Empirical EV Load Model for Distribution Network Analysis

Abstract

Electric vehicles (EVs) have introduced new operational challenges for distribution network service providers (DNSPs), particularly for voltage regulation due to unpredictable charging behaviour and the intermittent nature of distributed energy resources (DERs). This study focuses on formulating an empirical EV load model that characterises charging behaviour over a broad spectrum of supply voltage magnitudes to enable more accurate representation of EV demand under varying grid conditions. The empirical model is informed by laboratory evaluation of one Level 1 and two Level 2 chargers, along with five EV models. The testing revealed that all the chargers operated in a constant current (CC) mode across the applied voltage range, except for certain Level 2 chargers, which transitioned to constant power (CP) operation at voltages above 230 V. A model of a typical low voltage network has been developed using the OpenDSS software package (version 10.2.0.1) to evaluate the performance of the proposed empirical load model against traditional CP load modelling. In addition, a 24 h case study is presented to provide insights into the practical implications of increasing EV charging load. The results demonstrate that the CP model consistently overestimated network demand and voltage drops and failed to capture the voltage-dependent behaviour of EV charging in response to source voltage change. In contrast, the empirical model provided a more realistic reflection of network response, offering DNSPs improved accuracy for system planning.

1. Introduction

The transition to clean energy is driving the adoption of EVs on a global scale. In 2023, global EV sales reached nearly 14 million, representing 18% of total car sales, bringing the cumulative number of registered EVs to 40 million. Projections suggest this number could grow to 145 million by 2030 [1]. In addition, Figure 1 illustrates the forecasted DER installed capacity for Australia, where EV charging demand is projected to exceed 80 GW by 2048, approaching the total generating capacity of the National Electricity Market (NEM), which stood at approximately 88 GW as of the end of 2024 [2]. This trend indicates a rapid extension of the EV market, suggesting that EVs are going to constitute a significant portion of the electrical power system load in the near future.

In parallel with the increasing adoption of EVs, the accelerated integration of renewable energy sources (particularly rooftop photovoltaic (PV) systems) has introduced a distinct set of challenges for low-voltage (LV) distribution networks, especially in Australia [3,4]. The proliferation of rooftop solar PV has caused voltage regulation challenges within LV systems, particularly voltage rise, adversely affecting electrical appliances [5], including EVs.

Figure 1. DER installed capacity forecasted by AEMO (AEMO is the independent energy market and system operator and system planner for the National Electricity Market (NEM) and Western Australia’s Wholesale Electricity Market (WEM).) [6].

Figure 1. DER installed capacity forecasted by AEMO (AEMO is the independent energy market and system operator and system planner for the National Electricity Market (NEM) and Western Australia’s Wholesale Electricity Market (WEM).) [6].

Analysis shows that traditional distribution networks, designed for legacy load capacity and voltage regulation [3], are not equipped to manage the demands of widespread EV charging. The Australian power system, for instance, is configured to supply residential, commercial, and industrial consumers at a nominal 230 V (with ±10% tolerance) [7]; however, the integration of EVs and renewable energy risks violating these limits. This is particularly evident in the misalignment between residential EV charging peaks, which occur predominantly at night [8], and solar generation, which peaks during the middle of the day. This mismatch between local generation and load is leading to power quality issues, including voltage variation, reduced operating margins, phase imbalance, and harmonic distortion [9]. Given the predictions for increases in EV charging load, a more in-depth understanding of the relationship between EV charging loads and distribution network voltage magnitudes based on accurate EV charger electrical characteristics is essential for power system planning purposes.

1.1. Impact of EV Charging on Distribution Network

The recent surge in the implementation of EV chargers of various ratings has introduced challenges to conventional power systems, including their capacity to accommodate the increased load. Additional challenges have emerged, such as misconceptions concerning EV charging behaviour, the diversity in EV uptake across different regions, and grid optimization issues [10]. The increasing adoption of EV chargers has been associated with various power quality disturbances, including voltage sags, voltage unbalance, and voltage harmonics [11]. Numerous studies have examined the grid-level impacts of EV charging, revealing potential consequences, such as system voltage instability, power quality issues, transformer aging, and heightened peak demand [9]. The influence of EV charging loads on the power system varies based on their size, distribution network configuration, and proximity to service transformers, highlighting the need for a nuanced understanding of these factors in assessing the broader implications of EV integration.

At present, the market offers three distinct levels of electric vehicle chargers, with Level 2 chargers emerging as the most popular across both residential and commercial applications, while Level 1 chargers are common for residential use due to their lower changing capacity, as shown in

Table 1. EV charging levels [16].

	Level 1	Level 2	Level 3
Output Power	1.4–3.7 kW	7–22 kW	25–350 kW
Mounting Type	On-board	On-board	Off-board
Supply Voltage	120/230 VAC Single-phase	208/240 VAC Single-phase	208/240 VAC Three-phase and 300–800 VDC

. An EV load equipped with a Level 2 charger has the capacity to double the peak load demand of a single residential household [11], consequently increasing voltage drops in the LV network and potentially resulting in thermal constraints. The introduction of EV charging load has led to increased nighttime loading due to customer behaviour. The predominance of overnight home charging, accounting for approximately 80% of EV charging events [12], is primarily driven by the convenience of charging upon returning home from work. The study presented in [11] explores the impacts of EV load size on the distribution transformer bus, suggesting that higher currents contribute to larger secondary voltage drops. In essence, as the size of the EV charger increases from Level 1 to Level 3, the voltage drops also escalate accordingly. Furthermore, it has been shown that Level 3 chargers, normally operated at high switching frequency, have severe impacts on harmonics [13,14] and exacerbate the network voltage unbalance [15].

The impact of EV charger placement on the grid was also examined in [11]. The induced voltage drops are more considerable when the vehicles are charged at locations further from the substation transformer, which is due to a longer feeder length. Even if the total EV charging demand remains constant, their unplanned positioning will potentially affect the voltage magnitudes to some extent. Specifically, a concentrated load, for example, numerous EVs being charged at the same location on the feeder, could lead to more significant voltage drops compared to the situation where the same amount of EVs is evenly distributed along that feeder [11]. This outcome underscores the impact of clustering EV loads at a single location on the grid, which resembles the impact of large EV loads, such as electric buses or trucks, as discussed in [17,18]. Additionally, clustering EV loads can create voltage sags in one phase and subsequently cause voltage swells on the others, as well as introduce voltage unbalance into the network [19].

The findings in [20] reveal that Grid-to-Vehicle (G2V) operation has adverse impacts on voltage stability; however, the situation could be alleviated by the Vehicle-to-Grid (V2G) principle, or the implementation of capacitors, a distributed static compensator (DSTATCOM), renewable distributed generation (RDG), and a battery energy storage system (BESS) to mitigate voltage deviation. The findings presented in [21] agree that during V2G mode, the EV acting as a power source could improve the network stability and curtail the peak load demand. Clearly, large-scale EV deployment will have impacts on the grid as it increases the magnitude of the load, causing voltage to vary, making EV load planning and geographical allocation essential [22].

There are several studies examining the influence of widespread adoption of EVs on DNSP PQ disturbance magnitudes, which are critical considerations for long-term distribution network planning. A case study conducted in Malé (Maldives) [23] with an EV penetration rate of 50%, including electric two-wheelers, cars, and buses, showed that the EV load could be accommodated within the LV feeders, but it would lead to a substantial increase in the required generation capacity to handle the peak load. Optimised EV charging, along with local PV and BESS, could potentially help reduce additional costs and manage uncoordinated EV charging, specifically mitigating the sharp evening peak. In [24], the impact of EVs on a typical South Australian distribution network was investigated with varying EV penetration rates to determine the transformer loading profile. The results show that at 50% EV penetration, the network can adapt without compromising the recommended utilization rate of the transformer (80%). Notably, bus voltage variations did not exceed ±3%, but the power losses increased with higher EV connections.

However, in [14] it is demonstrated that significant grid impacts emerge at EV charging load penetration rates as low as 50%, with potential exacerbation by the inclusion of a few fast-charging stations. In addition, [25] reported that at just 20% penetration, uncontrolled charging using 15.7 kW Level 2 chargers caused voltage drops to 0.86 per unit (pu) in a suburban network, falling below the allowable lower limit of 0.94 pu. An evaluation of a medium-voltage (MV) network in Bangladesh [26] revealed that 20% of EV penetration was the critical limit for Level 2 chargers to avoid voltage violations and significant power losses. In [27], a case study of the distribution network in western Kentucky examined the thermal stress on the transformer, suggesting that transformer overload can be reduced by controlled EV charging schemes, which may slow down transformer ageing.

Existing studies exhibit inconsistencies in assessing the electrical power system impacts of widespread EV adoption, likely attributable to variations in substation configurations and feeder topologies across different study areas. Another contributing factor is the variation in the EV charger load modelling approaches employed, with studies adopting different methods for representing charging characteristics. These discrepancies underscore the need for a unified load model capable of accurately capturing EV charging behaviour for both steady-state and dynamic analysis applications.

1.2. Impact of Power Quality Disturbances on EV Charging

The existing literature highlights significant concerns regarding EV integration, particularly its potential to degrade distribution network voltage regulation and introduce multiple PQ disturbances. These PQ disturbances may subsequently affect EV chargers, forcing them to operate under suboptimal conditions. Such conditions can lead to unpredictable charging behaviour, complicating accurate EV load modelling for demand management and planning. In response, attention has been directed toward analysing the impact of common PQ issues on EV charging dynamics. EV chargers employ power electronic converters as core components for power transfer across different charging levels. The power electronic nature of these loads may lead to the assumption that they exhibit CP behaviour. However, their operation is unique compared to other power electronics equipment, such as PVs or wind turbines [28], and is far more complex than typical household electronic appliances. EV chargers incorporate advanced control schemes for constant current–constant voltage (CC-CV) charging phases, which often make it impracticable to maintain constant output power throughout the charging process. This complexity is further compounded by supply voltage variations during charging cycles. Consequently, there is a pressing need to investigate the behaviour of EV chargers when exposed to voltage variations.

Two simulation studies [29,30] revealed that prolonged voltage sags (lasting 10 s) directly extend EV charging cycles through reduced power delivery, indicating that the charger was unable to maintain CP operation. The comprehensive study detailed in [31] was undertaken to monitor and analyse the PQ data of a 6.6 kW Level 2 charging station connected to different EV models. The charger operated on a typical urban single-phase outlet with uncontrolled voltage magnitudes ranging from 208 to 240 V. The measurement results indicate that most EVs maintained a steady charging profile, except for two models that exhibited periodic on/off behaviour every 15 min. Notably, all the tested EVs consumed negative reactive power, indicating leading power factor behaviour, except Tesla vehicles, which consumed positive reactive power. Overall, the voltage and current THD, as well as PF, for all the vehicles remained within the recommended limits established by NERC/CMC/WECC [32]. Moreover, the power/voltage and current/voltage relationships of EVs were established, showing that they behaved as constant current load (CCL). These findings align with the experimental results in [33], where experimental PQ analysis under sustained voltage conditions also confirmed the CCL behaviour of various EV chargers.

Despite growing attention on PQ issues related to EV charging, the existing literature reveals a significant gap in long-term impact assessments. Prior studies have primarily focused on short-term root-mean-square (RMS) voltage disturbances (e.g., sags/swells) [29,30,34,35] or harmonic effects [36,37,38,39,40] on both Level 2 and Level 3 chargers. This gap is particularly relevant in the Australian context, where high penetration of renewable energy often results in significant voltage variation within distribution networks. Although Level 1 and Level 2 chargers dominate residential EV adoption, their performance under varying supply voltage magnitude remains insufficiently studied. Preliminary work completed by the authors in [33] is the only study offering a systematic investigation. To address this gap, this study extends the dataset from [33] and contributes in two further directions:

• Development of an expanded measurement database to enable more accurate static EV load modelling under real-world distribution network conditions.
• Calculation of conservation voltage reduction (CVR) factors for active power, which are critical for accurately assessing EV adoption electricity demand and impact on consumption, including economics aspects thereof.

1.3. Review of Modelling of EV Charging Load

As established in Section 1.2, EV chargers exhibit more complex operational behaviour than conventional CP load classifications [2]. This complexity stems from the hierarchical control architecture governing the charging process: software-mediated communication between the EV and EV supply equipment (EVSE) using pilot signal modulation, as shown in Figure 2, and hardware protection circuits based on electrical ratings and thermal constraints. These interdependent control systems create dynamic load characteristics that challenge traditional load modelling frameworks. Furthermore, in grids with high penetration of DERs, which frequently induce voltage regulation issues, EV charging power may exhibit adaptive variations or complete disconnection under extreme voltage magnitudes, triggering nonlinear charging responses and adding another layer of modelling complexity.

EV charging models can be categorised as either static or dynamic, each serving distinct analytical purposes. Static load models are particularly suitable for steady state analysis, including power flow studies and voltage regulation assessments. In contrast, dynamic load models provide critical insights into network behaviour during transient events, such as fault conditions, motor starting transients, and frequency excursions. This study specifically focuses on the development and relevant studies of the static EV load model to support distribution network planning and operational studies under steady-state operating conditions.

Numerous investigations have characterised EV chargers as constant power loads (CPLs) when assessing their influence on power system performance [11,21,24,42,43]. While this approach offers simplicity, it inherently neglects the voltage-dependent nature of EV charger power consumption, which is a significant limitation given the substantial voltage variations observed in modern distribution networks. To address this, recent studies [44,45,46,47,48,49,50,51] have adopted voltage-dependent load models to enable more accurate grid impact analyses, including both exponential and polynomial (ZIP) formulations. The exponential load model is represented by a nonlinear equation describing the relationship between power consumption and supply voltage, expressed by Equation (1).

$${\displaystyle \frac{P}{P_0}}=a_P{\left({\displaystyle \frac{V}{V_0}}\right)}^{n_P}+b_P$$

(1)

where

• $P_0$ is the nominal power at nominal voltage $V_0$;
• $n_P$ is the voltage exponent ($n_P=0, 1, 2$ for CPL, CCL, and CIL—constant impedance load, respectively);
• $a_P$ and $b_P$ are additional scaling coefficients.

Conversely, the polynomial ZIP load model is the expansion of the exponential model by combining CIL, CCL, and CPL components through a weighted summation. Both modelling approaches incorporate unknown parameters that require empirical estimation. These parameters can be determined through either

• Component-based methods—deriving values from manufacturer specifications and circuit characteristics, or
• Measurement-based methods—extracting parameters through regression analysis of operational charging data.

The choice between methods involves trade-offs between implementation complexity and model accuracy, with measurement-based approaches generally providing superior fidelity for dynamic grid studies. For example, the study detailed in [46] implemented an experimental setup comprising four distinct plug-in electric vehicles (PEVs) connected to a Level 1 charger, powered by a Staco Variable Transformer operating within a 100–130 V range. This configuration enabled systematic characterization of charging behaviour across different vehicle models under controlled voltage conditions. The derived ZIP load models revealed distinct charging characteristics among the PEVs: two vehicles exhibited near-CCL behaviour, one demonstrated multi-step CPL operation, and the fourth displayed hybrid CCL–CPL characteristics, though all of them had a power factor (PF) greater than 0.99. These findings corroborate the experimental observations from [33], where the majority of tested vehicles demonstrated CCL characteristics during Level 1 charging, while a subset exhibited hybrid CCL–CPL behaviour during Level 2 charging operations. Similarly, in [51], the authors developed a model for EV Level 2 charging that accounts for both voltage and SOC dependencies. Their findings indicate that the charger exhibits CPL behaviour when the voltage exceeds 1 pu while transitioning to a CCL characteristic for voltages below this threshold.

The study in [49] developed and compared EV ZIP load models against conventional CCL, CPL, and CIL representations. The analysis revealed that ZIP and CCL models produced comparable power flow results, both demonstrating moderately lower grid impact than the CPL model. This finding suggests that the CPL assumption may inadequately represent actual EV charging behaviour in distribution network analysis. The work described in [52] also found similar trends between the ZIP and CCL models for EV charging in the grid in terms of power losses, voltage profile index, and MVA capacity index. The consistency across studies [33,46,49,52] strengthens the evidence against using simplified CPL assumptions for accurate EV load modelling, highlighting that CCL characteristics may provide a more reliable representation for slow-rate charging scenarios.

The authors in [45,47] also adopted a measurement-based approach to determine unknown parameters for their ZIP load models. These works characterised the EV battery charger with varying supply voltage magnitudes of 180–230 V in a MATLAB simulation and in a laboratory experiment. The obtained ZIP load models were subsequently verified in a power flow analysis, showing that they could provide accurate values of power losses and bus voltages, and consumed less power than the conventional CPL models. These studies present two limitations that constrain their practical applicability. Firstly, the analysis excluded the higher voltage ranges (above 230 V), which is often observed in modern distribution networks. Secondly, the power electronics of the charger failed to incorporate protection circuits and advanced communications that characterise modern EV chargers. These simplifications particularly undermine the ability of the model to represent real-world charger behaviour during voltage excursions.

1.4. Paper Organization and Contribution

The introduction has demonstrated inconsistencies across studies and the adoption of distinct models for different charging levels, many of which mispresent EV charger loads as CPLs, potentially overlooking their true electrical behaviour. Therefore, this study aims to provide a definitive characterisation of EV charging loads at both Level 1 and Level 2, across the broad voltage ranges typically encountered in modern distribution feeders. These insights support the development of a more precise EV load modelling framework, which is also proposed by the authors to enhance accuracy in system planning and operational management.

The main contributions of the paper include the following:

• Comprehensive laboratory evaluation of the electrical characteristics of EV charging for a wide range of practical supply voltage magnitudes. Performance has been evaluated for a range of EVs models and several charger types, including Level 1 and Level 2.
• Development of static EV charging load models for both Level 1 and Level 2 chargers based on the experimental data, which accurately captures the relationship between supply voltage magnitude and EV charger electrical performance.
• Evaluation of the potential impacts of increasing penetration of EV charging loads when integrated into a case-study model of a PV-rich Australian distribution network. This modelling case study, implemented using the OpenDSS software package, also allows comparative analysis between the empirically informed EV charging load model and a CP EV load model.

The remainder of the paper is organised as follows:

• Section 2 presents the laboratory evaluation of EV charging performance under varying supply voltage magnitude conditions, along with the methodology and development of the empirical EV load model. The experimental procedures, results, and analysis are detailed.
• Section 3 describes the OpenDSS simulation model used for power flow analysis. It includes an assessment of the impact of increasing EV charger load on LV networks. This model also allows comparative evaluation of the empirical EV model and a CP load model and demonstrates the application of the model in an Australian distribution network.
• Section 4 concludes the paper and provides recommendations for future research to be completed over the next 12 months.

2. Experimental Evaluation of EV Charging Characteristics

2.1. Methodology

The charging characteristics of Level 1 and Level 2 chargers have been evaluated across a broad range of supply voltage magnitudes. The supply voltage was controlled using a California Instruments arbitrary power supply, while the EV performance parameters, including power parameters, RMS current, current THD (THDi), and displacement power factor (DPF), were measured using a Hioki 3198 PQ analyser, as depicted in Figure 3. Building upon the tests conducted in [33], the experimental setup was extended to include two additional EVs, bringing the total to five distinct EV models. These were evaluated using two Level 2 chargers, along with the Level 1 charger supplied with each of the vehicles.

The evaluated EV chargers are supplied to the Australian market with a nominal voltage of 230 V. The allowable voltage range for equipment connected to LV networks as specified by AS 60038 [7] and AS/NZS 3000 [53] is 230 V −15%/+10% (196 V to 253 V). As such, the chargers would be expected to be designed to operate as intended across this supply voltage range. Accordingly, the tested voltage range in this study encompasses the full expected operating range. The specifications of the chargers are as follows:

• Charger A—Level 1 Charger supplied with each vehicle with a rating of 10 A (2.3 kW at 230 V);
• Charger B—Level 2 Charger with a rating of 32 A (7.4 kW at 230 V);
• Charger C—Level 2 Charger with a rating of 32 A (7.4 kW at 230 V).

The EVs used for evaluation are commonly available in the Australian market, including the Nissan Leaf, BMW i3, Tesla Model 3, Tesla Model Y, and Peugeot E-Partner. These EVs were selected for testing due to their prevalence in the market. Together, these models accounted for 42.3% of total EV sales in Australia in 2024 [54]. The Level 2 chargers that were selected are from well-established suppliers. To ensure anonymity, the tested EVs are randomly referred to as EV1, EV2, EV3, EV4, and EV5. Figure 4 shows a photograph of the evaluated EVs. The combinations of the various chargers and EVs led to a total of 13 individual test scenarios. Performance assessment was conducted over a voltage range of 180 V to 260 V, in 10 V increments. For all the evaluations, the SOC of the EV batteries were maintained below 80% to prevent CV charging from influencing the collected data.

2.2. Analysis of Experimental Results

2.2.1. Level 1 Chargers

The outcomes for the Level 1 chargers tested are illustrated in Figure 5. The RMS current remained relatively stable across the applied voltage range, resulting in a proportional increase in active power as the voltage magnitude increased. THDi was also relatively stable across the applied voltage range, with EV1 having a THDi of 11%, whereas the others measured approximately 5%. The chargers consistently operated at unity PF, indicating minimal reactive power consumption. However, among the four EVs tested, one exhibited positive reactive power consumption, characteristic of an inductive load, while the others demonstrated capacitive characteristics.

Figure 6 illustrates the variation in charging current and active power as the input voltage increased from 180 V to 260 V. The RMS current exhibited minimal variation, with a maximum change of 3.58%, confirming the CCL behaviour. On the other hand, the active power significantly increased as supply voltage magnitude increased, clearly suggesting that the conventional CPL assumptions for Level 1 chargers may be overly simplistic.

2.2.2. Level 2 Chargers

For the Level 2 chargers, the active power profiles vary, as shown in Figure 7a, which corresponds to the RMS current responses observed in Figure 7c. As shown, the RMS current remained constant for only two combinations (Charger B–EV2 and Charger B–EV3). For the other test cases, the RMS current remained relatively unchanged at lower voltage levels but gradually declined as the input voltage increased. For example, EV1 is equipped with a standard 3.6 kW on-board charger, limiting its power draw to this value, despite both Level 2 chargers being capable of supplying 7.4 kW. As a result, power consumption initially increased but plateaued between 210 V and 260 V. Similarly, EV4 only accepted a maximum of 7 kW, causing a transition from CCL to CPL at 230 V with both chargers.

EV2, EV3, and EV5 could utilise the full charging capacity of both Chargers B and C. With Charger C, all three EVs exhibited hybrid CCL–CPL behaviour with the transition at 230 V as the charger reached its 7.4 kW limit. However, for Charger B, the same EVs maintained CC charging, with power consumption increasing up to 8.3 kW. This indicates that Charger B could deliver more power than its advertised specification. Similar to the case for the Level 1 chargers, the Level 2 chargers also demonstrated unity DPF, leading to insignificant reactive power consumption. Notably, EV3 and EV4, which exhibited negative reactive power with Level 1 charging, displayed positive reactive power consumption when charged using Level 2.

Figure 7e illustrates that the THDi remained below 5%. Notably, EV2 and EV5 exhibited a decreasing trend in THDi as the voltage increased, whereas the remaining vehicles maintained relatively stable THDi levels, around 3%.

Figure 8 presents the percentage variation in active power and RMS current for Level 2 charging. It is evident that power consumption generally increased with rising supply voltage, with observed increments ranging from approximately 30% to nearly 50%. The CPL assumption appears to be applicable solely in the case of EV1, whose onboard charger is constrained to a maximum capacity of 3.6 kW, thus limiting its power draw regardless of the charger’s available capacity.

2.2.3. CVR Factors

The general definition of the CVR factor is the ratio of the percentage change in a parameter of interest to the percentage change in voltage. For instance, the CVR factor for active power is calculated as follows:

$$CV{R}_P={\displaystyle \frac{\Delta P\left(\%\right)}{\Delta V\left(\%\right)}}$$

(2)

This expression quantifies how sensitive the active power response is to variations in supply voltage.

Table 2. Active power CVR factors.

Test Combination	CVRP
Level 1 Chargers
Charger A—EV1 Charger A—EV2 Charger A—EV3 Charger A—EV4	0.99 1.12 1.03 1.12
Level 2 Chargers
Charger B—EV1 Charger B—EV2 Charger B—EV3 Charger B—EV4 Charger B—EV5 Charger C—EV1 Charger C—EV2 Charger C—EV3 Charger C—EV4	0.37 1.03 1.05 0.63 0.68 0.28 0.73 0.73 0.64

summarises the calculated CVR_P across the entire voltage range for all the test combinations, including both Level 1 and Level 2 chargers. Overall, 6 out of the 13 tested combinations exhibited a CVR_P close to 1, suggesting CCL behaviour, in contrast to CPL characteristics, which are typically associated with a CVR_P of 0. In contrast, the remaining combinations, excluding EV1 under Level 2 charging, showed CVR_P values between 0.63 and 0.73, thereby confirming that these charging profiles cannot be accurately represented by a conventional CPL model.

2.3. Empirical EV Load Model

Based on the experimental results, it can be concluded that the often employed CPL representation is inadequate for characterising EV charger behaviour when subjected to voltage variations. The data consistently shows that Level 1 chargers are more accurately characterised as CCL loads across all the tested voltage levels. In contrast, Level 2 chargers were occasionally constrained by their rated power limits. Analysis indicates that Level 2 chargers typically transitioned from CC to CP mode at approximately 230 V (1 pu), where their rated power output was reached. Empirical power consumption curves for both Level 1 and Level 2 charging loads are proposed in Figure 9, with their DPF value set close to unity.

Figure 10 illustrates the percentage differences between the proposed empirical EV load models presented in Figure 9 and the measured data shown in Figure 5 and Figure 7. For the Level 1 charger, errors across all the EV/charger combinations did not exceed 2.5%. For the Level 2 chargers, errors generally remained below 4%, even in the most extreme cases. For EV1 with both Level 2 chargers, deviations of approximately 8–9% were observed across the lower voltage range (below 230 V); however, these differences can be considered acceptable. Notably, with Level 2 charger B, relatively larger errors of 9% and 14% were recorded for EV2 and EV3, within the higher voltage range of 250–260 V, where the EVs continued to operate in CC mode. It is important to note that such high voltage levels are not commonly encountered in typical distribution networks. It can be concluded that the output of the proposed EV charging load models closely aligned with the measured data in most cases, providing better accuracy in analysis when compared with the conventional CPL model, which will be validated in Section 3.

3. Impact of EV Load Model on Distribution Network Performance

3.1. Distribution Network Simulation Methodology

This section of the paper aims to evaluate the effects of EV adoption in a PV-rich distribution network. This evaluation proposes a model that considers the varying supply voltage magnitudes that will be observed along feeders due to both load behaviour and PV rooftop systems in the distribution network. The obtained load model from the conducted experiments will be utilised to perform an accurate power flow analysis of the potential impacts of high penetration of EV charger loads and includes a comparative analysis of the empirical load against CPL models.

In this study, the software packages used were OpenDSS for load flow analysis and MATLAB for data processing. In the OpenDSS simulation, a 24 h power flow for an LV network was implemented with a 1 min resolution, resulting in 1440 data points per day. The parameters used for the case studies will be presented in Section 3.1.1.

3.1.1. Australian Distribution Network Topology

In Australia, the three-phase distribution network typically comprises distribution lines, a distribution transformer, and shunt components, as depicted in Figure 11. Notably, for the medium-voltage (MV) network, a 3-wire arrangement is utilised, while the low-voltage (LV) feeder is configured as a three-phase 4-wire topology. Traditionally, power flow within the network is unidirectional, with power flowing from MV to LV networks. However, with the recent integration of renewable energy generation systems, reverse power flow can be observed within the network.

3.1.2. LV Network Modelling

To demonstrate the influence of EV charging stations on a PV-rich LV network, a model of an 11 kV/400 V delta-star distribution transformer feeding a 240 m radial overhead line, as depicted in Figure 12, has been used. There are 16 buses along the feeder, each of which supply 3 single-phase residential loads through 15 m cables. This would result in 16 residential loads per phase, equating to 48 houses in total. For the remainder of the case study, the EV charger penetration rate will be given in percentage of the total number of houses. For example, a 50% EV penetration rate will equate to 24 EV chargers connected to the network.

It is also noted that the buses are evenly spaced, giving a distance between each load connection point of 15 m. Each house is also equipped with a solar PV system with a rating of 5 kW peak. The type of cable used for the main feeder is 6/3.75ACSRGZ with line resistance and reactance of 0.583 Ω/km and 0.3523 Ω/km, respectively. Meanwhile, the cable used for connection from the bus to the residential load (service mains) is aerial bundled cable (ABC cable) with resistance and reactance of 0.842 Ω/km and 0.0853 Ω/km, respectively.

Table 3. LV network simulation parameters.

Components	Parameters
Residential Load	3 kWp at 0.95 pf lagging
Rooftop PV Generation System	5 kWp at unity pf
Substation Transformer	11/0.4 kV delta-wye, 200 kVA, 4% reactance on own base
Source Voltage	250 V (~1.09 pu)—50 Hz
Length of Main Feeder	240 m with 15 m between each load bus
Main Feeder Line Impedance	0.583 + j0.3523 Ω/km
Length of Service Mains Cable	15 m each phase
Service Mains Cable Impedance	0.842 + j0.0853 Ω/km

summarises the case study parameters.

Figure 13 describes the 24 h residential load, PV generation, and normalised EV charging profiles that have been adopted. It is observed that the typical residential load profile peaks during the evening hours when residents return home from work, a pattern that closely aligns with at-home EV charging behaviour obtained by data collected for 1000 EVs in [21].

3.2. Analysis of Simulation Results

3.2.1. Impact of EV Load Model on LV Network

To evaluate the performance of the developed empirical EV charger load model for both Level 1 and Level 2 chargers against the conventional CPL model, a worst-case scenario is simulated with a 100% EV penetration rate (assuming each of the 60 households owns one EV). Rooftop PV generation is excluded in this simulation to isolate the differences between EV charger load models.

Two simulation scenarios were carried out to assess the differences between the two EV charger models.

• The first scenario assumed that all EVs were charged using Level 1 chargers.
• In the second scenario, a mixed charger load was adopted: 50% of the EVs used Level 1 chargers, while the remaining 50% used Level 2 chargers. This distribution aligns with data reported by Evenergi for Ausgrid (Ausgrid is an Australian DNSP that supplies power to homes and businesses in Sydney, the Central Coast, and the Hunter Valley in New South Wales, serving over 4 million people) [55], which indicates that approximately 55% of EV owners in Australia own Level 2 chargers, while 41% rely on Level 1 charging. For the worst-case assessment, Level 2 chargers were connected to the most remote buses (Buses 9–16) of the network.

In both scenarios, key system metrics, including total active and reactive power flows and the voltage profile at bus 16, were monitored and compared. This comparison is shown in Figure 14, which includes emphasis on the evening peak hours. The results reveal that both load models demonstrated similar overall response trends; however, the CPL model consistently produced higher total active and reactive power than the empirical model.

For instance, during the evening peak under the 100% Level 1 charger scenario, the CPL model reached 288.5 kW of active power, while the empirical model peaked at 279.3 kW. In the mixed Level 1 and 2 charging scenario, the CPL model recorded a peak active power of 448.9 kW, compared to 410.3 kW for the empirical model. This means that treating EVs as CPLs resulted in an overestimation of the peak demand by 9.2 kW (3.3%) and 38.6 kW (9.4%), respectively. While the discrepancy between the load models may appear relatively small at the individual LV feeder level, when aggregated across all the distribution substations within a large distribution network, the cumulative impact will be substantial, leading to inaccuracy in system planning. Further, as a significant component of DNSP expenditure is related to servicing maximum demand, improper estimates of this quantity can have very meaningful impacts on DNSP expenditure, which, in turn, flow through to consumer energy costs. These impacts will start out small due to low EV penetration rates but will increase as EV take up increases.

Additionally, differences were observed in power losses between the two models, highlighting the limitations of relying solely on conventional CPL representations for EV load modelling. As detailed in

Table 4. Impact of EV load model on network analysis comparison.

Parameters	100% Level 1 EV Charging			50% Level 1–50% Level 2 EV Charging
	Empirical Load Model	CPL Model	Percentage Change (%)	Empirical Load Model	CPL Model	Percentage Change (%)
Peak active power demand (kW)	279.3	288.5	3.3	410.3	448.9	9.4
Peak reactive power demand (kVAr)	99.6	102.6	3.0	148.2	163.9	10.6
Peak active power losses (kW)	27.0	29.3	8.5	70.6	86.4	22.4
Peak reactive power losses (kVAr)	25.3	27.3	7.9	62.1	75.5	21.6
Lowest voltage magnitude (V)	204.1	202.5	−0.8	180.2	173.9	−3.5

, the CPL model significantly overestimated losses in the mixed charging scenario by 22.4% for active power losses and 21.6% for reactive power losses.

To better understand the impact of EV load models on network analysis across varying penetration levels, simulations were conducted for EV penetration rates ranging from 0% to 100% in 12.5% increments. For each penetration level, a 24 h simulation was performed, focusing on recording the average power consumption, as well as its minimum and maximum values for calculating the transformer loading factor, as illustrated in Figure 15.

The results reveal that differences between the EV load models are minimal when penetration is below 60%. However, beyond this threshold, discrepancies emerge, with the conventional CP EV load model consistently overestimating both the average and peak demand. This trend is likely to be further amplified in larger or more geographically extended LV networks, where longer feeder lengths and higher load densities exacerbate the sensitivity to load modelling inaccuracies. Additionally, Figure 15 also indicates that transformer overloading can occur during the evening peak even at moderate EV penetration levels, such as 37.5%, raising concerns related to long-term thermal stress and accelerated aging of distribution transformers. These findings highlight the need for more accurate EV load models in network planning to avoid insufficient infrastructure upgrades.

Voltage regulation can be achieved through adjustments to the source voltage, particularly in heavy loading conditions. Given that the differences between the EV load models are negligible at low penetration rates, this study examines how variations in source voltage affect network performance in scenarios with 50–100% EV penetration. To this end, the substation transformer secondary voltage was varied from 240 V to 250 V, and the corresponding total active power, power losses, and actual load demand were recorded, as presented in Figure 16.

The findings reveal distinct differences in outcomes for the two load models. When EVs are represented as CPLs, despite constant active power demand from the chargers with the increased source voltage, the total active power decreased. This decrease can be attributed to the decrease in active power losses. Conversely, the empirical EV load model exhibits a voltage-dependent characteristic. As the source voltage increases, both the total active power and actual load demand rise, even though losses decline. This contrasting trend highlights a key limitation of the CPL approach, which is its inability to capture the voltage-dependent nature of real EV charging loads.

3.2.2. Verification of the Empirical EV Load Model in a Typical Australian LV Network

In this section, the proposed empirical EV load model is employed to investigate how increasing adoption of EVs affects an LV network with high rooftop PV integration. The selected EV penetration rates are 0% (base case), 25%, 50%, 75%, and 75%, which are respectively equivalent to 0, 12, 24, 36, and 48 EVs connected to the network. To assess the worst-case scenario, at all EV penetration rates, all the EVs are placed at the most remote buses in the feeder. For example, at a 25% EV penetration rate, a total of 12 EVs are uniformly distributed across buses 13 to 16. The results of this investigation are shown in Figure 17.

In general, Figure 17a,b demonstrate that increasing EV penetration leads to higher active and reactive power flows throughout the network, primarily due to the dominance of nighttime charging patterns. This results in increased peak demand, leading to a load that exceeds the transformer’s rating when the EV penetration rate reaches 50%. At this threshold, transformer overload occurs for nearly two hours post-8 p.m. At 100% EV penetration, the transformer is overloaded by almost 50%, with the condition persisting for nearly 5 h. The heightened peak demand also exacerbates voltage drops during these high-load periods, as depicted in Figure 17c. In contrast, midday PV generation leads to voltage rise and reverse power flows. The temporal mismatch between PV output and EV demand means a significant portion of solar generation is unable to offset nighttime EV load, thereby reducing renewable integration effectiveness and intensifying voltage regulation issues.

In both the base case and at 25% EV penetration, the voltage at bus 16 approached the upper allowable limit. In contrast, nighttime undervoltage conditions progressively worsened with increasing EV penetration. At 50% penetration, the voltage at bus 16 reached the lower threshold, and at 75% and 100% penetration, it dropped below the acceptable limit. This issue is likely to be exacerbated if EV charging loads are represented using the CPL model rather than the empirical model, as shown in Figure 14, where the CPL representation resulted in greater voltage drops in both simulation scenarios.

Additionally, in terms of voltage regulation, the undervoltage issue is unlikely to be effectively resolved by simply increasing the substation transformer tap settings for two main reasons. First, the source voltage was set at 250 V, which is already in close proximity to the standard upper limit, leaving a minimal margin for upward adjustment. Secondly, as demonstrated in Section 3.2.1 using the empirical EV load model, increasing the source voltage will lead to higher load demand, increasing network voltage drops in turn. Overall, the results indicate that uncontrolled integration of EV chargers into LV networks will intensify voltage regulation challenges.

4. Conclusions

This study has revealed that the widespread assumption of representing EV charger loads as CPLs is an oversimplification. This misconception can lead to inaccurate demand forecasting, leading to ineffective planning. Through extensive and controlled laboratory experiments, which examined the EVSE–EV charging process under varying voltage conditions, the actual characteristics of EV charging loads were identified. Specifically, EVs charged using Level 1 chargers consistently exhibited CCL behaviour, while a hybrid CCL–CPL pattern was observed for Level 2 chargers.

These findings enabled the development of accurate static EV charger load models, which were subsequently validated through simulations using an Australian distribution network. The proposed empirical models accurately captured network parameters, including active and reactive power flows, losses, and voltage profiles, which the conventional CPL model failed to provide accurately. For example, under identical network conditions, the CPL model consistently overestimated network losses and demand and overexaggerated voltage drops, resulting in misleading information. Furthermore, while the empirical model appropriately reflected the voltage-dependent increase in load demand as the source voltage rose, the CPL model failed to capture this critical voltage-dependent behaviour. Additionally, validation in a PV-rich Australian distribution network further revealed that escalating EV penetration caused substantial voltage drops and transformer overloading. These issues are likely to be mischaracterised and could be insufficiently addressed if analysis relies on the CPL model, due to its inability to respond correctly to high degrees of voltage variation. Such misrepresentations may compromise network reliability and result in suboptimal investment planning. In contrast, the empirical model offers DNSPs more realistic insights into network demand margins and infrastructure upgrade requirements. This, in turn, enables optimised load shaping and active and reactive power demand inputs to planning functions. Furthermore, in high-penetration scenarios, the simplicity of the proposed load models may help overcome technical barriers to analysis, such as reducing simulation computational burden, enabling straightforward large-scale simulations involving thousands of EVs, and facilitating streamlined integration with traffic data.

For future work, similar investigations can be extended to Level 3 chargers, focusing on their response to supply voltage variations within the same range tested for Level 1 and Level 2 chargers (180–260 V). However, as Level 3 chargers are typically supplied by three-phase AC, both balanced and unbalanced test scenarios should be considered. Furthermore, given the harmonics-related issues identified in the literature for DC fast chargers, further evaluation of both the harmonic emissions of DC chargers across various EV charging scenarios, as well as the influence of distorted supply voltage waveforms could be investigated. The resulting data would support the development of accurate load models for fast-charging scenarios. Additionally, the proposed modelling framework can be expanded to include V2G operations, which introduce bidirectional power flow and further impact the behaviour of modern distribution networks.