[Objectives]In response to the significantly increased complexity of fault recovery in distribution networks caused by high-penetration distributed generation(DG)and the insufficient stability of islanded operation,a fault recovery method incorporating net load prediction is proposed for DG-integrated distribution networks.[Methods]A hybrid prediction model of convolutional neural networks and bi-directional long short-term memory networks(CNN-BiLSTM)incorporating a meteorological feature attention mechanism is designed to achieve high-precision net load prediction during fault periods.Then,a multi-objective power supply recovery model is developed,and an optimized genetic algorithm and Broyden-Fletcher-Goldfarb-Shanno is adopted to solve island partitioning and reconfiguration scheme based on net load prediction results.[Results]Simulation results based on the PG&E 69-node system show that the prediction accuracy of the proposed method is relatively improved by 21.8%,the continuous power supply time of one of the islands is increased by 150%,and the increased power supply reaches 198.464 kW·h.[Conclusions]The proposed strategy effectively solves the island instability problem caused by DG randomness,providing a new method for rapid self-healing of high-proportion new energy distribution networks.
ObjectivesDue to the difficulty in obtaining comprehensive and accurate raw activity data throughout the life cycle, it is challenging to evaluate the carbon emissions of renewable energy based on carbon footprint analysis. A low-carbon performance evaluation method based on the carbon emission contribution rate of key factors throughout the life cycle is proposed.MethodsConsidering activities with high energy consumption and high emissions in the life cycle of renewable energy power generation projects, key factors affecting carbon emissions of each activity are selected. On this basis, a low-carbon performance evaluation indicator system is established for renewable energy, covering stages such as manufacturing, equipment transportation, engineering construction, power station operation and maintenance, and waste disposal and recycling. The evaluation indicators are standardized according to the positive or negative sign and the level of the carbon emission contribution rate of each key factor, and then the corresponding weight values are determined.ResultsSix different types of renewable energy power stations in China are selected for case analysis of the low-carbon performance evaluation. The results are consistent with the carbon footprint, and in line with the actual carbon emission intensity of each life cycle stage of various renewable energy power stations. The case analysis proves the reliability and effectiveness of the proposed method.ConclusionsThe proposed evaluation method effectively addresses the problems of high data quality requirements and significant uncertainty in carbon footprint calculations. The evaluation results are relatively accurate and comprehensive, and the method holds great significance for tapping into the carbon emission reduction potential across the entire renewable energy industry chain.
ObjectivesTo address the collaborative optimization difficulties arising from participation of multiple market entities, deep integration of distributed resources, and multi-level decision coordination in new power systems, where traditional centralized optimization models struggle to adapt in terms of privacy protection, distributed computation, and interest coordination, this study aims to clarify the theoretical connotation, generalized framework, and applicability boundaries of integrating evolutionary game theory (EGT) and federated learning (FL) in power systems, thereby resolving the key scientific problem of multi-agent strategy coordination under privacy constraints.MethodsFollowing the main thread of “problem-driven-theoretical support-method construction-scenario implementation”, the respective mechanisms of EGT in characterizing the strategy evolution of bounded rational agents and of FL in achieving privacy-preserving distributed modeling are systematically reviewed. The coupling principles of their integration (EGT-FL) are then elaborated from four perspectives, namely dynamical isomorphism, information-theoretic consistency, learning-theoretic unification, and optimization-objective correspondence, and a generalized integration framework adapted to the physical constraints of power systems is constructed. On this basis, key methods including multi-agent game modeling, federated strategy evolution algorithm design, and incentive-privacy co-optimization are summarized, and simulation verification is conducted on a modified IEEE 33-bus system and a large-scale demand response case involving 1 050 agents.ResultsSimulation results demonstrate that the EGT-FL algorithm outperforms FL algorithms such as FedAvg and FedProx, as well as classical game-theoretic methods such as Nash equilibrium and Stackelberg game, across five dimensions: convergence, privacy protection, economic performance, computational efficiency, and robustness. The convergence speed is improved by 30%-50% compared with traditional methods. Under a differential privacy parameter of ε=1, more than 90% of the optimization performance is still maintained. The total system cost increases by only about 3.2%, while the privacy protection level reaches 95%. The system stability is maintained above 70% under a 20% proportion of malicious agents.ConclusionsThe integration of EGT and FL can effectively resolve conflicts of interest among multiple agents while preserving data privacy, and achieve system-wide benefit maximization, providing a new paradigm with both theoretical rigor and engineering feasibility for multi-agent collaborative decision-making in new power systems. The deep integration of federated reinforcement learning, cross-level collaborative optimization, and adaptive privacy mechanisms are key research directions for future breakthroughs, offering theoretical support and technical reference for the intelligent operation of new power systems.
ObjectivesIn recent years, transitional weather has occurred frequently, and the randomness and volatility of wind power generation have intensified. Existing day-ahead wind power prediction schemes struggle to balance the interval coverage rate and interval width of wind power prediction. Therefore, a robust error correction strategy is proposed to achieve high-quality day-ahead wind power interval prediction.MethodsFirst, a kernel-based fuzzy C-means clustering algorithm is proposed, which is combined with multi-step backward cloud transformation based on sampling with replacement to accurately cluster error types under complex transitional weather conditions. Then, a point prediction model based on temporal convolutional network-Transformer is established, and an improved kernel density estimation day-ahead wind power interval prediction method is designed to improve the fitting accuracy of power interval. Finally, an improved multi-objective dung beetle optimizer algorithm is designed to perform robust error correction across different clusters, and the prediction performance of the proposed method is verified using measured wind power data.ResultsThe day-ahead wind power prediction interval after robust error correction exhibits a smaller interval width and higher interval coverage rate. Under the 95% confidence level, prediction interval coverage probability increases by a maximum of 5.884%, and prediction interval normalized average width reduces by a maximum of 35.01%, validating the effectiveness of the proposed method.ConclusionsThe proposed method significantly improves the fitting accuracy of error distribution, avoids the problem of poor interval quality caused by local over-correction or under-correction, and greatly enhances the efficiency of the algorithm search, thereby achieving higher-quality wind power interval prediction.
ObjectivesImproving the operational flexibility of coal-fired units is a major requirement for building a new power system, which puts forward higher requirements for stable combustion of coal-fired boilers at ultra-low loads.MethodsTaking a 660 MW ultra-supercritical swirling opposed-firing pulverized coal boiler as the research object, the concept of local critical heat load of the boiler is put forward. Using numerical simulation method, the combustion stability of the boiler under different loads is analyzed with the volume-averaged temperature of the burner region as the index of local regional heat load. Based on this, the stable combustion strategy of increasing the power of a single burner under ultra-low loads is put forward, and the local critical heat load of the boiler is obtained.ResultsUnder the conventional operation mode, when the load is reduced from 25% to 20%, combustion instability occurs in the boiler. At this time, the volume-averaged temperature drop in the main combustion area and the local heat load area behind the burner nozzle are 415 K and 174 K respectively which are much higher than 31.5 K and 30.5 K in the process of reducing from 30%Pe (Pe is the rated load) to 25%Pe. After adopting the steady combustion strategy of increasing the heat load of a single burner, the volume-averaged temperature in the local heat load area under 20%Pe is raised to the level under 30%Pe, and the combustion stability of the boiler is obviously improved.ConclusionAt 20%Pe, the local critical heat load to ensure the stable combustion of the studied boiler is that the operating power of a single burner reaches 60% of its rated power. The research results can provide an effective reference for the deep peak regulation and ultra-low load stable combustion of coal-fired units.
ObjectivesUsing the power generation per ton of water as an evaluation index for system performance, this study investigates the variation patterns of various thermodynamic performance parameters in both regenerative and non-regenerative organic Rankine cycle power generation systems under different geothermal source fluid temperature conditions.MethodsThe EES software is used to establish models for both regenerative and non-regenerative power generation systems, with R245fa selected as the organic working fluid. The variation patterns of key thermodynamic parameters are explored in a power generation system with an installed capacity of 1 MW, operating at geothermal source temperatures from 100 ℃ to 140 ℃.ResultsAt a given geothermal source temperature, as the evaporation temperature of the organic working fluid increases, the power generation per ton of water reaches a maximum value for both regenerative and non-regenerative power generation systems. Additionally, the power generation per ton of water produced by the regenerative system is greater than that of the non-regenerative system. When the geothermal source temperature ranges from 100 ℃ to 140 ℃, the net efficiency and output power of the regenerative power generation system increase with rising geothermal source temperature, whereas the power consumption of the cooling system decreases as geothermal source temperature increases. The net efficiency and output power of the regenerative power generation system are greater than those of the non-regenerative system, while the power consumption of the cooling system is less than that of a non-regenerative power generation system.ConclusionsThe research findings can provide a reference for the selection of thermodynamic parameters for MW-scale regenerative and non-regenerative organic Rankine cycle power generation systems.
ObjectivesAlthough the hybrid photovoltaic-thermoelectric generation (PV-TEG) system can improve energy utilization efficiency by photovoltaic waste heat recovery, it still faces problems such as asynchronous optimization of electro-thermal parameters, rigid battery compensation mechanisms, and the tendency of traditional algorithms being prone to local optimality or unstable strategies under partial shading conditions (PSC), seriously restricting the system efficiency.Therefore a dynamic compensation strategy based on the improved soft actor-critic (ISAC) algorithm is proposed.MethodsBy optimizing PV array topology logic, TEG power regulation patterns, and battery adaptive compensation mechanisms, the synchronous optimization of electro-thermal parameters and dynamic environment adaptation are realized. The MATLAB/Simulink simulation platform is adopted to conduct performance verification on arrays of different scales such as 6×4 and 6×6, and the optimization effects are compared with those of traditional heuristic algorithms and traditional reinforcement learning algorithms.ResultsCase verification under 10 PSC scenarios indicates that, compared with traditional algorithms, the proposed strategy based on ISAC reduces the mismatch loss of 6×4 and 6×6 arrays by 12.3%-18.7% and 13.6%-19.2% respectively, with an overall power increase of 4.2%-7.3%.ConclusionsThe algorithm and its corresponding strategy effectively address the core issues of poor coordination and weak dynamic adaptability of existing hybrid PV-TEG systems, thereby providing reliable technical support for the efficient utilization of solar energy in complex shading environments. It holds significant implications for improving the operation efficiency and stability of renewable energy systems.
ObjectivesWith the gradual implementation of the “dual carbon” policy, the use of clean energy has been strongly promoted in recent years. Among them, the proton exchange membrane fuel cell (PEMFC), which uses hydrogen as fuel, has become an important trend in energy system development, and hydrogen fuel cell systems have therefore been widely promoted and applied. To extend the service life of PEMFC systems and improve their output performance, existing control strategies for fuel cell systems are reviewed and summarized.MethodsThe energy flow characteristics of fuel cell are described. In view of the three major control strategies that consider the intrinsic characteristics of PEMFC systems—air supply control, temperature control, and cold-start control—this study reviews and summarizes existing control strategies in light of the nonlinear, time-varying, strongly coupled, and multi-input-multi-output characteristics of the system, and analyzes and compares the advantages and disadvantages of various control strategies.ConclusionsKalman filter control enables real-time estimation and compensation of disturbances in the air supply system. Model predictive control can effectively address the large temperature lag in existing studies. For cold-start control, a stepped current density loading mode considering uncertainty in membrane water content can effectively reduce the cold-start time and improve output characteristics. These findings provide a theoretical basis for the development of hydrogen energy and PEMFC.
ObjectivesWith the widespread application of distributed renewable energy and flexible loads, the power interactions between the active distribution network (ADN) and the main grid have become increasingly complex. Traditional load models usually rely on fixed structures and parameters, making them unable to adapt to the time-varying and stochastic nature of loads. To address this, this study proposes a dynamic equivalence modeling method for ADNs based on a hierarchical double deep Q network (HDDQN).MethodsFirst, the ADN is initially modeled using multiple dynamic and static load models. The equivalence task is then abstracted as a two-stage, three-layer Markov decision process. Finally, the HDDQN algorithm is applied for online parameter identification. The HDDQN algorithm utilizes gated recurrent units for feature extraction from the input discrete time-series data and incorporates a competitive network and a prioritized experience replay mechanism to optimize traditional reinforcement learning algorithms.ResultsAfter sufficient training, the proposed method generates strategies in just 2.8 s. Its average equivalence accuracy for active and reactive power is 1.64 times and 8.10 times higher, respectively, compared to traditional methods.ConclusionsThe proposed method enables online identification of equivalent model parameters and significantly outperforms traditional methods in terms of convergence performance and equivalence accuracy. It is highly valuable for improving the efficiency and accuracy of real-time ADN modeling.
ObjectivesMost of the current prediction models for the combustion state of power station boilers can predict only a single objective. Due to different types of objectives, their data exhibit various temporal distribution characteristics. Besides, the simple long short-term memory (LSTM) neural network has limitations in predicting multiple different types of objectives. Given the above challenges, a multi-scale feature fusion enhanced LSTM neural network model is proposed.MethodsFirst, a multi-scale LSTM neural network model is developed and compared with the conventional LSTM neural network model. Then, on the basis of this multi-scale model, two optimization modules, including the spectral attention mechanism and the self-modulation feature fusion module, are gradually added to develop the multi-scale feature fusion enhanced LSTM neural network model, thereby improving the ability of the model to identify, enhance, and effectively integrate data variation features. Finally, the proposed model is evaluated on the same dataset.ResultsCompared with the simple LSTM neural network model, the multi-scale model achieves a significant improvement in accuracy, verifying the effectiveness of the model in capturing temporal characteristics of different types of prediction objectives. Moreover, with the increasing number of optimization modules, the model performance is further enhanced, indicating that the addition of modules strengthens the ability of the model to process temporal characteristics.ConclusionsThe proposed model effectively addresses the problem of insufficient prediction accuracy in boiler multi-objective prediction caused by different temporal distribution characteristics of objective types, resulting in more accurate predictions and providing significant value for predicting multiple types of objectives.
ObjectivesThe fluctuation of renewable energy generation and frequent load switching may cause frequent variations in the power flow direction of energy storage systems, compromising the stability of the bus voltage and system dynamic response performance. Therefore, this study proposes a virtual inertia control strategy for DC bus voltage in integrated energy system based on electric-hydrogen hybrid energy storage.MethodsVirtual inertia control is introduced into the voltage outer loop, and model predictive control is introduced into the current inner loop. The virtual inertia parameters are combined with the voltage change rate to establish an adjustment relationship between voltage and virtual capacitance. Building on this, a power coordination control method suitable for electric-hydrogen hybrid energy storage systems is designed, and the economic feasibility of different energy storage devices is compared. Simulation models are established using MATLAB/Simulink to verify the effectiveness of the proposed strategy.ResultsThe proposed strategy reduces voltage fluctuation range to 2.2%, ensuring efficient coordinated operation of the electric-hydrogen hybrid energy storage system while improving the safety of hydrogen storage tanks. Compared with single lithium battery storage and lithium battery-supercapacitor hybrid storage solutions, the electric-hydrogen hybrid energy storage solution achieves cost reductions of 10.55% and 3.45%, respectively.ConclusionsThe proposed strategy effectively mitigates bus voltage fluctuations caused by load power fluctuations, enhances the system dynamic response capability, and contributes to the stable operation of park-level integrated energy systems.
ObjectivesThe coordinated control of reactive power and voltage optimization based on optimal power flow theory can effectively reduce system voltage deviations during wind farm grid connection. However, traditional methods rely on detailed parameters of wind farms and have a long solution time, which poses challenges to the online application of these methods. To overcome this challenge, taking typical offshore wind farms as the research objects, this study proposes an optimization strategy for the coordinated control of reactive power and voltage in offshore wind farms based on improved deep deterministic policy gradient (iDDPG).MethodsFirst, an optimized operation model for reactive power-voltage coordination in offshore wind farms is established with the objectives of minimizing voltage deviations and system losses. Second, a method is proposed to transform the coordinated optimization problem of reactive power and voltage into a Markov decision process (MDP). By defining system states, actions, and a reward function, the multi-constraint optimization problem is converted into an unconstrained deep reinforcement learning problem. Then, combined with the random power output data of wind turbines, the iDDPG is employed to solve the coordinated optimization decision for reactive power and voltage in offshore wind farms. Finally, the effectiveness of the proposed model and algorithm is validated through simulation cases.ResultsThe results indicate that compared to traditional methods, the proposed method has advantages in model solution accuracy and real-time response speed.ConclusionsThe proposed method can improve the voltage stability of offshore wind farms.
ObjectivesTo mitigate voltage fluctuations at the grid-connected points of doubly-fed induction generator wind farms while enhancing their reactive power margin, thereby strengthening their transient voltage support capability, a reactive power margin optimization method based on an improved proximal policy optimization algorithm is proposed for doubly-fed induction generator wind farms.MethodsTaking doubly-fed induction generator wind turbines and static var generators as the main reactive power regulation equipment, a mathematical model for reactive power margin optimization of a doubly-fed induction generator wind farm is established and transformed into a Markov decision process model. Addressing the issue of improper setting of the clipping coefficient in the original proximal policy optimization algorithm, which makes it difficult to balance the exploration and exploitation behaviors of the agent, a proximal policy optimization algorithm based on a variable clipping coefficient is proposed. Based on the improved IEEE-39 node system, the correctness and effectiveness of the proposed algorithm are verified.ResultsThe proposed algorithm can improve the convergence speed of agents, and is superior to traditional proximal policy optimization algorithms and heuristic algorithms in terms of decision speed and reactive power margin optimization.ConclusionsThe proposed algorithm effectively improves the reactive power margin of doubly-fed induction generator wind farms, enhances their voltage support capability under transient conditions, and is of great significance for better solving reactive power optimization problems in power systems with uncertain sources and loads.
ObjectivesArray-type heat exchangers, renowned for their excellent power regulation capability, are one of the important means to ensure the rapid response of advanced adiabatic compressed air energy storage (AA-CAES) systems to control commands. However, existing research on the use of array-type heat exchangers in AA-CAES is relatively limited, and control methods based on the structure of array-type heat exchangers urgently need to be investigated.MethodsFirst, an array-type heat exchanger structure suitable for AA-CAES systems is proposed, which can alter its connection configuration to respond to power change commands of different magnitudes. On this basis, a power generation system model of the AA-CAES system incorporating the array-type heat exchanger is established. This model fully considers the influence of fluid physical properties on each equipment of the power generation system, thereby improving model accuracy. Then, a PID-based control strategy for the AA-CAES power generation system and a small-amplitude power control strategy are developed to enable power command tracking. Finally, the effectiveness of the control strategies is validated through simulation case studies, and the system’s power tracking performance is analyzed.ResultsAA-CAES can achieve power tracking over a wide range after employing the array-type heat exchanger. Additionally, adopting small-amplitude power control can effectively enhance the system’s ability to respond to rapid, small-amplitude power change commands when consuming new energy generation.ConclusionsProposed control strategy can broaden the operational capability of the AA-CAES power generation system under variable conditions and enhance system flexibility.
ObjectivesZhundong coal is characterized by large reserves and favorable combustion properties, but its high alkali metal content tends to lead to fouling on boiler heating surfaces. Traditional mechanism-based and data-driven methods face challenges such as difficulty in mechanism simplification and insufficient labeled data in fouling monitoring modeling. Therefore, there is an urgent need for accurate and reliable models for fouling quantification characterization to guide sootblowing decisions and optimize diagnostics. To address these issues, a modeling method for heating surface fouling monitoring based on machine learning is proposed.MethodsTaking the boiler of a 1 000 MW power plant as the research object, a high-precision dynamic simulation model incorporating the control system is established. On this basis, new characteristic parameters are established, and a fouling quantification characterization modeling method based on autoencoder (AE) and long short-term memory (LSTM) neural network is proposed.ResultsThe fouling monitoring model for heating surface based on new characteristic parameters such as heat transfer deviation and outlet steam temperature deviation shows false alarm rates of 0.6% under stable load conditions and 1.2% under variable load conditions, effectively mitigating the influence of load fluctuations on thermal parameters. The fouling monitoring model based on AE demonstrates high accuracy and robustness.ConclusionsThe proposed model can accurately monitor the fouling trends of heating surfaces during sootblowing cycles, providing new insights for boiler heating surface fouling monitoring.
ObjectivesUnmanned and intelligent high-efficiency inspection is of great significance for improving the operational reliability of wind turbines and reducing operation and maintenance costs. To address the severe challenges posed by strong turbulence in turbine wakes during operation and the complex multi-obstacle environment in mountainous wind farms, an intelligent unmanned aerial vehicle (UAV) inspection path planning method considering wind farm wake effects is proposed.MethodsFirst, a global inspection problem model is established with inspection path cost and UAV endurance time as optimization objectives, and UAV flight characteristics as constraints. A virtual inspection scenario is then designed to reflect the complex terrain and wake characteristics of a mountainous wind farm. Next, a policy interactive target bias rapidly-exploring random trees (PITB-RRT) algorithm is proposed to improve path efficiency and computational efficiency by optimizing sampling and extension strategies. Finally, the non-dominated sorting genetic algorithm Ⅱ (NSGA-Ⅱ) is introduced to optimize the sequence of inspection points.ResultsFor a typical mountainous wind farm with 12 wind turbines, the proposed method efficiently plans 66 feasible paths between turbine inspection points. The globally optimal inspection path obtained has a total length of approximately 9 342 meters and a corresponding flight time of 25.3 minutes, reducing the path cost by 28.8% compared to traditional inspection sequences.ConclusionsThe proposed method exhibits good applicability and high efficiency in complex wind farm scenarios, providing technical support for the intelligent and automated execution of UAV inspection tasks.
ObjectivesCurrent studies on energy saving of ash removal systems mainly rely on equipment modifications combined with numerical simulation, but their timeliness is limited. In contrast, the Bayesian optimization algorithm can efficiently explore the parameter space, improve the analysis efficiency of energy saving research in ash removal system, shorten the optimization process time, promote the timeliness of numerical simulation, and enhance the efficiency of ash removal system by optimizing the parameters of aided blowing, thereby achieving energy saving and consumption reduction, and improving the economic performance of power plants. Therefore, this study is conducted.MethodsThe computational particle fluid dynamics (CPFD) is used to conduct numerical simulation of the modified ash removal system, and the Bayesian optimization algorithm is employed to perform iterative calculation of different aided blowing parameters (aided blowing speed, aided blowing distance, aided blowing angle, and aided blowing position z).ResultsA total of 30 iterative cycles of optimization on the aided blowing parameters of ash removal system of power plants are carried out through the Bayesian optimization algorithm and CPFD. The energy consumption of the system is the lowest when the aided blowing distance is 2.4 m, the aided blowing speed is 2 m/s, the valve position is z1, and the aided blowing angle is 48°.ConclusionsThe system with valve optimization achieves an energy saving of 24% compared to the system without valve optimization, and an energy saving of 54.4% compared to the system without valves, demonstrating significant optimization performance.
ObjectivesThermal energy storage technology can effectively mitigate the intermittency and volatility of renewable energy sources and holds great potential for enhancing the stability of energy systems. Solid particle packed beds, with their simple structure, low cost, and high thermal energy storage temperatures, offer broad prospects for development. To address the issue of optimal flow channel layout in solid particle packed beds, this study investigates flow channel configurations and thermal energy storage performance across a wide temperature range.MethodsUsing rock particles as the thermal storage medium, a computational fluid dynamics model of a solid particle packed bed is established. The performance of the packed bed is analyzed for different baffle configurations to determine the optimal flow channel layout. Based on this, the study systematically examines the effects of parameters such as the packed bed’s height-to-length ratio, inlet air flow rate, temperature, particle diameter, and porosity on the overall performance of the thermal energy storage system.ResultsThe double-baffle configuration with a parallel staggered layout effectively enhances thermal energy storage performance and improves the temperature distribution within the packed bed. The system’s overall performance is optimized when the height-to-length ratio of the packed bed is 0.557. Both the system’s thermal energy storage rate and pump power consumption increase with rising inlet air flow rate and temperature, while thermal energy storage efficiency and exergy efficiency decrease as the inlet air temperature rises but increase as the inlet air flow rate increases.ConclusionsInlet air flow rate, particle diameter, and porosity have the most significant regulatory effects on the overall thermal performance of the system. The research findings provide a basis for structural optimization of solid particle packed bed systems, matching of operating conditions, and multi-parameter co-optimization.
ObjectivesTo address the core challenges that traditional optimal power flow algorithms have limited adaptability to complex multi-objective optimization problems and that existing methods fail to simultaneously meet real-time and accuracy requirements in large-scale power grids, this study proposes an innovative algorithm which integrates a high-dimensional reduction mapping with a distributed SARSA(λ) collaborative optimization framework.MethodsFirstly, a high-dimensional reduction mapping mechanism is introduced that performs feature extraction and dimensionality compression on the action space of the power grid to map the original high-dimensional control variables to a low-dimensional representation space, thereby fundamentally overcoming the curse of dimensionality. Secondly, a multi-agent collaborative learning framework based on distributed SARSA(λ) is established. Each agent has both online learning and multi-step backtracking optimization capabilities. While independently performing strategy iteration locally, the agents achieve the integration of distributed learning with centralized optimization through cross-agent boundary information exchange and a global reward allocation mechanism.ResultsValidation on the IEEE 118-node standard test system demonstrates that the proposed method significantly enhances the solution efficiency and decision-making quality for multi-objective optimal power flow problems while maintaining compliance with physical constraints.ConclusionsThe findings provide a new approach with both theoretical innovation and engineering feasibility for the intelligent optimization of power systems in high-dimensional uncertain environments.
ObjectivesAgainst the backdrop of energy transition and large-scale grid integration of wind power, and aiming at the increasingly complex electro-mechanical coupling characteristics of wind turbines with frequency-supported control, this study conducts theoretical and simulation analyses on the load characteristics of direct-driven permanent magnet synchronous generators (D-PMSG) under virtual synchronous generator (VSG) control.MethodsBased on multi-body dynamics and small-signal analysis theory, a linearized electro-mechanical coupling model of VSG-PMSG is constructed to characterize the dynamic relationship between loads and source-grid excitation, followed by an analysis of the response mechanisms of VSG-PMSG under impact and fatigue loads. A nonlinear electro-mechanical coupling model of VSG-PMSG with multi-time-scale interactions and multi-dynamic link effects is established. Simulation analysis of load characteristics of VSG-PMSG is conducted under typical disturbance scenarios, such as turbulent wind and grid impacts. Indicators including maximum load increase and equivalent fatigue load are selected to evaluate the effects of VSG control on turbine impact and fatigue loads.ResultsUnder turbulent wind excitation, VSG control can reduce the drivetrain fatigue load and the tower bottom side-to-side fatigue load. In addition, under grid-side frequency excitation, VSG control can induce drivetrain impact load and the tower bottom side-to-side impact load. However, these are significantly smaller than the transient impacts caused by three-phase short-circuit faults.ConclusionsVSG control alters the electrical and load characteristics of wind turbines, particularly affecting the drivetrain mode and the tower bottom side-to-side mode. The research findings provide a theoretical basis for control strategies and mechanical structure design of frequency-supported wind turbines.
