Emergency Frequency Control Strategy Using Demand Response Based on Deep Reinforcement Learning

An emergency frequency control strategy using demand response based on deep reinforcement learning is proposed. According to the system response before and after the fault, the control strategy is used to predict the amount of load shedding by the trained prediction model. Firstly, genetic algorithm is used to optimize the multi-objective function of the optimal load-shedding using demand response after the fault of the target power grid. Q-learning is used to establish the action mapping relationship between the system response before and after the fault and the optimal load-shedding strategy using demand response, so as to obtain the sample data. After training sample data by convolutional neural network (CNN), a load-shedding strategy prediction model is obtained to adapt to more fault scenarios. Finally, validity of this method is verified by IEEE 39-bus system.