基于深度期望Q网络算法的微电网能量管理策略

随着光伏发电在微电网中的渗透率不断提高,其发电出力的不确定性和时变性为微电网的经济运行带来了挑战.在构建经济调度模型时,就需要适当模拟不确定变量并相应地发展高效求解算法.在此背景下,文中提出能够有效计及不确定性因素的深度强化学习算法,以实时求解微电网的优化运行问题.首先,采用马尔可夫决策过程对微电网优化运行问题进行建模...

Energy Management Strategy for Microgrid Based on Deep Expected Q Network Algorithm

With the increasing penetration of photovoltaic units in microgrids,the uncertainty and time-variability of its power generation have brought challenges to the economic operation of microgrids.When constructing an economic dispatch model,it is necessary to properly simulate the uncertain variables and develop the efficient solving algorithms accordingly.In this context,this paper proposes a deep reinforcement learning algorithm that can effectively account for uncertain factors to solve the problem of optimal operation of microgrids in real time.Firstly,the Markov decision process(MDP)is used to model the optimization operation of microgrids,and the real-time reward function is used to replace the objective functions and constraint conditions,and the interaction with the environment is used to find the optimal strategy.Secondly,the uncertain learning environment is modeled with the help of Bayesian neural networks,and then the stochastic process of state transfer is effectively considered in the MDP.Therefore,a double depth expected Q network algorithm is proposed.By considering the randomness of the state transfer,the Q iteration rules of the general deep Q network algorithm are optimized to significantly improve the convergence speed of the algorithm.Finally,a case is used to verify the effectiveness of the proposed model and algorithm.