基于强化学习的园区综合能源系统多主体运行优化

Multi-agent Operation Optimization of District Integrated Energy System Based on Reinforcement Learning

园区综合能源系统中多主体利益诉求不同,需借助多主体博弈理论进行能源的优化调度,然而多主体互动机制复杂,常用的粒子群优化算法在求解该过程时耗时较长。为解决上述问题,本文提出了一种基于强化学习的园区综合能源系统多主体运行优化方法。首先,构建了含能源供应商、园区服务商和用户的多主体园区综合能源系统模型;其次,提出了一种基于强化信号的博弈搜索法,以提升多主体博弈求解的速度;然后,对园区综合能源系统多主体博弈过程实行分层控制,将不同主体间的供需博弈作为上层,将各主体自身的优化运行作为下层;最后,以某园区综合能源系统为算例进行仿真,并与传统粒子群优化算法进行对比,验证了所提方法的有效性与快速性。

Since the interest demands from multiple agents in a district integrated energy system(DIES)are different,it is necessary to use the multi-agent game theory to optimize the energy scheduling.However,due to the complexity in the multi-agent interaction mechanism,the commonly used particle swarm optimization(PSO)algorithm takes a long time in finding the solution.To solve this problem,a multi-agent operation optimization method for DIES based on reinforcement learning is proposed in this paper.First,a multi-agent DIES model including energy suppliers,district service providers and users is constructed.Second,a game search method based on reinforcement signal is proposed to improve the speed when solving the multi-agent game.Third,the multi-agent game process of the DIES is controlled hierarchically,i.e.,the supply and demand game between different agents is taken as the upper layer and the optimal operation of each agent itself as the lower layer.Finally,one DIES is taken as an example to perform simulations,and the comparison with the traditional PSO algorithm verifies the effectiveness and rapidity of the proposed method.