Intermittency and volatility of photovoltaic (PV) power generation will cause voltage fluctuations in inverter-based microgrids, which require inverter-based voltage control (IBVC) to maintain the grid-wise voltage magnitudes. Considering limited computational and communication abilities of PV inverters, a number of decentralized reinforcement learning (RL) methods have been proposed. However, existing decentralized RL methods cannot provide persistent safety guarantees, which are crucial for IBVC problems, thus hindering their broader application in real world entities. In this paper, we utilize the reachability analysis as a novel safety certificate to provide persistent safety, which adaptively computes the maximum voltage violation throughout the entire control process to characterize the safe set. Under the safe set characterized by reachability analysis, we formulate the IBVC problem as a novel reachability constrained Markov game, and utilize the state-wise Lagrangian method to handle the introduced reachability constraints. Finally, a reachability constrained decentralized RL method is proposed for IBVC in microgrids, which incorporates the multi-agent soft actor-critic algorithm with the reachability analysis and state-wise Lagrangian method, thereby providing persistent safety. The proposed method significantly improves the applicability of current decentralized RL algorithms. Comprehensive comparisons and numerical simulations on various test cases demonstrate the effectiveness of our proposed method.

The rapid evolution of energy structures has positioned microgrids as pivotal components of next-generation power systems, offering enhanced resilience and renewable energy integration. However, the inherent low inertia, complex dynamics, and poor model conditions of microgrids necessitate advanced data-driven frequency control strategies. Although reinforcement learning (RL) has demonstrated certain potential and advantages, existing RL methods often struggle with generalization across diverse microgrid configurations and lack adaptability to unseen environments, particularly when explicit system parameters are unavailable. To address these challenges, in this paper, we introduce a novel prompt decision transformer (Prompt-DT) architecture for microgrid frequency control. Unlike traditional approaches that rely on hard-to-obtain environmental characteristic parameters, the proposed method leverages few-shot expert historical trajectories as prompts to guide autonomous perception and adaptive decision-making. In addition, we propose a context-aware training and execution mechanism utilizing self-supervised contrastive learning to enhance environment recognition and prompt utilization efficiency. In addition, a physics-informed prompt design technique that filters prompts based on cumulative reward and frequency volatility is proposed, ensuring high-quality physical guidance during online execution. Finally, to ensure generalization in unseen environments with limited data, we develop a lightweight finetuning approach that achieves performance comparable to full-parameter finetuning with minimal adjustments.

The growing integration of distributed photovoltaics (PVs) into active distribution networks (ADNs) has exacerbated operational challenges, making it imperative to coordinate diverse equipment to mitigate voltage violations and enhance power quality. Although existing data-driven approaches have demonstrated effectiveness in the voltage control problem, they often require extensive trial-and-error exploration and struggle to incorporate heterogeneous information, such as day-ahead forecasts and semantic-based grid codes. Considering the operational scenarios and requirements in real-world ADNs, in this paper, we propose a hybrid knowledge-data-driven approach that leverages dynamic collaboration between a large language model (LLM) agent and a reinforcement learning (RL) agent to achieve two-stage voltage control. In the day-ahead stage, the LLM agent receives coarse region-level forecasts and generates scheduling strategies for on-load tap changer (OLTC) and shunt capacitors (SCs) to regulate the overall voltage profile. Then in the intra-day stage, based on accurate node-level measurements, the RL agent refines terminal voltages by deriving reactive power generation strategies for PV inverters. On top of the LLM-RL collaboration framework, we further propose a self-evolution mechanism for the LLM agent and a pretrain-finetune pipeline for the RL agent, effectively enhancing and coordinating the policies for both agents. The proposed approach not only aligns more closely with practical operational characteristics but also effectively utilizes the inherent knowledge and reasoning capabilities of the LLM agent, significantly improving training efficiency and voltage control performance. Comprehensive comparisons and ablation studies demonstrate the effectiveness of the proposed method.

With the integration of massive distributed energy resources and the widespread participation of novel market entities, the operation of active distribution networks (ADNs) is progressively evolving into a complex multi-scenario, multi-objective problem. Although expert engineers have developed numerous domain specific models (DSMs) to address distinct technical problems, mastering, integrating, and orchestrating these heterogeneous DSMs still entail considerable overhead for ADN operators. Therefore, an intelligent approach is urgently required to unify these DSMs and enable efficient coordination. To address this challenge, this paper proposes the ADN-Agent architecture, which leverages a general large language model (LLM) to coordinate multiple DSMs, enabling adaptive intent recognition, task decomposition, and DSM invocation. Within the ADN-Agent, we design a novel communication mechanism that provides a unified and flexible interface for diverse heterogeneous DSMs. Finally, for some language-intensive subtasks, we propose an automated training pipeline for fine-tuning small language models, thereby effectively enhancing the overall problem-solving capability of the system. Comprehensive comparisons and ablation experiments validate the efficacy of the proposed method and demonstrate that the ADN-Agent architecture outperforms existing LLM application paradigms.

The increasing penetration of distributed energy resources into active distribution networks (ADNs) has made effective ADN dispatch imperative. However, the numerous newly-integrated ADN operators, such as distribution system aggregators, virtual power plant managers, and end prosumers, often lack specialized expertise in power system operation, modeling, optimization, and programming. This knowledge gap renders reliance on human experts both costly and time-intensive. To address this challenge and enable intelligent, flexible ADN dispatch, this paper proposes a large language model (LLM) powered automated modeling and optimization approach. First, the ADN dispatch problems are decomposed into sequential stages, and a multi-LLM coordination architecture is designed. This framework comprises an Information Extractor, a Problem Formulator, and a Code Programmer, tasked with information retrieval, optimization problem formulation, and code implementation, respectively. Afterwards, tailored refinement techniques are developed for each LLM agent, greatly improving the accuracy and reliability of generated content. The proposed approach features a user-centric interface that enables ADN operators to derive dispatch strategies via simple natural language queries, eliminating technical barriers and increasing efficiency. Comprehensive comparisons and end-to-end demonstrations on various test cases validate the effectiveness of the proposed architecture and methods.

In recent years, the rapid development of smart grid technologies and energy markets has brought massive new entities into the power system, such as virtual power plant (VPP) operators, distribution retailers, electric vehicle (EV) aggregators, and end prosumers. These entities with little knowledge of power system modeling, dispatch, and optimization require a more intelligent and flexible approach to ensure their efficient participation. Trained on a vast corpus of text data, generative pre-trained transformer (GPT) based large language models (LLMs) are embedded with massive world knowledge. With powerful comprehension and reasoning capabilities, GPT based LLMs have the potential to enable intelligent and customized dispatch. However, despite extensive experimentation, existing studies have neither addressed how LLMs should be applied in power systems nor provided a generalized framework. In this paper, we propose a generalized framework of LLMs for power system operation and discuss its implementations and limitations in detail. Comprehensive case studies and test results demonstrate the effectiveness of the proposed framework and several critical factors to ensure the performance of LLMs.

A residual deep reinforcement learning (RDRL) based on an approximate-model-driven optimization approach is proposed for inverter-based volt-var control (IB-VVC) in active distribution networks. A modified Markov decision process is introduced to formulate the model-based and RDRL-based IB-VVC simultaneously, and then RDRL learns a residual action based on the action of the model-based approach with an approximate model. It inherits the control capability of the approximate-model-based optimization and enhances the policy optimization capability by residual policy learning. Since the approximate model acquired by operators is generally relatively reliable, the action solved by model-based optimization approaches is not far away from the optimal one. This allows RDRL to search for the residual action in a smaller residual action space, which further improves the approximation accuracy of the critic and reduces the search difficulties of the actor. Simulations demonstrate that RDRL improves the optimization performance considerably throughout the learning stage and verifies their three rationales for superior performance point-by-point on 69 and 141 bus balanced distribution networks.

Automation, sharing and electrification will be the important development trend of automobile and transportation system in the future, and will have a profound impact on energy and environment. In recent years, the research of shared autonomous electric vehicle has become a hot spot, which needs systematic summary and analysis. This paper focuses on the electric vehicle, and summarizes the interaction of automatic driving, sharing travel, wireless charging, V2G and other technologies. Then, aiming at the shared autonomous, focusing on the charging part, this paper summarizes the modeling methods in the aspects of charging facility planning, charging load modeling and charging management. Then, the paper summarizes the interaction of influence of shared autonomous electric vehicle fleet with energy infrastructures including power systems and charging stations. Finally, the paper analyzes the shortcomings of the existing research, and looks forward to the possible research direction in the future.

This paper proposes a novel adaptive data-driven secondary control method for islanded microgrids (MGs) dominated by grid-forming (GFM) inverters, without depending on accurate network parameters. Using the measured input/output data of the MG, a data-driven dynamic linearization model related to disturbances is built at each operating point of the closed-loop control system. Next, we develop a robust recursive regression method to learn the dynamic linearization parameter that is updated online. To estimate the lumped disturbance, an adaptive disturbance observer with an observer gain updated online is introduced. Subsequently, the adaptive data-driven cooperative controller with an iterative structure is developed. This method can effectively regulate the voltage and frequency of the islanded MG with adaptability to system changes. The effectiveness of the proposed method is verified through both simulations and field tests in a real 100% PV powered islanded microgrid.

As large-scale distributed energy resources are integrated into the active distribution networks (ADNs), effective energy management in ADNs becomes increasingly prominent compared to traditional distribution networks. Although advanced reinforcement learning (RL) methods, which alleviate the burden of complicated modelling and optimization, have greatly improved the efficiency of energy management in ADNs, safety becomes a critical concern for RL applications in real-world problems. Since the design and adjustment of penalty functions, which correspond to operational safety constraints, require extensive domain knowledge in RL and power system operation, the emerging ADN operators call for a more flexible and customized approach to address the penalty functions so that the operational safety and efficiency can be further enhanced. Empowered with strong comprehension, reasoning, and in-context learning capabilities, large language models (LLMs) provide a promising way to assist safe RL for energy management in ADNs. In this paper, we introduce the LLM to comprehend operational safety requirements in ADNs and generate corresponding penalty functions. In addition, we propose an RL2 mechanism to refine the generated functions iteratively and adaptively through multi-round dialogues, in which the LLM agent adjusts the functions' pattern and parameters based on training and test performance of the downstream RL agent. The proposed method significantly reduces the intervention of the ADN operators. Comprehensive test results demonstrate the effectiveness of the proposed method.

With the advanced reasoning, contextual understanding, and information synthesis capabilities of large language models (LLMs), a novel paradigm emerges for the autonomous generation of dispatch strategies in modern power systems. In this paper, we propose an LLM-based experience-driven day-ahead Volt/Var schedule solution for distribution networks, which enables the self-evolution of LLM agent's strategies through the collaboration and interaction of multiple modules, specifically, experience storage, experience retrieval, experience generation, and experience modification. The experience storage module archives historical operational records and decisions, while the retrieval module selects relevant past cases according to current forecasting conditions. The LLM agent then leverages these retrieved experiences to generate new, context-aware decisions for current situation, which are subsequently refined by the modification module to realize self-evolution of the dispatch policy. Comprehensive experimental results validate the effectiveness of the proposed method and highlight the applicability of LLMs in power system dispatch problems facing incomplete information.

With the increasing number of distributed energy resources in active distribution networks (ADNs), the complexity of dispatch modeling and optimization has led to extremely heavy burdens for distribution system operators. The natural language dispatch, which integrates Large Language Models (LLMs), can reduce the operators' heavy burden and thereby reduce the probability of making mistakes. However, traditional LLM application methods suffer from issues such as low accuracy. This paper proposes an LLM-based dispatch framework for ADNs. The framework constructs a sequential-agent architecture with four stages-problem identification, problem analysis, problem solving, and instruction translation-using LLMs to solve dispatch problems described in natural language and generate dispatch instructions. By encapsulating common constraints and objective functions and leveraging the enhanced Retrieval-Augmented Generation (RAG) technique, LLMs can successfully master the PyOptInterface mathematical modeling language and effectively handle complex ADN dispatch problems. Experimental results show that the proposed framework achieves accuracy rates of 100% and 90% in two different dispatch problems, significantly outperforming other comparative baselines. The generated dispatch instructions are trustworthy, highlighting the potential of the framework for application in power system dispatch optimization.

Increasing penetration of renewable energy has led to reduced system inertia and increased challenges in maintaining frequency stability. In this paper, a detailed DFIG-based wind turbine model has been developed and integrated into a modified IEEE 9-bus system within the RSCAD/RTDS, which comprises aerodynamic model, drive train, and converters. The control strategies combine maximum power point tracking (MPPT) with primary frequency support through droop and virtual inertia control. In the case study, a 300 MW wind farm operates with three 100 MW synchronous generators in the IEEE 9-bus system. The simulation scenario is set as a sudden 50 MW load increase. Two cases are considered here: one without wind farm participation in frequency regulation and one with both droop and virtual inertia support enabled. RTDS simulation results demonstrate that allowing the wind farm to provide primary frequency control significantly reduces the frequency nadir and accelerates frequency recovery.

To address the challenges posed by the volatility and randomness of wind power to power systems, numerous studies on probabilistic forecasting of wind power have emerged in recent years. In this paper, a novel probabilistic forecasting approach is proposed by combining statistical method with deep learning model, named ARIMA-Assisted-Neural-Network (AANN). In AANN, the autoregressive integrated moving average model is seamlessly integrated with a lightweight deep learning architecture. The hybrid framework employs a two-stage temporal encoding mechanism, utilizing continuous sampling for short-term pattern capture and interval sampling for long-term dependency extraction, thereby achieving complementary integration of statistical principles and deep learning capabilities. The proposed method addresses two key limitations in probabilistic wind power forecasting: a) By incorporating statistical modeling, it enhances conventional deep learning frameworks through improved prediction interval reliability and robustness; b) A post-processing technique resolves quantile crossing issues in quantile regression by enforcing monotonic constraints, ensuring mathematically consistent quantile relationships without distorting distributional properties. Experimental evaluations on real-world wind farm datasets demonstrate AANN's superior performance over state-of-the-art models (Autoformer, Mamba), particularly when processing extended input sequences. The results substantiate that hybrid statistical-deep learning approaches offer advantages in renewable energy forecasting accuracy, providing methodological insights for handling stochastic power generation in modern electrical grids.

As numerous distributed energy resources (DERs) are integrated into the distribution networks, the optimal dispatch of DERs is more and more imperative to achieve transition to active distribution networks (ADNs). Since accurate models are usually unavailable in ADNs, an increasing number of reinforcement learning (RL) based methods have been proposed for the optimal dispatch problem. However, these RL based methods are typically formulated without safety guarantees, which hinders their application in real world. In this paper, we propose an RL based method called supervisor-projector-enhanced safe soft actor-critic (S3AC) for the optimal dispatch of DERs in ADNs, which not only minimizes the operational cost but also satisfies safety constraints during online execution. In the proposed S3AC, the data-driven supervisor and projector are pre-trained based on the historical data from supervisory control and data acquisition (SCADA) system, effectively providing enhanced safety for executed actions. Numerical studies on several IEEE test systems demonstrate the effectiveness and safety of the proposed S3AC.

Benefiting from precise perception, real-time prediction and reliable planning, autonomous driving systems have exhibited exceptional performance in research. However, the high complexity and opacity prevent its application in practice. To introduce a user-friendly autonomous driving system, we propose a driving captioner to generate real time description and explanation of self-driving systems in natural language. Specifically, we unify the end-to-end autonomous driving and video captioning tasks into a single yet effective framework by introducing an additional captioning head to describe the action of the vehicle and explain the reasons. Besides, we exploit an effective accelerating method to accelerate the inference process, which decreases the average inference time from 0.670 s to 0.298 s. Through extensive experiments on both simulation datasets and real-world datasets, we show the superior generalization ability and robustness of the proposed framework.

Wind power development presents large-scale characteristics. In order to enhance the electrical quality of wind farm integrated hybrid HVDC system and reduce the footprint and cost of platforms, this paper proposes an adaptive harmonic current compensation method. Firstly, the topology structure and harmonic characteristics of the hybrid HVDC system are introduced, where the high-voltage valve group adopts uncontrolled diode rectifier (DR) inverter and the low-voltage valve group adopts modular multilevel converters. Then, the sliding window iterative Discrete Fourier Transform (DFT) harmonic detection method and the adaptive harmonic compensation control strategy based on it are presented. Finally, a model of the hybrid HVDC grid system is built in the PSCAD/EMTDC which to verify the effectiveness of the adaptive harmonic current compensation control strategy based on sliding window iterative DFT in steady-state and dynamic operation of the system. The study shows that this control strategy provides a new option for the filtering method of the hybrid HVDC system.

A residual deep reinforcement learning (RDRL) approach is proposed by integrating DRL with model-based optimization for inverter-based volt-var control in active distribution networks when the accurate power flow model is unknown. RDRL learns a residual action with a reduced residual action space, based on the action of the model-based approach with an approximate model. RDRL inherits the control capability of the approximate-model-based optimization and enhances the policy optimization capability by residual policy learning. Additionally, it improves the approximation accuracy of the critic and reduces the search difficulties of the actor by reducing residual action space. To address the issues of "too small" or "too large" residual action space of RDRL and further improve the optimization performance, we extend RDRL to a boosting RDRL approach. It selects a much smaller residual action space and learns a residual policy by using the policy of RDRL as a base policy. Simulations demonstrate that RDRL and boosting RDRL improve the optimization performance considerably throughout the learning stage and verify their rationales point-by-point, including 1) inheriting the capability of the approximate model-based optimization, 2) residual policy learning, and 3) learning in a reduced action space.