Hybrid Deep Adversarial Network Rolling Reinforcement Learning Algorithm For Smart Generation Control

With the continuous access of large-scale intermittent distributed energy and emerging load to power system,the connection between regional grid network and energy network is increasingly close.The expanding energy system has put forward higher and higher requirements for the optimal operation and control of power grid.Economic and high-quality energy supply has become an urgent problem for electric power workers.To this end,this dissertation is dedicated to researching smart power generation control in the context of microgrids and integrated energy systems,and designing smart control algorithms from multiple objectives such as economy,network loss,and environmental protection,and taking into account control performance.The specific research content and results are as follows:(1)The basic idea of generative adversarial networks is analyzed,and then the modeling of generative adversarial networks is introduced from the perspective of game theory;then the training process of generative adversarial networks is introduced by deriving the optimal solutions of generators and discriminators;the methods and research progress of generative adversarial networks in performance evaluation are discussed;the methods to improve the convergence speed of generative adversarial networks and the improvement methods are discussed;the implementation process of generative adversarial networks is introduced.(2)To solve the problem of uncoordinated operation of the traditional multi-time-scale generation mode of power systems and the problem of mismatch between optimization algorithms and control algorithms,a unified time-scale smart generation control framework is proposed for the generation control of microgrids in order to solve the discrete problem in generation data and improve the control performance of the system,and a time series generative adversarial network is proposed for this framework.First,the reinforcement learning is combined with the generative adversarial network,and the generative adversarial network learns the historical states and historical actions values of the system and fuses them in the reinforcement learning framework;then,to solve the problem of discrete data in microgrids,the strategies in reinforcement learning are applied to the generative adversarial network;finally,the unified time-scale smart generation control framework and the time series generative adversarial network algorithm are proposed and applied in Hainan power grid,IEEE300 node system and IEEE1951 node for simulation verification,which verifies the effectiveness and feasibility of the algorithm.(3)In order to solve the problem of uncoordinated multi-timescale scheduling mode of integrated energy,and to solve the problem of uneven energy distribution while realizing multi-energy complementarity,a smart generation control framework with unified time-scale regulation and control integration and a relaxed deep generative adversarial networks algorithm are proposed for the integrated energy system.First,the smart generation control framework with unified time-scale regulation and control integration is designed to replace the traditional conventional combined framework;then,the relaxation operator is fused in the generative adversarial network,and a multi-objective optimization model involving energy supply cost,carbon emission and control performance is built to apply the proposed relaxed deep generative adversarial networks algorithm to the multi-objective optimal control problem of the integrated energy system,so as to meet the system performance requirements;finally,simulations are performed to verify the effectiveness of the algorithm in the built simple integrated energy system and the IEEE 14-bus with 10natural-gas systems,which can improve the control performance of the system,consume renewable energy,improve the system stability,achieve multi-energy complementarity,and avoid energy surplus.