A battery management strategy in microgrid for personalized customer requirements

Different microgrid (MG) users possess different preferences, which effectively stimulates the energy management needs of MG. Based upon different preferences, this paper proposes a battery management strategy in MG for personalized customer requirements, aiming to lower the cost of electricity and stabilize the state of charge (SOC) of energy-storage devices. Firstly, the Long Short Term Memory (LSTM) algorithm is used to predict the user's load power and generation power of renewable energy, followed by the construction of a corresponding reward function for different preferences of users. Secondly, according to the prediction, the Deep Deterministic Policy Gradient (DDPG) algorithm is used to carry out massive and intensive trainings on the decision model. Furthermore, the L2 regularization correction is introduced to avoid over-fitting during neural network training. Then, based on the predicted power data for the forecast day, an optimal control strategy of battery that meets the needs of differently preferred users is obtained. Finally, comparative experiments have been conducted in four scenarios for economic user, conservative user, compromised user, and user who did not adopt a decision-making framework. The results show that the battery management strategy proposed in this paper can meet the needs of differently preferred users, among whom, the profit of compromised user using the framework has increased by 55.19% and the fluctuation range of SOC have decreased by 67.05% compared with user who did not adopt the decision-making framework. (C) 2019 Elsevier Ltd. All rights reserved.