Intelligent Decision-making Method For Dynamic Energy Management Of A Microgrid

Energy management of a microgrid is an important way to utilize renewable energy sources effectively.The current dynamic energy management methods have difficulties in effectively dealing with the discreteness,nonlinearity and stochasticity of the microgrid.To this end,this paper introduces deep learning and Markov decision processes into microgrid dynamic energy management,and proposes a new set of intelligent decision-making methods for dynamic energy management of a microgrid.The main work is as follows:(1)This paper proposes a dynamic energy management method for a microgrid based on Markov decision process.The microgrid dynamic energy management model is transformed into a Markov decision model.By calculating the return value of the currently feasible discrete decision,the dynamic decision-making process of each stage is given the farsighted effect.By decoupling the discrete nonlinear optimization problem to make decisions,the approximate optimal value is obtained.The prediction scenarios of stochastic variables are generated by Monte Carlo method to fully consider the stochasticity of the microgrid.In order to reduce the time to calculate the return value,a deep neural network is used to simulate the strategy evaluation process of microgrid dynamic energy management,which can significantly improve the calculation efficiency with little loss in calculation accuracy.(2)A microgrid dynamic energy management method based on model predictive control theory is proposed to make full use of the real-time prediction information.A deep long short-term memory neural network prediction model is designed,and the long-term and short-term historical information is utilized to predict stochastic variables in subsequent periods,which can significantly improve the prediction accuracy.Under the framework of model predictive control theory,deep neural networks are used to make predictions and decisions at the same time,forming an efficient intelligent decision-making method for dynamic energy management of a microgrid,which can better deal with the strong stochasticity of the microgrid.(3)From different perspectives,simulation is performed to analyze the decision effec-tiveness and the prediction accuracy of a deep neural network and the far-sighted effect of the Markov decision process.And the proposed methods are compared with greedy strategy and traditional model predictive control algorithm.The results presented illustrate the effectiveness of the proposed method.