Cooperative multi-target encirclement by Unmanned Surface Vessel (USV) swarms poses a challenging problem in maritime operations, due to the increasing difficulty of maintaining effective coordination as swarm size grows. To address this challenge, this paper proposes a novel Transformer-enhanced multi-agent proximal policy optimization (TMAPPO) framework. Specifically, we design a policy network that integrates a Transformer encoder to focus on critical spatial information, while a GRU module is further introduced to enhance temporal decision-making capability. In addition, we design a comprehensive reward function that jointly considers target pursuit, collision avoidance, and encirclement uniformity. Meanwhile, a more realistic training environment is constructed by incorporating USV dynamic characteristics and environmental disturbances. Numerical experiments show that the proposed method achieves higher success rates and rewards than representative baselines, with its advantage becoming more pronounced as the task scale increases. Moreover, the learned policy achieves autonomous target allocation and safe multi-agent encirclement in complex scenarios.

Remotely operated vehicles (ROVs) require accurate hydrodynamic parameters for reliable model-based control. Online identification remains difficult because measurements are noisy, parameters can differ by several orders of magnitude, and weakly excited parameters cause only small changes in the observed motion. This study proposes an enhanced physics-informed neural network (EPINN) to identify 15 hydrodynamic parameters online. The framework integrates sign-preserving log-scale parameterization, robust ensemble training, sensitivity-guided refinement, and sliding-window updates to improve physical consistency, noise robustness, low-sensitivity parameter estimation, and time-varying tracking. Across three noise and colored-disturbance cases, EPINN reduced the overall and low-sensitivity parameter errors by 69.8-89.8% and 65.4-94.7%, respectively, compared with the basic PINN. It also outperformed an augmented-state unscented Kalman filter (AUKF) and recursive least squares (RLS) under severe and time-varying conditions while satisfying the real-time identification requirement. EPINN provides an accurate, robust, and adaptive solution for hydrodynamic identification and high-fidelity model updating, supporting advanced underwater robot control with potential improvements in accuracy, reliability, and safety.

This study proposes a control strategy for multi-tugboats pushing operations based on structure-constraint and presents numerical simulations of a barge performing turning-in-place scenario with complex environmental disturbances. Unlike conventional pushing simulations, which simplify tugboats as barge-side thrusters, the model used in this study incorporates the dynamic influence of fenders. The results demonstrate that the proposed control strategy can effectively achieve the turning of the barge. Furthermore, this study observes that during pushing operations, first-order wave loads cause repeated and unavoidable collisions between the tugboat and barge, resulting in a pulse-like pushing force on the barge. This differs significantly from the steady thrust typically produced by thrusters. Environmental disturbances also create a discrepancy between the average pushing force and the average thrust, and the difference corresponds to the thrust that the tugboat requires to counteract its own environmental disturbances and damping. The conventional pushing operation simulation method cannot simulate these phenomena, leading to unrealistically small tracking errors for the barge in simulation. Therefore, this study asserts that incorporating fender dynamics into pushing operation simulations is critically important.

Competitive decision-making for unmanned surface vessels (USVs) is challenging because adversarial opponents create highly dynamic and uncertain interactions, making it difficult for an intelligent agent to learn a stable and effective policy. To address this challenge, this paper proposes a counterfactual opponent-aware proximal policy optimization (COAPPO) algorithm for competitive decision-making of USVs. COAPPO employs an opponent-aware critic to estimate the Q-function conditioned on actions of both the agent and the opponent, thereby capturing the influence of adversarial behaviors on the decision-making process of the agent. In addition, COAPPO introduces a counterfactual baseline that marginalizes the agent’s own actions while keeping the opponent’s actions fixed, allowing the model to estimate the individual contribution of the friendly agent. Moreover, to enhance training stability, a temporal-difference (TD) learning scheme with λ-returns is incorporated to reduce estimation bias in the opponent-aware critic, while the proximal policy optimization (PPO)-clip mechanism is adopted to constrain policy updates. To evaluate the performance of the proposed algorithm, a two-USV competitive simulation platform was developed for training and testing. Experimental results show that COAPPO outperforms representative reinforcement learning (RL) baselines, demonstrating improved competitive capability and better performance under the tested opponent variations. COAPPO consistently develops distinct adaptive tactics and maintains a high win rate against diverse opponents, indicating robust decision-making capability in dynamic competitive environments.

As a flexible regulatory resource, hybrid energy storage system (HESS) is capable of providing multiple reliable ancillary services, which improves the adaptability of the distribution system to large-scale grid connection of the distributed generation (DG) and alleviate the pressure of peak load and frequency response. In this context, this paper proposes an optimal dispatch strategy of a HESS for DG electricity production and multiple auxiliary service markets to create stackable benefits for HESS operators. Firstly, different types of energy storage system (ESS) (energy-based and power-based) are unified to the joint optimal framework of peak shaving (PS), frequency containment reserves (FCR), and secondary frequency regulation (SFR). By constructing a virtual ESS model based on the idle-time reuse response, an optimal bidding strategy for HESS under this revenue combination is proposed to achieve the optimal utilization of resource allocation. Furthermore, an "hourly-minute-secondly" progressive time series is introduced, and a multi-timescale hierarchical dispatch model named "daily baseline-regulation basepoint-real-time regulation" is constructed. Specifically, the PS capacity is allocated day-ahead and a two-stage capacity allocation method for FCR and SFR is proposed in the intraday, which realizes the parallel optimal of HESS at the scale of full clearance in the auxiliary service markets. Results show that compared to the combined benefits of two types of ESS, the proposed method achieved the comprehensive income increased by 4.87 % and the auxiliary services income increased by 15.2 %. Under this revenue combination, the SFR market can create an additional 60 %-90 % economic value for HESS operators. The proposed hierarchical optimal model can better adapt to the trading rules of different auxiliary service markets and provide guidance for HESS and DG to further participate in the electricity market.

Extreme natural disasters can easily cause large-scale power outages in distribution networks (DN), and energy storage system (ESS) contributes to an essential part of integrated solutions to this problem owing to its flexible regulation and rapid response characteristics. A two-stage robust optimization model for ESS that considers the resilience enhancement of a DN under extreme weather conditions is proposed. First, the impacts of secondary hazards on the component failure rates were quantified, and a time-varying matrix of distribution line failures was constructed. Second, an overall recovery index of the DN and an important load recovery index were proposed. Finally, a two-stage robust optimization model for the ESS is established to improve DN resilience with the objective of minimizing the comprehensive economic cost of the ESS and the annual comprehensive weighted load loss, which is solved using the column-and-constraint generation algorithm (C&CG). Furthermore, numerous simulations were performed on the IEEE 33-node system, and it showed that the proposed method can not only ensure the optimal comprehensive economics of the ESS and fully tap the support potential of the ESS, but also maximize the resilience of the DN. Compared to the DN without energy storage system, the proposed method improves the overall resilience and important load recovery of the DN by about 15.9% and 4.3%, respectively.

As electric vehicles (EVs) are cross-domain entities with dual attributes related to mobile loads and transportation, their travel modes and charging behaviors are stochastic and uncertain in time and space. Large-scale uncoordinated EV charging may cause problems such as the congestion of charging stations (CSs) and overloading of distribution networks (DNs). To address these challenges and the current research gap, and to optimize the spatial-temporal distribution of EV charging loads, a vehicle-road-network collaborative operation framework is constructed in this paper, and a two-stage spatial-temporal scheduling method for EV charging is proposed. In the first stage, real-time traffic and CSs information are introduced to improve Dijkstra's algorithm, and the value preferences of different types of users are considered to further improve the objective function of the algorithm as well. Thus, a dynamic personalized charging navigation model based on improved Dijkstra's algorithm is proposed to provide real-time guidance allowing users to choose their travel paths and CSs. In the second stage, we consider the charging demands of EV users and the operation status of the DN, and an orderly charging model is established to minimize the peak/valley load difference of the DN. Through comprehensive comparisons of experimental results, the variance in the average utilization rates of CSs and the peak valley difference for the DN are reduced by 80.77 % and 16.91 %, respectively. In addition, the time and economic costs of users can be reduced by 9 min and 7.9 RMB, respectively, thereby reducing travel and toll costs and increasing their willingness to participate in scheduling. The simulation experiment proves that the proposed method achieves multi-agent collaborative optimization of EV users, CSs, and DN.

With the growth in the electricity market (EM) share of photovoltaic energy storage systems (PVSS), these systems encounter several challenges in the bidding process, such as the uncertainty involved in photovoltaics, limited bidding ability, and single-revenue structure, which significantly impact the market revenue. To address this research gap, a two-stage bidding strategy based on a non-cooperative game is proposed for PVSS to participate in energy and regulation markets. Considering the complexity of the PV output from adjacent multi-PVSSs, a scenario generation method considering spatiotemporal correlation is proposed. Furthermore, a two-stage bidding strategy is constructed, which includes a bi-level offer price model for the day-ahead (DA) market and a bi-level offer capacity model in the intraday (ID) market. In the DA stage, this study balances the interests of the PVSS and market-clearing costs and considers offer prices in the transaction process. In the ID stage, the market balance cost is considered to further increase the revenue of the PVSS. Moreover, the multi-PVSSs ‘competition relationship is analyzed to coordinate market revenues based on a non-cooperative game. The PVSS adjusts its offer by considering that other PVSSs plans to achieve the Nash equilibrium. The superiority of the proposed strategy was validated using an improved IEEE30 nodes system. Compared to the DA stage bidding strategy, adopting the two-stage bidding strategy can increase the revenue of the PVSS by 5.608%. Specifically, the proposed bidding strategy can increase the revenue of the PVSS by 4.993% compared with a cooperative bidding mode.

With the increasing demand of users for distributed energy storage (ES) resources and the emerging development of peer to peer (P2P) transaction technology, shared energy storage (SES) has great potential to contribute into new business models of demand-side ES. In order to compromise essential elements like safety, stability and efficiency of P2P trading, as well as to improve the utilization rate of demand-side ES, this paper devotes to construct a P2P transaction framework based on a partially decentralized topology and proposes a two-stage trading optimization strategy of SES in a P2P market, considering the equilibrium state of supply and demand flow. In the first stage, this paper simultaneously balances the interests of buyers and sellers and brings the carbon trading mechanism into the transaction process. The interaction of interests of bilateral parties with consideration of carbon trading mechanism has been investigated, and a SES capacity sharing model is, then, established based on the bargaining game theory. In the second stage, a unique pricing mechanism for SES leasing fee is designed based on a multi-strategy evolutionary game model, considering bounded rational decision-making for SES operators and communities. Finally, numerical simulation verified the feasibility and superiority of the proposed P2P trading strategy of SES.

This study proposes a hybrid network model based on data enhancement to address the problem of low accuracy in photovoltaic (PV) power prediction that arises due to insufficient data samples for new PV plants. First, a time-series generative adversarial network (TimeGAN) is used to learn the distribution law of the original PV data samples and the temporal correlations between their features, and these are then used to generate new samples to enhance the training set. Subsequently, a hybrid network model that fuses bi-directional long-short term memory (BiLSTM) network with attention mechanism (AM) in the framework of deep&cross network (DCN) is constructed to effectively extract deep information from the original features while enhancing the impact of important information on the prediction results. Finally, the hyperparameters in the hybrid network model are optimized using the whale optimization algorithm (WOA), which prevents the network model from falling into a local optimum and gives the best prediction results. The simulation results show that after data enhancement by TimeGAN, the hybrid prediction model proposed in this paper can effectively improve the accuracy of short-term PV power prediction and has wide applicability.

Accurate estimation of the State of Health (SOH) for lithium-ion batteries is necessary for the stable operation of the battery system. To accurately estimate the SOH for lithium-ion batteries, we propose an SOH estimation method based on the features of the variation coefficient of partial charging curves, feature processing, and Gaussian Process Regression (GPR). Firstly, the features of the variation coefficient are extracted from the partial charging voltage and current curves as health indicators. The extracted features are efficient and practical, and can effectively reflect the aging phenomenon of batteries. Subsequently, to suppress existing noises, Box-Cox transform (BCT) and discrete wavelet packet transform (DWPT) are employed for the extracted feature signals, thus improving the correlation between the features and the SOH, and ensuring the reliability of the overall framework. Moreover, aiming at the parameters selection problem of the GPR model, an improved particle swarm optimization algorithm with mutation factor and self-adaptive weight adjustment according to population diversity is introduced. Finally, the proposed SOH estimation framework is verified on the NASA battery data set. The experimental results show that the estimation error of the proposed model can be kept within 1.5 % based on different training sample sizes. The results show that the proposed model has high estimation accuracy, generalization, and adaptability.

This study proposes a co-deployment strategy of distribution-level phasor measurement units (D-PMUs) and feeder terminal units (FTUs) for active distribution networks (ADNs) considering node vulnerability to efficiently monitor and perceive the operational state of ADNs. The proposed measurement deployment strategy preferentially configures measurement devices with high precision and superior monitoring functions on vulnerable nodes while ensuring state estimation (SE) accuracy and cost economy. Based on complex network theory and risk assessment, a vulnerability index reflecting the "bridge" characteristics of nodes was defined. Considering the system monitorability, SE accuracy, and measurement cost, a measurement co-deployment strategy based on a combination weighting method was developed. An improved discrete particle swarm optimization algorithm was proposed to solve the co-deployment model by establishing a new particle coding format, improving the dynamic inertia weight and learning factors, and introducing chaos theory. Simulation results based on the IEEE 33-bus and 119-bus distribution systems verify the effectiveness and show that the proposed strategy not only improves system monitorability but also enhances SE robustness.

Owing to the increasing structural complexity and operational flexibility of active distribution networks (ADNs), vulnerable links in networks must be accurately identified for system operation planning and accident prevention. In this study, a node vulnerability assessment method for ADNs considering topological structure and operational characteristics is proposed. First, a network model of the ADN's real-time operating status is established based on the parameterized P-box model and risk theory. Second, an electrical bridgeness index and an operational vulnerability index are proposed to analyze node vulnerability in terms of structural and operational vulnerability, respectively. The subjective and objective weights of the indexes are obtained via a combination weighting method, and the comprehensive vulnerability of the nodes is evaluated and ranked based on the Technique for Order Preference by Similarity to Ideal Solution. To prove the effectiveness and adaptability of the proposed method, a verification analysis is performed based on the IEEE 33-bus distribution system, the 34-bus real test system, and the 118-bus test system. The simulation results show that the accurate identification and effective monitoring of nodes with high vulnerability in ADNs can prevent large-scale system failures.

This paper presents a coordinated islanding partition and scheduling strategy for service restoration of active distribution networks (ADNs) to maximize weighted restored energy considering the minimum sustainable duration (MSD) of controllable islands (CIs). To avoid the forming of uncontrollable islands, the concept of dominated buses is proposed and integrated into the topological model of ADNs for islanding partition. The co-optimization of mobile distributed generator (MDG) dispatch and dominated bus identification is carried out to accelerate the restoration process. The MSD of CIs is explicitly modeled, and the CIs are scheduled in a risk-adjusted way, whose conservatism can be adjusted by controlling the predefined risk confidence level. Interruptible loads are considered as flexible resources in the islanding partition strategy considering the effects of the MSD of CIs. To improve the practicability of the presented method, time-varying power demands and distributed renewable generation outputs are incorporated into the presented framework. The studied problem is formulated as a mixed-integer second-order cone programming model, which can be efficiently solved by commercial solvers. Finally, a modified version of the Pacific Gas and Electric Company 69-bus distribution system, a practical medium-voltage distribution system in Denmark, and a modified IEEE 123-bus distribution system are employed to demonstrate the performance of the developed model.

A regional integrated energy system (RIES) guarantees the fulfillment of a diversified load demand of users by coordinating all types of energy equipment and is a two-layer optimal scheduling method that takes into consideration the flexibility of the characteristics of the combined heat and power (CHP) system is proposed in this paper. First, the thermoelectric output characteristics of hydrogen fuel cell (HFC) and CHP units are analyzed to establish an energy equipment model that takes into consideration the variability of the operating conditions. Subsequently, a two-layer optimal scheduling model of the RIES is established, where the upper layer takes into consideration the operating costs, carbon emission penalties, and renewable energy consumption capacity to determine the strategy for allocating the energy flow, and the lower layer considers the energy efficiency as the optimization objective to determine the output modes of the HFC and CHP units. Finally, an adaptive closed-loop control strategy is used to avoid the continuous startup and shutdown of the CHP unit. The simulation results show that the proposed method can effectively reduce operating costs and carbon emissions, improve energy efficiency and renewable energy consumption, and promote the efficient utilization of integrated energy resources.

To overcome the difficulty in tracking the operation state of distribution networks (DNs) when the specific distribution of system noise and measurement errors is unknown and the measurement is insufficient, a dynamic state estimation (DSE) method based on adaptive set membership filter (SMF) is proposed in this article. First, for the sampling period and measurement delay differences of various measurements, a multisource data fusion strategy was proposed to achieve the synchronization of measurement data at the sampling moment. Subsequently, considering unknown but bounded (UBB) noise, an ellipsoid-based DSE model was established, which unified the form of multisource data through measurement transformation strategies and linearized the measurement function. Then, an adaptive SMF considering bad data detection was proposed to solve the proposed DSE model. The state variables at different moments were iteratively solved through three steps: time update, bad data adaptive detection, and measurement update. Finally, the effectiveness and robustness of the proposed method were verified based on the IEEE33-bus distribution system, the 118-bus test system, and the 34-bus real test system.

To improve the comprehensive utilization efficiency of energy, a multi-objective optimization control strategy applied to the energy hub (EH) within the system is proposed to address the electrical and thermal load distribution of the integrated energy system (IES) and the low-carbon economic operation. First, a model of the electrical and thermal energy outputs is established based on the characteristics of the IES network and the multidimensional "load parameter" evolution law. Moreover, a distributed control strategy is proposed that utilizes the information interaction between neighboring EHs to accurately share the electrical and thermal loads, which reduces the communication burden of the system while avoiding overloads. Furthermore, to coordinate the optimal operation of the devices within the hub, based on the energy conversion characteristics of the EH, a multi-objective optimization model is proposed that considers low-carbon and economic aspects to realize efficient energy use. Simulation results show that the proposed strategy effectively improves the robustness of the system while realizing proportional load distribution and low-carbon economic operation. Under the same load, when focusing on system economics, the operating cost is 2.182/& YEN; & YEN; lower than when focusing on low-carbon systems, but carbon emission is 1.6753/kg CO2 2 higher.

高比例新能源并网使得配电网对通信系统的依赖性不断提高.为降低通信设备故障对分布式控制系统运行的影响,提出一种考虑多重通信故障的配电网失联分布式电源(DG)群优化控制方法.首先,建立了分布式通信网络的通信链路矩阵,并基于通信链路搜索得到了典型的故障场景.然后,考虑通信故障场景及源荷的不确定性,以综合风险最小为目标,建立了失联DG群双层优化模型,模型外层优化失联DG多场景控制策略,模型内层优化未失联DG控制策略.最后,采用改进粒子群优化算法求取失联DG群控制节点的最优控制策略.以IEEE 33节点算例系统为例,验证了该控制方法的有效性.结果 表明所提方法能够有效降低系统运行风险,提高供电可靠性.

Given the difficulty of traditional state estimation methods in meeting the computational efficiency and accuracy requirements of active distribution networks, we propose a distributed state estimation (DSE) method for active distribution networks based on the weighted least squares-adaptive Kalman filter (WLS-AKF) hybrid algorithm. A multi-criteria partition model of an active distribution network that is suitable for DSE has been established. The model comprehensively considers the impact of partitioning results on DSE calculation accuracy and efficiency from the three perspectives of structure, measurement, and performance and solves the model using an improved genetic algorithm. A partition decoupling method is proposed herein, which completely decouples the sub-regions without requiring the measurement configuration of the sub-region boundary nodes and effectively re-duces the DSE calculation scale. Furthermore, the paper proposes a DSE algorithm based on WLS-AKF that uses AKF to provide accurate pseudo-measurement of boundary nodes for WLS, which improves the calculation ef-ficiency while ensuring the DSE calculation accuracy. The proposed method is analyzed and verified using the improved IEEE118 node system. The results show that the proposed method has high computational accuracy and efficiency, and can obtain high-precision estimation results in the case of missing data.

Power prediction can effectively mitigate the uncertainty in photovoltaic power generation, enabling better operation and scheduling of power grids. Therefore, in this study, a multi-step interval prediction method for ultra-short-term photovoltaic power from time-series-segment (TSS) feature analysis is proposed. First, three TSS indicators are defined to determine the fluctuation characteristics of historical data and combined with fuzzy C-means clustering to address the time mismatch problem. Subsequently, a deterministic multi-step prediction method is proposed based on the optimal membership search using deep recurrent neural networks, improving the prediction stability. Finally, based on the difference in the TSS types, an improved interval prediction method is proposed in combination with Gaussian process regression, narrowing the average interval width. Experiments are conducted to compare the performances of the conventional and proposed methods using measured data from Australia. Compared with the baseline scheme, the proposed scheme enhances the accuracy of multi-step prediction by 19.7%, and the average error of each step does not exceed 5%.The average interval width is reduced by 45.6% while ensuring more than 95% interval coverage in the probability interval prediction. The experimental findings demonstrate that the TSS feature analysis can effectively reveal the potential patterns of PV power output under various weather conditions. This enables the algorithm to learn clearer sample features and thus enhances the performance of multi-step probability interval prediction.