Efficient and accurate short-term load forecasting (STLF) is critical to ensuring the safe, stable, and economical operation of power systems. To address the limitations of existing load forecasting models—specifically their inadequate capture of characteristics during peak and off-peak periods and relatively low prediction accuracy—this study proposes a two-stage integrated STLF architecture. This approach combines baseline forecasting with localized adjustments during critical periods, implemented using a temporal convolutional neural network (TCN) integrated with a convolutional block attention module (CBAM) and a bidirectional gated recurrent unit (BiGRU) model. First, historical load data are clustered using the k-medoids algorithm. Representative daily load curve features and corresponding historical data for each load type are reconstructed and input into the prediction model to generate baseline predictions. Subsequently, a shared CBAM-TCN-BiGRU model architecture independently trains peak and valley load predictions to accurately capture critical period characteristics. High-precision peak-valley forecasts from the sub-models replace the corresponding segments in the baseline curve, and optimized final predictions are achieved through key-point correction and global fusion. Finally, the New England-ISO load dataset is used for case validation. Results demonstrate that the final model predictions, refined by the peak-valley optimization strategy, achieve further accuracy improvements compared with the baseline prediction model.

Multi-objective task scheduling in electric vehicle battery software updates provides a key technology for the efficient and intelligent management of the battery's entire lifecycle in the era of software-defined vehicles. However, existing scheduling methods have the following main limitations: i) mainstream approaches treat task allocation and scheduling as separate processes, focusing only on local optimization during allocation and failing to prevent systemic bottlenecks globally; ii) most studies rely on simple weighted sums or single-objective opti mization, which makes it difficult to effectively characterize and optimize the Pareto-optimal trade-off between updating efficiency and resource load balancing, thereby limiting the flexibility and quality of decision-making. To address these issues, this paper proposes a physical potential-assisted variable neighborhood search (CPA-VNS) framework tailored to this problem. The framework first introduces a physics-inspired connection potential model that quantifies multidimensional factors into system potential. By solving a mixed-integer programming model, an optimized initial task allocation scheme is generated, improving solution quality from the outset. Then, within a Pareto optimization framework, a variable neighborhood search algorithm with a customized neighborhood structure is designed. This algorithm efficiently explores the solution space under efficiency-fairness constraints, systematically mapping the entire Pareto front, and providing decision-makers with a set of high-quality non-dominated scheduling solutions. Comparative experiments against seven benchmark algorithms across six simulation scenarios of varying scale and complexity demonstrate that the proposed framework achieves the lowest makespan and load variance in nearly all scenarios. Furthermore, Pareto analysis proves that sacrificing approximately 3% in efficiency can yield an improvement of over 50% in fairness.

The integration of new rural industries, particularly photovoltaic greenhouses, facilitates flexible regulation of distribution networks and reduces investment costs. This paper proposes a distribution network line sizing method that takes into account the aggregation regulation of photovoltaic greenhouses. It addresses the high costs associated with current distribution network planning and the insufficient exploration of the power regulation potential of photovoltaic greenhouses. A method for multi-stage energy demand modeling and the power regulation potential evaluation of photovoltaic greenhouse is proposed, taking into account the dynamic changes in crop demand for environmental factors throughout the entire growth cycle and the coupling effects among these factors. Based on this, a Stackelberg game model is established, which considers the participation of photovoltaic greenhouse aggregator in the regulation and control of the distribution network. In this model, the distribution network and photovoltaic greenhouse aggregators engage in a game with the objectives of minimizing line investment and maximizing economic benefits to achieve a balance of benefits among multi-agent. A genetic algorithm and the commercial software CPLEX are employed to solve the model. The simulation results demonstrate that the proposed method enhances the economic viability of the planning scheme.

Real-time peer-to-peer (P2P) electricity markets dynamically adapt to fluctuations in renewable energy and variations in demand, maximizing economic benefits through instantaneous price responses while enhancing grid flexibility. However, scaling expert guidance for massive personalized prosumers poses critical challenges, including diverse decision-making demands and lack of customized modeling frameworks. This paper proposed an integrated large language model-multi-agent reinforcement learning (LLM-MARL) framework for real-time P2P energy trading to address challenges such as the limited technical capability of prosumers, the lack of expert experience, and security issues of distribution networks. LLMs are introduced as experts to generate personalized strategy, guiding MARL under the centralized training with decentralized execution (CTDE) paradigm through imitation learning. A differential attention-based critic network is designed to enhance convergence performance. Experimental results demonstrate that LLM generated strategies effectively substitute human experts. The proposed multi-agent imitation learning algorithms achieve significantly lower economic costs and voltage violation rates on test sets compared to baselines algorithms, while maintaining robust stability. This work provides an effective solution for real-time P2P electricity market decision-making by bridging expert knowledge with agent learning.

The expansion of data centers (DCs) drives a sustained increase in electricity demand and associated water withdrawals at generation sites. These withdrawals occur at generation sites and are virtually allocated to demand based on network power flows. Consequently, the actual water footprint of a specific load varies dynamically with generation dispatch and network conditions. Existing approaches typically rely on static statistical accounting to quantify these water footprints. However, such static methods fail to capture how dispatch optimization and workload relocation dynamically affect water withdrawals. As a result, static statistical accounting approaches remain decoupled from the optimization process, rendering them incapable of guiding workload relocation or power dispatch to mitigate water stress. To address this limitation, this paper develops an operational electricity-computation-water (ECW) nexus framework that internalizes virtual water impacts directly into power system dispatch. The framework represents dispatch optimization as a differentiable optimization layer embedded within a deep learning architecture, enabling efficient end-to-end learning of coordination policies while preserving operational feasibility. Combined with fixed-point coordination, the framework enforces consistency between virtual water attribution and physical generation-side withdrawals. Case studies on the IEEE 30-bus and 118-bus test systems demonstrate reliable convergence, exact power-water consistency, and reductions of approximately 3-5

ABSTRACT In recent years, extreme cold disasters have occurred frequently worldwide. The security of rural distribution network is threatened. The current research on enhancing the resilience of distribution network fails to consider the changes in the operating characteristics and efficiency of equipment under extreme cold disasters. To address this, this paper uses hydrogen‐integrated energy systems (HIES) to enhance the resilience of rural distribution network under extreme cold disasters. The method considers ensuring the power supply of rural lifeline load. A power support capability evaluation model of HIES is established. The model takes into account the impacts of low‐temperature conditions on the operational characteristics and efficiency of hydrogen energy equipment and photovoltaic equipment. An energy demand model considering livelihood security lifeline load, public service lifeline load and cultivation‐breeding lifeline load under extreme cold disasters is established. A rural distribution network resilience enhancement model is developed. Its goal is to maximise the restored energy demand of lifeline load. The example shows that the lifeline load recovery rate is increased by 38.98% and the primary load recovery rate is increased by 5.44% using the method proposed in this paper.

The tightening of global environmental regulations is accelerating the adoption of Electric Trucks (ETs) in logistics; however, the integrated optimization of their routing, charging schedules, and cargo loading presents a formidable challenge. This complex coupling inherently introduces a Generalized Multiple Knapsack Problem and results in a Mixed-Integer Linear Programming model that is typically NP-hard and computationally demanding for standard solvers. To address this, we propose a novel Graph-Benders algorithm based on an exact algorithmic framework that strategically decomposes the problem into an ET routing master problem and a Vehicle-to-Grid (V2G) scheduling subproblem, reformulated within a graph-based modeling approach. This structure ensures all delivery tasks are fulfilled while tightening the feasible region for grid coordination. By synergizing feasibility and optimality cuts from the linearly relaxed subproblem with no-good cuts to address integer infeasibility, the method efficiently eliminates infeasible solutions,yielding a high-quality near-optimal solution, which is at least as good as the optimal V2G schedule for the best feasible routes. Through experiments in various geographic scenarios and split-delivery tasks, the algorithm demonstrates its versatility and achieves a breakthrough in efficiency without sacrificing solution quality compared to Gurobi. In large-scale tests, for instances with 19,585 decision variables, the proposed Graph-Benders decomposition yields a superior, high-quality near-optimal solution in merely 28 seconds, outperforming Gurobi’s 10,000-second time-limit result. Notably, even at a scale of 127,489 variables, the algorithm yields high-quality near-optimal solutions in only 130 seconds.Consequently, the proposed algorithm effectively tackles the strong coupling of large-scale ET routing and V2G coordination, offering a powerful tool for practical low-carbon logistics deployment.

Existing highly intelligent, large-scale, and industrialized facility agricultural parks (FAPs) often exhibit uncoordinated electricity consumption patterns, which not only lead to line overloads and voltage violations in rural distribution networks, but also increase the operating cost of FAPs. To address the insufficient characterization of crop growth characteristics in FAPs and the difficulty for flexible distribution networks to simultaneously ensure secure, economical operation and satisfy FAP crops growth requirements, this paper proposes a bi-level optimal dispatch method for distribution networks and FAPs considering the regulation capability of power electronic devices and crop coupled growth characteristics. First, the light-thermal-carbon coupled growth mechanisms of FAP crops are analyzed, and a light-thermal-carbon load balance model together with coupled crop growth indicators is established to quantitatively characterize crop growth demands. Second, a bi-level optimal dispatching model for the flexible distribution network and FAP is formulated. In the upper-level model, network losses, peak-valley difference, and line overload are minimized subject to power balance constraints and equipment regulation limits. In the lower-level model, the operating cost of the FAP and crop growth performance indicators are jointly optimized to determine the optimal electricity consumption strategy. Through iterative coordination between the upper and lower levels, collaborative optimization between the flexible distribution network and the FAP is achieved. Finally, case studies demonstrate that the proposed method can effectively reduce the operating cost of FAPs while ensuring normal crop growth, and simultaneously enhance the operational security and economic performance of the distribution network.

Island distribution networks suffer from weak external support, difficulties in renewable energy integration, high carbon emissions, and vulnerability to extreme disasters. To address these challenges, this study proposes an optimal hydrogen energy storage (HES) configuration method that balances economic performance, resilience, and low-carbon operation. The research develops physics-based models of electrolyzers, hydrogen storage tanks, and fuel cells, and designs operating strategies for both normal and extreme scenarios to capture renewable utilization, emergency supply capability, and Carbon emission reduction capacity. On this basis, a tri-objective optimization framework is constructed using generalized Nash equilibrium (GNE) theory, where device capacities are modeled as independent strategic variables and the three objectives are defined as utility functions of different players. The model is solved using a mathematical program with equilibrium constraints (MPEC), which transforms each objective into KKT conditions and enables stable, interpretable solutions under coupled physical and resource constraints. Case studies on a typical coastal island demonstrate that, compared with electrochemical storage, HES achieves higher renewable absorption and carbon reduction benefits under normal operation, while significantly enhancing resilience during extreme events. Results also reveal nonlinear couplings and high sensitivity among the three objectives, where small capacity adjustments lead to substantial fluctuations in performance. The findings confirm that the proposed GNE-based method effectively captures complex multi-objective interactions, avoids imbalanced configurations caused by traditional weighted or hierarchical approaches, and provides theoretical and practical guidance for resilient and low-carbon development of island distribution networks.

While multi-device coordination significantly enhances the capability for comprehensive power quality management in distribution networks, its practical effectiveness is severely hindered by communication imperfections such as delays, packet loss, and intermittent interruptions. To address these challenges, this paper proposes a communication-resilient coordinated power quality management framework based on multi-agent reinforcement learning (MARL). The non-ideal communication process across the MARL training and execution phases is considered. Incomplete or unreliable data caused by packet loss and interruptions is recovered via data-driven spatiotemporal reconstruction, while impacts of communication delays are mitigated through a timestamp-verified event-triggered execution mechanism. Furthermore, to mitigate the policy learning bias induced by training data disturbances from communication imperfections, a novel MARL algorithm is proposed. By modeling the distribution rather than only the expected value of decision outcomes, the proposed method captures uncertainty from noisy and incomplete data, improving robustness to communication impairments. Extensive simulations on a modified IEEE 123-node system demonstrate that the proposed method significantly outperforms baselines under imperfect communication conditions and maintains robust effectiveness under severe communication impairments.

Electric vehicles (EVs) significantly influence modern urban energy systems, posing both challenges and opportunities in sustainability and efficiency. This research tackles these challenges by integrating Electric Mobility on Demand(EMoD) into the power distribution network, focusing on reducing the carbon footprint and increasing the use of renewable energy. A multi-flow model was developed, covering power, passenger, and vehicle rebalancing flows, which facilitates precise energy distribution and resource management. Additionally, a mixed-integer second-order cone programming model incorporates carbon-aware routing to strategically schedule shared EVs, optimizing their charging and discharging times in sync with renewable energy availability and aiming for maximum carbon emission reductions. Findings reveal a significant decrease in carbon emissions, amounting to 232.5 kg, and an increase in renewable energy usage by 612.1 kW, attributed to the optimization of EV charging schedules to align with renewable energy production. Furthermore, by employing power flow tracing techniques, the research crucially tracks energy outputs and carbon emissions across the network which is vital for refining energy distribution strategies and ensuring that the EMoD operates within optimal environmental parameters. These methodologies collectively enhance environmental impacts and grid efficiency, underscoring their importance in future EMoD system planning and management.

With the rapid advancement of generative artificial intelligence, AI agents, as the core of next-generation human-computer interaction and automated decision-making, are gradually being integrated into the core processes of real-time communication. However, a fundamental conflict exists between the flexible control flow of agents and the fixed data processing patterns of the media plane, which severely restricts the in-depth application of AI technologies in real-time communication scenarios. To address this challenge, this paper proposes a plug-in and agent collaborative scheduling architecture for the converged media plane oriented towards real-time communication networks. A General Media Framework is constructed to support the efficient embedding of agents into the media pipelines in the form of plug-ins. For high-concurrency media stream processing scenarios, a zero-copy memory mechanism is proposed to significantly reduce latency during stream processing. Furthermore, a two-level scheduling optimization model based on a universal agent is presented to address the redundant computation problem caused by multiple agents running in parallel. Experimental results demonstrate that various AI plug-ins can be stably integrated into the GMF pipeline, and the proposed approach satisfies the end-to-end latency requirements of real-time communication, providing both a theoretical foundation and engineering reference for the AI-native evolution of real-time communication networks.

Decarbonizing transportation requires approaches that embed renewable generation into existing infrastructure. Here we show that roadside photovoltaic deployment along China's roads and railways can be quantified using a geospatial framework that links segmented transport corridors to meteorological grids. The approach maps 480,019 km of transport infrastructure to 4,133 meteorological grids and provides a scalable alternative to coarse regional averaging. Across all deployment scenarios, roadside photovoltaic systems could support 40.91-202.84 GW of installed capacity and generate 56.6-239.2 TWh of electricity annually. Under the baseline scenario, annual generation reaches about 100.6 TWh, equivalent to about 50% of current transport-sector electricity demand. The resulting carbon reduction reaches 33.62-143.97 Mt CO2 annually. The results reveal strong regional heterogeneity, with North and Central China showing the highest near-term potential, while Northwest China could act as a generation-export region. These findings provide a basis for region-specific infrastructure planning and more coordinated transport-energy system integration.

Mobile Soft Open Point (MSOP) technology improves the flexibility and utilization of interconnection resources by enabling mobile deployment across low-voltage distribution networks (LVDNs). However, efficient MSOP scheduling depends on accurately identifying the stage-wise complementarity among interconnectable LVDNs under time-varying source-load conditions. This paper proposes a bi-level equilibrium scheduling framework for MSOP by integrating stage-wise complementarity segmentation with coordinated scheduling optimization. First, a Gaussian mixture model is used to construct the complementary probability matrix among LVDNs, and Bayesian dynamic segmentation is employed to identify statistically stable complementarity stages from its temporal evolution. Then, by incorporating the segmented complementarity stages as stage-wise inputs, a bi-level MSOP scheduling model is developed to coordinate the trade-off between the net benefit of MSOP and the mobile scheduling cost. The scheduling problem is formulated as a Pareto-Nash-equilibrium-based multi-objective coordinated optimization problem. Simulation results verify the effectiveness of the proposed method.

Achieving deep decarbonization of power systems requires large-scale integration of distributed renewables, yet such integration is increasingly constrained by the limited capacity and flexibility of existing distribution networks. Besides, the energy transition has triggered significant load growth, necessitating the capacity expansion and upgrade of numerous distribution assets. Flexible interconnection technology offers a promising solution by enabling coordinated power exchange among neighboring distribution systems, but its deployment has been hindered by high upfront costs and the lack of large-scale assessments. This study develops a nationwide expansion framework that explicitly incorporates scale-dependent cost evolution of flexible interconnection technology and regional heterogeneity in electricity demand across rural, urban, and industrial networks in China. Results show that large-scale deployment can reduce unit costs of interconnection device by more than 50%, reaching approximately 135.81 $/kVA, while enabling annual carbon emission reductions of up to 247 million tons. These findings highlight the critical role of scale-cost interactions in shaping the techno-economic viability and decarbonization potential of distribution-level flexibility technologies.

The proliferation of electric vehicles (EVs) introduces transformative opportunities and challenges for the stability of distribution networks. Unregulated EV charging will further exacerbate the inherent three-phase imbalance of the power grid, while regulated EV charging will alleviate such imbalance. To systematically address this challenge, this study proposes a two-stage bidding strategy with dispatch potential of electric vehicle aggregators (EVAs). By constructing a coordinated framework that integrates the day-ahead and real-time markets, the proposed two-stage bidding strategy reconfigures distributed EVA clusters into a controllable dynamic energy storage system, with a particular focus on dynamic compensation for deviations between scheduled and real-time operations. A bi-level Stackelberg game resolves three-phase imbalance by achieving Nash equilibrium for inter-phase balance, with Ka-rush-Kuhn-Tucker (KKT) conditions and mixed-integer second-order cone programming (MISOCP) ensuring feasible solutions. The proposed coordinated framework is validated with different bidding modes includes independent bidding, full price acceptance, and cooperative bidding modes. The proposed two-stage bidding strategy provides an EVA-based coordinated scheduling solution that balances the economic efficiency and phase stability in electricity market.

The frequent occurrence of typhoon disasters poses a severe threat to the safe operation of distribution networks. Due to issues such as weak network frameworks and insufficient emergency resources, low-voltage distribution networks face significantly increased difficulties in post-disaster recovery. This paper proposes a typhoon post-disaster distribution network restoration method with energy storage as the core, aiming to enhance the power supply resilience of distribution networks under extreme weather conditions. Firstly, by characterizing the source-load characteristics of distribution networks before and after typhoon disasters, this study analyzes the impact mechanism of typhoon disasters on distribution network lines, sorts out typical fault scenarios in the aftermath of typhoons, and interprets the demand changes of critical loads after disasters. On this basis, with the goal of maximizing the recovery volume of critical loads, a fault optimization model utilizing energy storage as the key restoration means is constructed. Finally, a typhoon post-disaster distribution network restoration method with energy storage at its core is proposed.