Key Technologies of Novel Power Systems: From Optimal Scheduling to Intelligent Operation and Maintenance Review

As the goals of peak carbon emissions and carbon neutrality continue to progress, the power system is rapidly evolving toward a cleaner, low-carbon, and more efficient structure with increasing integration of renewable generation, which brings new research concerns regarding low-inertia operational characteristics and relay protection adaptation for renewable dominated power grids [1,2]. Nevertheless, the inherent randomness, intermittency, and strong output volatility and uncertainty of renewable energy impose unprecedented pressure on traditional distribution networks in terms of scheduling flexibility and power supply reliability. Under such circumstances, digital twin-enabled situational awareness technologies provide promising tools to perceive these complex grid operating states [3,4], while active network reconfiguration offers an effective way to cope with the fluctuation-induced operational risks [5]. To further improve grid adaptability to high-penetration renewable energy and achieve source–grid–load–storage-coordinated optimization, multiple technical pathways have been investigated. Beyond typical flexible regulation resources represented by energy storage systems (ESSs) and soft open points (SOPs), demand response also acts as an effective approach to alleviate grid operational stress [6,7]. As power electronic-based regulation facilities, SOPs mitigate the adverse impacts of large-scale renewable integration on distribution networks: they regulate feeder power through fully controlled power electronic devices to lower risks of voltage violation and reduce network losses [8,9,10]. Diverse energy storage forms, represented by compressed CO₂ energy storage and hybrid mobile stationary energy storage, possess flexible regulation and energy time shift capacity. These resources can support both normal system operation and multi-resource coordinated service restoration under contingencies, further enhancing the overall resilience of power systems [11,12,13,14]. Apart from flexible resource scheduling, the stable operation of novel power systems relies on a complete suite of supporting technologies, including low-carbon optimization for electricity–heat–gas–hydrogen integrated energy systems [15], data-driven power forecasting, intelligent monitoring and fault diagnosis for power equipment, as well as relay protection and stability improvement for interconnected power grids.

To tackle these practical challenges, substantial research efforts have been devoted to microgrid economical scheduling and distribution network power electronic regulation, forming an important technical research branch. To cut the comprehensive running expense of hybrid microgrids assembled with wind farms, PV arrays, conventional thermal generators and diversified energy storage devices, Dong and Lee designed an upgraded second-order oscillating chaotic mapping particle swarm optimization (SCMPSO) algorithm with improved convergence performance (contribution 1). To mitigate voltage fluctuation and violation risks caused by high-penetration distributed photovoltaic integration in AC-DC hybrid distribution networks, Zheng et al. proposed a spatio-temporal hierarchical distributed voltage-optimization framework. The work realizes coordinated scheduling of slow discrete regulation resources and fast inverter-based reactive power assets for network-level voltage control (contribution 2). In order to cope with unpredictable fluctuation of combined wind and photovoltaic generation, Lin et al. constructed a data-driven Wasserstein uncertainty set and put forward a two-tier allocation scheme to reasonably arrange flexible active and reactive resources (contribution 3). Focusing on the transient unequal current among parallel-connected converters, Fu et al. analyzed the root cause stemming from discrete feedback coefficients and proposed an optimized virtual impedance droop regulation strategy to boost instantaneous current sharing capacity (contribution 4). Without requiring complete network topology and impedance parameters, Li et al. invented a data-driven predictive control (DeePC) approach to adjust bus voltages by coordinating the reactive potential of PV inverters and battery storage units (contribution 5). For the problem of inaccurate electromechanical simulation of large-scale renewable power bases, Zhao et al. created a full calibration framework relying on rapid-response generator testing to finish error positioning and parameter revision of station simulation models (contribution 6). Despite the decent simulation performance of the above-mentioned optimization and control schemes, most approaches have not been fully verified with actual field data, restricting their large-scale promotion in real distribution grids with complex operating conditions.

Low-carbon scheduling research targeting electricity–heat–gas–hydrogen integrated energy systems (IESs) has grown rapidly under carbon emission restriction policies. Cui et al. set up a multi-source–grid–load interactive operating framework alongside graded demand response rules and adopted the Shapley–Topsis evaluation method to quantify the low-carbon benefits brought by segmented user participation (contribution 7). Taking the long-cycle operating traits of electrolytic aluminum factories into scheduling design, Yang et al. made use of variational mode decomposition to classify load response characteristics and matched thermal power with energy storage to locally consume surplus green power (contribution 8). After comparing the cost gaps of pipeline, tank truck and liquid hydrogen delivery modes, Liu et al. established mixed integer programming formulas to optimize multi-energy flow distribution within hydrogen-containing comprehensive energy parks (contribution 9). On the basis of Nash bargaining game theory, Shao et al. applied the alternating direction method of multipliers (ADMM) to realize decentralized energy trading among geographically separated energy parks (contribution 10). Sun et al. combined master–slave game with carbon capture plus power-to-gas techniques to formulate low-carbon operational plans under carbon trading constraints (contribution 11). Incorporating green hydrogen quota market and distribution robust optimization methods, Yang et al. further built multi-objective scheduling models to balance economic profit, carbon reduction and anti-interference capacity against renewable volatility (contribution 12). Most existing IES models simplify real-time price swings of hydrogen and fossil fuels, leading to visible gaps between simulated results and practical market implementation requirements.

Driven by continuous advancement in deep learning techniques, numerous novel prediction frameworks are developed to satisfy diversified power forecasting needs. To precisely calculate grounding current at substation sites, Zhang et al. put forward a frequency-enhanced Transformer model that splits raw data into trend and fluctuation components via frequency-domain decomposition (contribution 13). Aiming at the predictive maintenance demands of high-voltage direct current (HVDC) converter cooling equipment, Sun et al. constructed a dynamic spatial–temporal graph neural network (GNN) to mine spatial coupling and temporal changing rules of cooling parameters (contribution 14). For lightweight deployment on wind field edge terminals, Chen et al. developed a normalized feature learning model (NFLM) with streamlined core modules for short-term wind power prediction (contribution 15). By combining grey relational screening of impact factors and frequency sequence splitting, Jin et al. proposed grey relational analysis-frequency enhanced decomposition Transformer (GRA-FEDformer) to enhance medium- and long-term bulk load prediction accuracy (contribution 16). For community-level heating, ventilation and air conditioning (HVAC) aggregated load probabilistic forecasting, Pan et al. designed multi-scale cross-variable Transformer equipped with dual attention modules to capture complicated cross-factor load variations (contribution 17). Although these Transformer-based algorithms outperform traditional statistical prediction tools, the opaque black-box property of deep neural networks limits their widespread onsite application in power industry.

Intelligent monitoring and fault identification based on computer vision and optimized intelligent algorithms become mainstream solutions for power equipment health management. Using part affinity fields (PAFs) feature extraction technology, Huo et al. built a three-layer computational framework to finish structural posture recognition of various high-voltage transmission towers (contribution 18). Leveraging meta-learning optimization thinking, the same research team built meta-learning YOLO (MA-YOLO) integrated with adaptive feature fusion units to identify wildfire near power lines under scarce sample circumstances (contribution 19), and later upgraded masked 2D Transformer–high-resolution network (Mask2Former-HRNet) with deformable convolution to lower insulator defect detection error rate (contribution 20). Ji et al. adopted contrastive learning and dual-path classification architecture to ease sample imbalance when judging the health state of transmission bolts (contribution 21). To reduce large-scale blackout risks of urban distribution systems under typhoon disasters, Zhu et al. developed a resilience-oriented spatio-temporal co-optimization framework that couples power grids, transportation networks and distributed energy resources. It implements pre-disaster resource pre-deployment and post-disaster multi-source coordinated load restoration at system level (contribution 22). Combining upgraded YOLO11 and Segment Anything Model (SAM), Dai and Fang developed an automatic assessment system to realize pixel-level quantitative measurement of armature component defects (contribution 23). Sun et al. constructed invariant mixture of experts (InvMOE) multi-task learning framework to spot diversified hidden faults inside converter stations (contribution 24). To handle the output uncertainty of distributed renewable energy within multi-regional integrated-energy-system clusters, Cui et al. established a distributionally robust collaborative dispatch model considering multi-energy sharing of electricity, heat and oxygen, and formulated an internal cluster energy-pricing mechanism to balance system robustness and operational economy (contribution 25). Facing renewable-output uncertainties in transmission–distribution coupled power grids, Chen et al. constructed a sequential bi-layer optimization framework based on information-gap decision theory for aggregated electric-vehicle dispatch. The method minimizes system supply cost and distribution-network nodal loss without requiring complete probability information for renewable generation (contribution 26). Aiming at low coordination efficiency of multi-energy park-level integrated energy systems under high renewable penetration, Yue et al. put forward a cluster-division approach based on improved Louvain algorithm with demand-response potential evaluation, and carried out cluster-oriented optimal dispatch to enhance local renewable-energy consumption (contribution 27). Insufficient real malfunction datasets collected from field operation cause poor generalization performance of most intelligent diagnosis algorithms under unfamiliar working scenarios.

Beyond equipment condition monitoring and fault diagnosis, flexible resource scheduling represents another critical approach to accommodate high-penetration renewable energy in distribution networks. Mobile energy storage systems (MESSs) and distributed electric vehicles (EVs) act as typical flexible regulation resources to improve distribution network economic performance and voltage stability. Ji et al. established a joint scheduling framework linking MESSs and SOPs, making full use of the movable features of mobile storage to narrow peak–valley load difference and restrain abnormal voltage drift (contribution 28). Taking actual urban road layout into modeling constraints, Miao et al. blended Dijkstra path optimization and non-dominated sorting genetic algorithm III (NSGA-III) to arrange operating routes and power output of hybrid energy carriers (contribution 29). After summarizing the temporal–spatial charging distribution rules of EV users, Sun and Yu optimized the coordinated running scheme of movable storage and scattered electric cars to raise station income and cut users’ charging expense simultaneously (contribution 30). On the basis of source-load clustering processing for distributed charging piles, Zhan et al. set multi-dimensional optimization targets and adopted non-dominated sorting genetic algorithm II (NSGA-II) to design interactive charging control tactics for improving grid power quality (contribution 31). To scientifically decide the location and rated capacity of large-scale EV charging stations amid volatile new energy output, Liu et al. utilized Gaussian mutation multi-objective thermal exchange optimization (MOTEO-GM) to suppress voltage oscillation and lower overall network loss (contribution 32). Most existing coordinated scheduling schemes ignore users’ random charging willingness, leading to obvious gaps between calculated optimization results and real implementation effects.

Progress in relay protection and interconnected grid stability optimization provides reliable guarantee for high-new-energy grids. Novozhilov et al. utilized zero-sequence mutual inductor and annular measuring converter to develop full-selective single-phase earth fault protection applicable for bundled cable groups (contribution 33). Considering unique power output features of grid-connected power-electronic equipment, Liu et al. put forward an iterative short-circuit current computation approach to support setting revision of protection devices for high-voltage grids with massive renewable access (contribution 34). Benbouhenni and Bizon constructed cascade neural network dual-loop control to stabilize output power and restrain torque fluctuation of doubly fed induction generator (DFIG) wind turbines (contribution 35). By introducing conditional value at risk (CVaR) as the risk evaluation indicator, Wang et al. built a risk-oriented tie-line power optimization model to reduce cross-border power over-limit risks and boost overall renewable absorption capacity of interconnected power grids (contribution 36). Issabekov et al. developed a preselective ground fault detection approach relying on vector reactive asymmetry for medium voltage distribution networks (contribution 37). Nevertheless, most existing protection and stability strategies are validated under ideal fault scenarios, and their adaptability toward complex operating conditions dominated by power-electronic-interfaced renewables still requires further field verification.

Overall, these papers systematically build a complete technical chain covering source-side multi-energy low-carbon scheduling, midstream intelligent prediction and equipment diagnosis as well as grid-side protection upgrade. Nevertheless, existing studies suffer from multiple common limitations. On the one hand, numerous models oversimplify practical factors including volatile energy market prices and stochastic user behavioral willingness. On the other hand, deep-learning-driven prediction and diagnosis techniques are confronted with black-box interpretability issues and the scarcity of real-world fault samples. Most importantly, the majority of technical solutions are developed within idealized simulated environments without sufficient practical-operation-data support, which constitutes a major bottleneck for real-world engineering deployment of relevant technologies. Focusing on recurring voltage-violation challenges in high-renewable-penetration distribution feeders, Zhang et al. proposed a dual-agent-driven dynamic intelligent management scheme adopting mixed-integer quadratic programming joint optimization, which achieves high-efficiency voltage violation mitigation for practical distribution-network operation (contribution 38).

Looking to the future development of new power systems, five core research directions deserve continuous exploration. First, establish refined full-chain multi-energy coupling models covering hydrogen production, storage and diversified transportation links to mitigate unrealistic energy interaction assumptions embedded in current modeling work. Second, develop emergency control and rapid restoration technologies to strengthen grid anti-risk capability against extreme meteorological disasters, which has gained growing attention under the frequent occurrence of extreme weather events. Third, explore physics-informed AI algorithms to improve the interpretability of black-box deep-learning-based prediction and diagnosis models. Fourth, accelerate digital twin industrialization to realize full-life-cycle visual monitoring, predictive maintenance and closed-loop simulation–verification for power equipment, so as to narrow the gap between simulation analysis and field practice. Fifth, promote coordinated formulation of carbon trading, green certificate and demand response incentive policies, and further incorporate stochastic user willingness of flexible resources, to realize combined drive of market mechanism and technical optimization for low-carbon grid construction.