Wireless on-road charging is an emerging charging method in addition to plug-in charging. And it is a promising application for the future smart grid. Hence, in this paper, a reputation-based wireless on-road vehicle-to-vehicle (V2V) energy trading strategy is formulated in vehicular energy networks. First, a three-stage wireless V2V energy trading algorithm is established to capture the interaction between charging electric vehicles (EVs) and discharging EVs and obtain the optimal energy trading matching results. Second, the trustworthiness of the discharging EV is evaluated using the proposed reputation index. Both explicit reputation and implicit reputation indices are incorporated to rigorously derive the real-time reputation index based on the consortium blockchain system. Third, two irrational behaviors of EV users, namely, the weighting effect and range anxiety, are mathematically modeled based on Prospect theory. Numerical results indicate that efficient wireless energy matching can be achieved. Moreover, the proposed wireless V2V energy trading strategy is effective in increasing the utility of both charging EVs and discharging EVs.

The increasing penetration of renewable energy introduces significant uncertainties in power systems, necessitating advanced strategies to maintain economic efficiency and frequency stability. This article addresses the multireserve constrained unit commitment (MRCUC) problem by simultaneously allocating spinning reserve and frequency regulation reserve in systems with high renewable energy penetration. First, spinning reserve requirements are accurately determined by an optimization model addressing imbalances caused by renewable energy volatility and load variations. Second, primary and secondary frequency regulation requirements are quantified through detailed frequency-domain simulations utilizing each unit's frequency-to-active-power-output transfer function, thereby overcoming limitations of traditional first-order frequency approximations applicable mostly to synchronous generators. The proposed comprehensive MRCUC model, characterized by partial observability of load uncertainties and high-ordernonconvexity stemming from precise frequency simulations, cannot be efficiently solved using conventional optimization methods. Therefore, a hybrid optimization framework combining RL with convex optimization techniques is developed to address these complexities, ensuring feasible and effective decision-making. Extensive case studies conducted on IEEE 39-bus and 118-bus test systems confirm the efficacy of the proposed method, highlighting enhanced economic performance, accurate reserve allocation, and robust frequency stability.

While Electric Vehicles (EVs) have the potential to contribute to low-carbon transportation, the burgeoning adoption of EVs and uncoordinated charging can place a significant burden on the power grid. This paper addresses the challenges of the integration of massive EVs into smart grids by introducing a comprehensive framework with heterogeneous flexibility aggregation. First, a Spatio-Temporal Heterogeneous Graph Neural Network (STH-GNN) is proposed to accurately predict EV charging demands by analyzing complex spatio-temporal relationships. Furthermore, a bi-level coordinated charging scheduling framework optimizes grid operations and individual EV charging schedules, relieving grid burden and enhancing efficient energy use based on the STH-GNN prediction. It solves the gaps between prediction errors and real-time actual charging demand. Additionally, the EV flexibility set is modelled to effectively aggregate heterogeneous EV resources, allowing for optimized deployment of large-scale EV assets by constructing the inner approximation of the Minkowski sum of the individual flexibility sets. The proposed method is verified in simulation, utilizing data from a realistic urban setting with diverse EV charging behaviours and grid conditions. Simulations demonstrate the effectiveness of the STH-GNN in accurately forecasting EV charging demands across different times and locations. The bi-level charging scheduling framework successfully manages grid burden while showcasing significant improvements in operational efficiency and cost reduction. Results validate the proposed model’s robustness and scalability, proving its potential applicability in real-world smart grid environments.

Federated Learning (FL) has received much attention in recent years. However, although clients are not required to share their data in FL, the global model itself can implicitly remember clients' local data. Therefore, it’s necessary to effectively remove the target client's data from the FL global model to ease the risk of privacy leakage and implement "the right to be forgotten". Federated Unlearning (FU) has been considered a promising solution to remove data without full retraining. But the model utility easily suffers significant reduction during unlearning due to the gradient conflicts. Furthermore, when conducting the post-training to recovery the model utility, it’s prone to move back and revert what have already been unlearned. To address these issues, we propose Federated Unlearning with Orthogonal Steepest Descent (FedOSD). We first design an unlearning cross entropy loss to overcome the convergence issue of the gradient ascent. A steepest descent direction for unlearning is then calculated in the condition of being non-conflicting with other clients’ gradients and closest to the target client's gradient. This benefits to efficiently unlearn and mitigate the model utility reduction. After unlearning, we recover the model utility by maintaining the achievement of unlearning. Finally, extensive experiments in several FL scenarios verify that FedOSD outperforms the SOTA FU algorithms in terms of unlearning and the model utility.

Machine unlearning in the domain of large language models (LLMs) has attracted great attention recently, which aims to effectively eliminate undesirable behaviors from LLMs without full retraining from scratch. In this paper, we explore the Gradient Ascent (GA) approach in LLM unlearning, which is a proactive way to decrease the prediction probability of the model on the target data in order to remove their influence. We analyze two challenges that render the process impractical: gradient explosion and catastrophic forgetting. To address these issues, we propose Multi-Objective Large Language Model Unlearning (MOLLM) algorithm. We first formulate LLM unlearning as a multi-objective optimization problem, in which the cross-entropy loss is modified to the unlearning version to overcome the gradient explosion issue. A common descent update direction is then calculated, which enables the model to forget the target data while preserving the utility of the LLM. Our empirical results verify that MoLLM outperforms the SOTA GA-based LLM unlearning methods in terms of unlearning effect and model utility preservation. The source code is available at https://github.com/zibinpan/MOLLM.

In recent years, the rapid growth of distributed energy resources (DERs) and the emergence of local energy markets (LEMs) have dramatically transformed the energy trading landscape, emphasizing the crucial role of DER aggregators in optimizing bidding strategies. Traditional model-based methods for optimizing DER aggregator bidding in LEMs face significant challenges, including information asymmetry, an inability to adapt to changing market dynamics, and issues with computational scalability in real-time decision-making. Recognized as a promising alternative, deep reinforcement learning (DRL) forms the basis of our proposed solution. This article introduces a human–machine (HM) framework that utilizes a multiagent adversarial inverse reinforcement learning (MA-AIRL) approach to address these challenges. The HM framework enables the DER agent to imitate human demonstrations and leverages a HM hybrid experiment to augment insufficient data, effectively tackling the problem of data inadequacy in new market environments. Concurrently, the MA-AIRL algorithm employs inverse reinforcement learning to capture underlying reward functions, risk preferences, and expert knowledge, significantly enhancing the model’s adaptability to dynamic market conditions. Additionally, the adversarial learning component allows the DER agent to robustly respond to uncertainties and the strategic maneuvers of rival agents, thereby mitigating information asymmetry. Moreover, this DRL-based approach is designed to ensure rapid responsiveness without compromising scalability in real-time contexts. Through extensive case studies, we have verified that the proposed HM framework and MA-AIRL algorithm offer a more robust, data-efficient, and adaptive approach for optimizing DER aggregator bidding in LEMs.

On-road wireless charging is an emerging charging method, in addition to plug-in charging, that is a promising application in the future smart grid. Hence, in this paper, a competitive pricing strategy for the wireless charging lane (WCL) is proposed to maximize the economic benefits of the WCL operator. First, the competitive pricing strategy is formulated based on a non-cooperative game between the WCL operator and the charging station (CS). An iterative optimal pricing searching algorithm is developed to find the Nash equilibrium of the game. Second, a tri-level framework is established to derive the optimal competitive price considering the interaction among the WCL operator, the power distribution network (PDN) operator, and EV users. Third, the range anxiety of EV users is mathematically modeled based on Prospect theory. Numerical results indicate that the pricing strategy is effective in enhancing the attractiveness and profitability of the WCL operator. In addition, the utility of EV users is increased as well. Moreover, the PDN loss cost can be reduced, and downward voltage violation can be avoided.

With extensive pretrained knowledge and high-level general capabilities, large language models (LLMs) emerge as a promising avenue to augment reinforcement learning (RL) in aspects, such as multitask learning, sample efficiency, and high-level task planning. In this survey, we provide a comprehensive review of the existing literature in LLM-enhanced RL and summarize its characteristics compared with conventional RL methods, aiming to clarify the research scope and directions for future studies. Utilizing the classical agent-environment interaction paradigm, we propose a structured taxonomy to systematically categorize LLMs' functionalities in RL, including four roles: information processor, reward designer, decision-maker, and generator. For each role, we summarize the methodologies, analyze the specific RL challenges that are mitigated and provide insights into future directions. Finally, the comparative analysis of each role, potential applications, prospective opportunities, and challenges of the LLM-enhanced RL are discussed. By proposing this taxonomy, we aim to provide a framework for researchers to effectively leverage LLMs in the RL field, potentially accelerating RL applications in complex applications, such as robotics, autonomous driving, and energy systems.

Fairness has become an important concern in Federated Learning (FL). An unfair model that performs well for some clients while performing poorly for others can reduce the willingness of clients to participate. In this work, we identify a direct cause of unfairness in FL - the use of an unfair direction to update the global model, which favors some clients while conflicting with other clients’ gradients at the model and layer levels. To address these issues, we propose a layer-wise fair Federated Learning algorithm (FedLF). Firstly, we formulate a multi-objective optimization problem with an effective fair-driven objective for FL. A layer-wise fair direction is then calculated to mitigate the model and layer-level gradient conflicts and reduce the improvement bias. We further provide the theoretical analysis on how FedLF can improve fairness and guarantee convergence. Extensive experiments on different learning tasks and models demonstrate that FedLF outperforms the SOTA FL algorithms in terms of accuracy and fairness. The source code is available at https://github.com/zibinpan/FedLF.

This paper introduces a novel approach that leverages Large Language Models (LLMs) and Generative Agents to enhance time series forecasting by reasoning across both text and time series data. With language as a medium, our method adaptively integrates social events into forecasting models, aligning news content with time series fluctuations to provide richer insights. Specifically, we utilize LLM-based agents to iteratively filter out irrelevant news and employ human-like reasoning to evaluate predictions. This enables the model to analyze complex events, such as unexpected incidents and shifts in social behavior, and continuously refine the selection logic of news and the robustness of the agent's output. By integrating selected news events with time series data, we fine-tune a pre-trained LLM to predict sequences of digits in time series. The results demonstrate significant improvements in forecasting accuracy, suggesting a potential paradigm shift in time series forecasting through the effective utilization of unstructured news data.

Electric vehicles (EVs) are believed to be effective in reducing the use of fossil fuels. However, the increasing penetration of EVs can lead to challenges in managing EV charging/refueling. Typically, when EVs simultaneously gather at the stations, long queueing lengths and waiting times might occur, which might result in EV disutility. Thus, in this paper, novel decision-making strategies considering the irrational behaviors of EVs are formulated to ensure the utility of EVs. Three types of EVs are investigated, i.e., plug-in electric vehicles (PEVs), fuel cell electric vehicles (FCEVs), and plug-in hybrid hydrogen and electric vehicles (PH2EVs). First, three typical types of irrationalities are modeled based on prospect theory and irrational herding theory, including risk attitude irrationality, probability distortion, and irrational herding behaviors. Second, decision-making strategies, including station selection, navigation, and energy purchasing, are modeled based on EV irrationalities. Additionally, when making the station selection decision, the selection ranking index is proposed to balance energy price and waiting time. Third, the energy substitution of PH2EV is investigated to determine the optimal energy-purchasing mixture of electricity and hydrogen. Simulation results show that the average waiting time and charging/refueling time of EV users of the proposed model are reduced compared with the other two cases where EV users are assumed to be rational. In addition, the overall utility of EV users is enhanced.

Developing the deep learning (DL) technique is a promising way to improve renewable energy forecasting accuracy and offset the negative impacts of renewable energy on the power system. However, the application of the DL technique brings novel cyberthreats to the renewable energy forecast, and its cybersecurity has not received enough attention in previous literatures. To fill the gap, the vulnerability of renewable energy forecasting is, among the first, studied in-depth in this article. First, a novel cyberattack named adversarial learning attack (ALA) is proposed. The ALA is achieved by tampering with the meteorological data obtained by online weather forecasts from external application programming interfaces to undermine the renewable energy forecasting performance, which jeopardizes the power system operation. Then, an iterative algorithm is proposed to solve the ALA-based optimization problem. As the DL model is involved as optimization constraints, the optimization problem is nonconvex and NP-hard, which is unable to be solved by traditional approaches. The proposed algorithm utilizes the proximal gradient descent principle and is effective in iteratively exploring the near-optimal solution. At last, the impact of the ALA strategy on the power system operation is assessed, which considers the economic loss incurred and the potential hazards. The feasibility and efficacy of the ALA strategy are validated by conducting comprehensive and extensive experiments on the IEEE 30-bus benchmarks. The simulation results reveal that the ALA is able to impose severe economic losses on the operation and even induces disastrous hazards, such as power system collapse.

Fairness has been considered as a critical problem in federated learning (FL). In this work, we analyze two direct causes of unfairness in FL - an unfair direction and an improper step size when updating the model. To solve these issues, we introduce an effective way to measure fairness of the model through the cosine similarity, and then propose a federated multiple gradient descent algorithm with fair guidance (FedMDFG) to drive the model fairer. We first convert FL into a multi-objective optimization problem (MOP) and design an advanced multiple gradient descent algorithm to calculate a fair descent direction by adding a fair-driven objective to MOP. A low-communication-cost line search strategy is then designed to find a better step size for the model update. We further show the theoretical analysis on how it can enhance fairness and guarantee the convergence. Finally, extensive experiments in several FL scenarios verify that FedMDFG is robust and outperforms the SOTA FL algorithms in convergence and fairness. The source code is available at https://github.com/zibinpan/FedMDFG.

Achieving carbon neutrality is widely recognized as the key measure to mitigate climate change. As the basis for achieving carbon neutrality, corporate carbon footprint (CCF) estimation is mainly based on the disclosed information of corporates to roughly estimate the direct carbon emission, but the estimation may not be comprehensive, timely, and accurate. In this article, the CCF estimation problem is formulated and a novel estimation methodology is proposed for the first time to estimate the direct and indirect carbon emissions of factories in real time. An appliance identification method based on the multihead self-attention mechanism and gated recurrent unit is proposed to identify the device states, and then, calculate the corresponding direct carbon emission. The indirect carbon emission is derived from the electricity consumption of the factory and the marginal carbon emission factor of the connected bus. A dataset containing load and device state data from six different industries is released and used to verify the effectiveness of the proposed method. Experiments show that the proposed appliance identification method is significantly superior to the benchmarks in the literature, and the proposed method can achieve a comprehensive and accurate estimation of the minute-level CCF.

Maritime transportation takes a major responsibility in public travel between islands while producing a large quantity of greenhouse gas (GHG) emissions. All-electric ships (AESs) can be applied to mitigate GHG emissions through energy storage systems (ESS), renewable integration, and cold ironing. In this paper, we propose a flexible voyage scheduling strategy for AESs based on the temporal-spatial dynamics (TSD) to satisfy the transportation demand while mitigating the burden of the AESs on the power grids during charging. The interaction between the AESs and island-based microgrids is modeled. Furthermore, the AESs are utilized to enhance the resilience of the power grids. The AESs can be dispatched to realize the load restoration under contingencies. The proposed methodology is verified on a three-island system. It can be concluded that under normal operation, the proposed voyage scheduling can reduce the total energy consumption cost of the AESs and the islands. Besides, the voltage violation can be improved. Under the emergency, the proposed method can help the grids restore more critical and normal loads.

With the burgeoning of plug-in electric vehicles (PEVs), as well as the emergence of fuel cell vehicles (FCVs) and plug-in hybrid electricity/hydrogen vehicles (PH2EVs), the synergistic effect of electricity, hydrogen, and transportation networks should be investigated. To enhance the security of the coupled network and ensure the reliability of the charging/refueling services of EVs, a vulnerability assessment is required to assist in system contingency planning as well as post-contingency measures. In this paper, a vulnerability assessment strategy is formulated for coupled transportation and multi-energy networks. First, a novel graph representation for the coupled transportation and multi-energy networks is proposed. Then, a critical asset identification tool is applied to find the vulnerability point of the coupled networks, and the transfer margin ratio (TMR) is put forward to assess the dynamic vulnerability level under cascading contingencies. Finally, a vulnerability envelope (the lower bound and upper bound of vulnerability) is found based on a bi-level optimization problem. To demonstrate the effectiveness of the proposed methodology, case studies are performed on the IEEE 39-bus electricity network coupled with a 25-node transportation network and a 50-node hydrogen/gas network. It is verified that the vulnerable point of the coupled network can be found. Besides, it is concluded that the penetration of FCVs and PH2EVs can enhance energy flexibility through the energy substitution effect and thus mitigate the system vulnerability.

The cloud-based application is a major developing feature of smart grids. Apart from centralized Internet data centers (IDCs), distributed cloud resources (CRs) can also provide cloud computing services with low latency and high reliability. For the distributed cloud computing, CRs aggregators (CRAs) will integrate the distributed idle computing resources, which are dispersed energy consumers in the system, to form virtual IDCs. This article presents a market-based computing resource allocation method for distributed cloud computing services. The computing resource allocation refers to the approach to allocating the computing workloads to different CRs. First, the batch workload scheduling (BWS)-based virtual energy storage system (VESS) model and thermal inertia (TI)-based VESS model are proposed to help CRAs better aggregate the distributed CRs and characterize the energy consumption flexibility of the virtual IDCs. Then, the energy trading behavior of the CRAs in the transactive energy market is modeled in the resource allocation process. Case studies are conducted on a 55-bus electricity system. It can be found that energy consumption costs can be reduced by applying the proposed methodology, and revenues from providing cloud computing services can be increased.

Due to vulnerabilities exposed to cyberattacks in the cyber physical power system, increasing concerns have been paid to its cybersecurity, especially on the so-called false data injection attack. Timely recovering true values of measurements and states after encountering cyber-attacks is of paramount importance for ensuring the subsequent controls and operations of the cyber physical power system. This article, for the first time, discovers a measurement data inertia effect, and uses this effect to deduce coarse values of preattack measurements as a preliminary work for data recovery. Then, based on the deduced coarse values and suggested state bounds, an optimization model is proposed to recover the measurements and states contaminated by attacks in-time. Moreover, an error criterion named interval error is proposed to assess the entire performance of the proposed recovery scheme. Extensive and comprehensive experiments are implemented on the IEEE 30-bus test benchmark to verify the feasibility and effectiveness of the proposed recovery scheme. The numerical studies reveal that the proposed method can achieve high accuracy and efficient timeliness for data recovery.

Cooperative wind farm control is a complex problem due to wake effect, and it is hard to find the proper model. Reinforcement learning can find the optimal policy in a dynamic environment using “trial and error,” but may damage the machine and cause high cost during the learning process. In order to address this challenge, this article proposes the knowledge-assisted reinforcement learning framework by combining the low-fidelity analytical model with a reinforcement learning framework. Moreover, the knowledge-assisted deep deterministic policy gradient (KA-DDPG) algorithm and three kinds of knowledge-assisted learning methods are proposed based on the framework. The proposed methods are tested in nine different scenarios of WFSim. The simulation results show that the KA-DDPG algorithm can reach the maximum power output and ensure safety during learning. In addition, the learning cost is reduced by accelerating the learning process.

A key feature of an electricity distribution market is that it may be dominated by renewable generation with zero marginal cost. Existing market mechanisms are likely to fail in this context since it cannot generate a reasonable price signal to compensate for the investment cost of renewable generators. Given this background, a double-sided auction market mechanism is presented for pricing the zero marginal cost renewable generation in the distribution system. Honesty is proved to be a dominant strategy for participants, which would enable the proposed mechanism to develop into a set-and-forget bidding market. The proposed market mechanism is also shown to be compatible with the nodal pricing system. Finally, case studies are carried out, and the results show that under the proposed market mechanism, the problem of always bidding a zero price by renewable generators in some existing markets can be avoided. Even when only renewable generation units with zero marginal costs participate in the bidding, the proposed mechanism can still produce a reasonable market clearing price. When adopting the average pricing market mechanism, merits of nodal pricing can still be retained and contribute to the enhancement of the operating efficiency of the distribution network.

This paper proposes an operational planning framework for large-scale thermostatically controlled load (TCL) dispatch. The proposed framework consists of a day-ahead scheduling stage and a real-time operation stage. A thermal comfort model is employed to estimate the occupants' thermal comfort degree. A self-adaptive TCL grouping method is proposed to group the TCLs based on the similarity of the TCL model parameters. Then, a hierarchical day-ahead scheduling model is proposed to make the optimal dispatch plan for the TCL aggregators based on the day-ahead forecasted information. In the real-time operation stage, a predictive control model is proposed for the TCL aggregators to make the real-time TCL dispatch decision based on the updated real-time information. The simulation results prove the efficiency of the proposed framework.

This paper proposes a new evolutionary algorithm (EA), which is called the natural aggregation algorithm (NAA). NAA is inspired by the collective decision making intelligence of the group-living animals. Distinguished from other EAs, NAA distributes individuals to several sub-populations (called `shelters'), and uses a stochastic migration model to dynamically mitigate the individuals among the shelters. The inter-individual attraction effect and crowding effect are considered in the migration model to balance the exploration and exploitation. In each generation, both of the located search and generalized search are performed simultaneously, and the distributions of the individuals are self-adaptively updated. 7 benchmark functions with different dimensionality settings are used to validate the efficiency of NAA, and the results clearly show that NAA has strong performance for solving the real-parameter optimization problems.

In the smart grid paradigm, growing integration of large-scale intermittent renewable energies has introduced significant uncertainties to the operations of an electric power system. This makes real-time dynamic security assessment (DSA) a necessity to enable enhanced situational-awareness against the risk of blackouts. Conventional DSA methods are mainly based on the time-domain simulation, which are insufficiently fast and knowledge-poor. In recent years, the intelligent system (IS) strategy has been identified as a promising approach to facilitate real-time DSA. While previous works mainly concentrate on the rotor angle stability, this paper focuses on another yet increasingly important dynamic insecurity phenomenon-the short-term voltage instability, which involves fast and complex load dynamics. The problem is modeled as a classification subproblem for transient voltage collapse and a prediction subproblem for unacceptable dynamic voltage deviation. A hierarchical IS is developed to address the two subproblems sequentially. The IS is based on ensemble learning of random-weights neural networks and is implemented in an offline training, a real-time application, and an online updating pattern. The simulation results on the New England 39-bus system verify its superiority in both learning speed and accuracy over some state-of-the-art learning algorithms.

In this paper, an economic dispatch model, which can take into account the uncertainties of plug-in electric vehicles (PEVs) and wind generators, is developed. A simulation based approach is first employed to study the probability distributions of the charge/discharge behaviors of PEVs. The probability distribution of wind power is also derived based on the assumption that the wind speed follows the Rayleigh distribution. The mathematical expectations of the generation costs of wind power and V2G (vehicle to grid) power are then derived analytically. An optimization algorithm is developed based on the well-established particle swarm optimization (PSO) and interior point method to solve the economic dispatch model. The proposed approach is demonstrated by the IEEE 118-bus test system.

Following the recent devastating blackouts in North America, UK and Italy, blackout prevention has attracted significant attention, though it is known as a notoriously difficult task. To prevent the blackout, it is essential to accurately predict the instable status of power network components. In the large-scale power network however, existing analysis tools fail to perform accurate and in-time prediction of component instability, because of the sophisticated structure of real-world power networks and the huge amount of system variables to be analyzed. To prevent the blackout, we need an accurate and efficient method that (a) can discover interesting features and patterns relevant to the blackout, from the highly complex structure and ten thousands of system variables of a power network, and (b) can give accurate and fast prediction of system instability whenever required, so that the network operator can take necessary actions in time. In this paper, we report our tool developed for power network instability prediction. The proposed method consists of two major stages. In the first stage,a novel type of patterns namely Local Correlation Network Pattern (LCNP) is mined from the structure and system variables of the power network. Correlation rules, which are useful for the network operator to locate potentially instable components, can be further generated from the LCNP. In the second stage, a kernel based network classification method is developed to predict the system instability. By testing on a real world power network (the New England system), we demonstrate that the proposed tool is effective in predicting system instability and thus highly useful for blackout prevention.