基于混合整数线性规划和强化学习的微电网经济优化调度  

Economic optimal scheduling of microgrids based on mixed integer linear programming and reinforcement learning

作　　者：宋潇磊 王致杰(指导)[1] 王鸿[1] SONG Xiaolei;WANG Zhijie;WANG Hong(School of Electrical Engineering,Shanghai Dianji University,Shanghai 201306,China)

机构地区：[1]上海电机学院电气学院,上海201306

出　　处：《上海电机学院学报》2023年第6期311-316,329,共7页Journal of Shanghai Dianji University

基　　金：上海市自然科学基金资助项目(15ZR1417300)。

摘　　要：针对微电网中可再生能源的不确定性和波动性,以及负荷的动态变化带来的复杂性问题,提出了一种将混合整数线性规划(MILP)和强化学习(RL)相结合的MILP-RL算法。首先,在算例分析中,采用粒子群优化(PSO)算法和MILP对微电网的负荷、风机、光伏预测数据进行经济调度策略求解;然后,将求解的结果作为RL的初始调度策略,构建了PSO-RL和MILP-RL两种组合算法;最后,为了模拟可再生能源的不确定性和波动性以及负荷的动态变化,在预测数据的基础上添加了扰动并采用PSO、RL及组合算法PSO-RL、MILP-RL对扰动数据进行求解。结果表明,相比PSO算法和RL,所提出的MILP-RL组合算法在应对可再生能源的不确定性和负荷的动态变化方面表现出更高的经济性和更快的收敛速度。To address the uncertainty and volatility of renewable energy in microgrids and solve the complexity problem caused by the dynamic changes in loads,a hybrid algorithm combining mixed integer linear programming(MILP)and reinforcement learning(RL)is proposed(MILP-RL).First,in the case analysis,the particle swarm(PSO)algorithm and MILP are used to solve the economic dispatch strategy for the load,wind turbine,and photovoltaic forecast data of the microgrid.Then,the solution results are used as the initial dispatch strategy of RL to construct two combined algorithms,i.e.PSO-RL and MILP-RL.Finally,in order to simulate the uncertainty and volatility of renewable energy and the dynamic change of loads,a perturbation is added based on the forecast data.The perturbation data are solved by PSO,RL,and the combined algorithms PSORL and MILP-RL.The results show that compared with the PSO algorithm and RL,the proposed MILP-RL combination algorithm can obtain a higher economy and faster convergence speed to deal with the uncertainty of renewable energy sources and the dynamic changes in loads.

关 键 词：微电网 混合整数线性规划 强化学习 MILP-RL组合算法 

分 类 号：TM73[电气工程—电力系统及自动化]