Dynamic Collaborative Learning For Smart Generation Control Of Power Systems

In order to effectively control air pollution and deal with the energy crisis,the power system must actively transform into an intelligent decentralized renewable energy power system.With the large-scale interconnection of renewable energy,the randomness and uncertainty of power system will increase geometrically.How to maximize the compatibility between renewable energy and power system has become a hot issue in the energy field at home and abroad.In this paper,a new energy-oriented distributed intelligent generation control(SGC)strategy is explored from the viewpoint of automatic generation control(AGC).Based on stochastic game theory,the stochastic optimal control problem of the above-mentioned distributed SGC is modeled as a multi-agent stochastic game problem,which is solved by multi-agent reinforcement learning and deep reinforcement learning respectively.In order to deal with the strong random disturbance caused by large-scale grid-connected renewable energy,and promote the compatibility between renewable energy and power system.Firstly,the severe challenges faced by centralized AGC are analyzed in detail from three perspectives of new energy utilization efficiency,coordinated control of multi-regional power grids and fast dynamic response.Based on this,a new distributed SGC strategy for new energy is proposed.At the same time,the necessary technologies of distributed SGC are analyzed in detail from three perspectives of distributed control framework,multi-objective optimization and strategy solution.Breakthroughs and feasible ways.Secondly,a novel multi-agent reinforcement learning method,PDWo LF-PHC(λ),is proposed to synthesize the optimal solution of distributed SGC.By using the variable learning rate,the global dynamic performance of the system can be improved.At the same time,the asynchronous decision-making problems faced by traditional reinforcement learning in multi-agent systems and the multi-solution problems caused by the increase of the number of agents in large-scale tasks can be solved.Based on this,the proposed method can effectively explain the interaction and optimal equilibrium state of distributed SGC,so as to deal with the strong random disturbance caused by large-scale grid-connected new energy.The two-area power system model,the central China power grid model and the intelligent distribution network model with a large number of new energy sources are simulated.The results show that the method has fast convergence and strong robustness,and can reduce carbon emissions while improving CPS qualification rate and new energy utilization efficiency.Finally,an efficient deep reinforcement learning method,DDRQN-AD,is proposed to realize the optimal coordinated control of distributed SGC.The proposed method can solve the problem that knowledge matrix storage of traditional reinforcement learning takes up so much space that it is difficult to solve,thus significantly reducing the information processing burden of SGC.It can solve the problem of limited action set in traditional reinforcement learning,so as to improve the dynamic accuracy of SGC regulation instructions.Based on this,the proposed method can effectively deal with the random load disturbance caused by large-scale grid-connected new energy.The two-area microgrid model incorporating a large amount of new energy and the Guangdong power grid model considering carbon emissions are simulated.The results show that,compared with other algorithms,the proposed algorithm has stronger dynamic learning ability and robustness,and can reduce carbon emissions while improving CPS qualification rate and new energy utilization efficiency.