as a sub-second closed-loop PQ controller. This conservative treatment deliberately avoids presenting an untraceable small-sample bar chart as if it were a census of the literature.

6.3. Field, Pilot, and Experimental Evidence

Table 13 separates field and pilot demonstrations from laboratory, prototype, and system-model evidence that is sufficiently specific to test the transferability logic developed in the preceding sections. The evidence level is therefore part of each entry and is not treated as interchangeable across rows. The entries are used to establish what has actually been demonstrated under stated technical and institutional conditions, not to imply universal transferability across feeders, jurisdictions, battery chemistries, charger populations, communication architectures, or market arrangements. In the project labels, NREL denotes the National Renewable Energy Laboratory, SCE denotes Southern California Edison, REVS denotes Realising Electric Vehicle-to-Grid Services, and VPP denotes virtual power plant; these abbreviations are retained only where they are the established project or industry names.

Table 13. Field, pilot, laboratory, and system-model evidence for PV hosting and battery–grid integration.

Project/Deployment	Technology and Validation Level	Reported or Supported Grid Capability	Transferability and Deployment Limits	Reference
NREL–SCE Fontana field demonstration	High-penetration PV with advanced inverter functions; utility field demonstration	Tested inverter functions on an operating distribution circuit with high PV penetration and connected local control behavior to feeder-level voltage and PV integration.	One utility circuit with project-specific feeder impedance, inverter population, monitoring, and operating procedures; the result is evidence of feasibility rather than a universal setting.	[49]
UK low-voltage PV–storage field trial	Customer-owned rooftop PV and small-scale storage; field PQ measurements combined with desktop network studies	Field measurements assessed coordinated storage operation and power quality. A separate modeled 100% storage-uptake case produced 4.1094% third-harmonic voltage against a 4.0% study limit when other nonlinear loads were included.	The 100% uptake and harmonic result are modeled extrapolations, not observations from full-penetration field deployment; the controller was not established as AI.	[217]
REVS, Australia	51 EVs using bidirectional charging; fleet-scale pilot	Field availability evidence: 38 of 51 chargers had usable data and 29 had discharge data. In 2022, recorded discharge was 0.146 MWh versus 18.4 MWh imported; this is sparse contingency-service duty, not continuous regulation.	The trial comprised 51 bidirectional chargers; the final analytical dataset included 38 chargers, with discharge data available for 29. Australian regulatory conditions, market arrangements, charger technology, and customer participation limit direct transfer.	[50]
Project Sciurus, United Kingdom	More than 300 domestic V2G units; large residential pilot	Demonstrated household-scale bidirectional charging, customer behavior, data collection, aggregation, and portfolio operation at a scale beyond laboratory prototypes.	UK distribution-network connection rules, tariffs, domestic load patterns, and trial hardware are jurisdiction and technology specific.	[106]
Advanced VPP Grid Integration, South Australia	Distributed PV, batteries, and dynamic export limits; utility–aggregator demonstration	An 18-month field trial at 1000 sites increased available VPP export capacity from 5 MW to 6–8 MW during solar hours, with dynamic limits and communication-loss fallback. This is export capacity, not measured annual net PV-energy recovery.	Implementation depends on the utility/VPP interface, network models, communication availability, and Australian governance arrangements.	[58]
PV-coupled EV charging in low-voltage feeders	EV charging facilities coordinated with local PV; application study on LV feeders	Modeled how charging timing and location alter feeder loading and can improve local PV utilization when charging is coordinated with network conditions.	Results depend on feeder topology, charger penetration, driving/charging patterns, and assumptions regarding controllability and customer behavior.	[56]
Three-phase off-board EV charger voltage support	Bidirectional charger prototype; laboratory implementation	Demonstrated practical converter-based voltage regulation and showed that an EV interface can provide grid-support functionality beyond uncontrolled charging.	Prototype ratings, controller bandwidth, measurement filters, and laboratory network conditions do not establish fleet-scale availability or field interoperability.	[48]
Active harmonic rejection in EV charging	Converter control with experimental waveform support	Demonstrated that charger power electronics can participate in harmonic mitigation, supporting the multifunctional-converter interpretation adopted in the synthesis.	Harmonic performance depends on grid impedance, background distortion, filter design, current headroom, switching frequency, and the measurement boundary.	[40]
Lab-scale renewable microgrid with hybrid storage	PV–wind generation with battery–supercapacitor storage; analytical and laboratory experimental validation	Demonstrated how transient storage buffering changes DC-bus droop and AC-side voltage behavior under source and load disturbances.	Lab-scale ratings, hybrid storage architecture, and disturbance profiles do not establish distribution-feeder hosting or field interoperability.	[30]
PV–storage electric-bus charging station	PV, BESS, and electric-bus charging with day-ahead scheduling under uncertainty; system-level 2026 study	Quantified coordinated PV–storage charging under uncertain renewable production and charging demand, reinforcing the need to couple storage scheduling to charging-service requirements.	The optimization objective, bus timetable, tariff assumptions, efficiency model, and station architecture determine the preferred schedule and limit direct feeder-to-feeder transfer.	[218]
MW-scale zinc–iron flow battery source–grid–load–storage control	1 MW-class flow-battery system model with SOC estimation and coordinated control; 2026 battery–grid study	Modeled battery SOC estimation, power allocation, and source–grid–load–storage coordination, showing that large-battery support depends on internal battery state and system loss.	Battery chemistry, control objectives, source/load profile, dispatch interval, and frequency-support context differ from PV-rich LV feeders and require revalidation before transfer.	[31]

The REVS observation counts also affect interpretation. Usable charger data covered 38/51 installations (74.5%), and discharge records covered 29/51 (56.9%). The remaining installations cannot be treated as zero-service devices or assumed available. Incomplete observation reduces confidence in fleet-wide extrapolation, and the report does not establish that missing records are random. The observed 0.146 MWh discharge relative to 18.4 MWh import (approximately 0.8%) supports a low-throughput interpretation for the specific contingency service but does not establish low wear under continuous regulation or daily energy shifting. No statistical correction is applied without the underlying dataset. REVS is therefore used as evidence of practical availability and implementation requirements rather than an estimate of universal V2G effectiveness [50].

The South Australian VPP result provides a complementary network observation: export capacity increased from 5 MW to 6–8 MW, equivalent to a calculated 20–60% increase, at 1000 sites over an 18-month trial. The report connects that result to dynamic operating envelopes and local fallback [58]. It does not isolate an AI contribution or report net annual recovered PV energy under a common battery-aging model. These differences explain why numerical project results inform the synthesis without being pooled as equivalent performance effects.

The field evidence changes the interpretation of maturity. V2G should no longer be described simply as a simulation-stage concept, but pilot operation does not remove battery-life, customer-availability, standards, communications, or settlement constraints. Dynamic export limits likewise show that network-aware coordination can release additional renewable export capacity, but they require a trusted utility–aggregator interface and fail-safe local behavior. These functions have been demonstrated under specific technical and institutional conditions; they are not universally deployment ready. Broader implementation studies also show that coordinated EV-grid support depends on charger capability, feeder conditions, control architecture, and the interface between customer service and network objectives rather than transferring a controller diagram directly into field operation [56,80,147,219].

Nduka et al. [217] provide a PV–storage field comparator with power-quality measurements. Their network-wide 100% storage-uptake harmonic analysis is a desktop extrapolation informed by field data, not a measured full-penetration deployment: with other nonlinear loads, the modeled third-harmonic voltage reached 4.1094% against the study’s 4.0% limit. The result illustrates why device-level conformity cannot substitute for feeder-level PQ assessment; it does not verify AI control or establish universal compliance margins.

6.4. Scope Limitations and Evidence Boundaries

Expert selection can omit relevant studies, and the supplementary title-query check cannot establish complete coverage. Its publication window and automated classification are distinct from the wider bibliography and claim-specific source appraisals. Heterogeneous feeders, battery duties, baselines, and assessment intervals prevent pooled effect estimates or universal control rankings. Field and laboratory findings remain conditional on the tested equipment, communication architecture, chemistry, and local connection rules. The broader AI–BESS reviews [21,22] provide complementary method coverage; the present synthesis concentrates on PQ constraints, battery duty, and source-attributed PV delivery.

7. Conclusions

PV utilization depends on the binding voltage, waveform, thermal, protection, or converter constraint. Battery services differ in their energy and current requirements: reactive and harmonic support consume converter headroom and can impose DC-side ripple, especially in single-phase or unbalanced operation, whereas sustained active support also requires usable energy and mobility reserve. Experimental ripple studies report different aging outcomes under different cells and duties [128,129]; a small mean energy exchange alone cannot establish negligible battery impact. AI should therefore be evaluated inside a verified electrical and battery feasibility layer.

Field projects establish that V2G and network-aware export control can operate outside simulation, subject to charger, communication, customer, and local-code constraints. None of the 30 selected sources in Supplementary Tables S1 and S5 demonstrates the combined field claim of AI coordination, independently assessed PQ compliance, battery-health limits, and long-duration net renewable-energy recovery. This combination remains a research gap within the assessed evidence. Comparisons should report PV-origin delivery at a declared balance point, grid charging and terminal energy separately, together with battery duty and compliance margin.

The quantitative evidence illustrates these boundaries: the South Australian trial expanded available export capacity from 5 MW to 6–8 MW, whereas REVS recorded only 0.146 MWh of discharge against 18.4 MWh imported; the two results characterize different services [50,58]. Cell-level shallow-cycle aging evidence also depends on the comparison with high-SOC storage [28]. These findings support conditional, service-specific conclusions rather than universal hosting or lifetime gains.

Future comparisons should evaluate controllers on the same feeder, battery model, converter limits, uncertainty set, and validation platform, using a common energy-attribution convention. This would distinguish decision-making gains from differences in physical assumptions.