Strategy optimization of emergency frequency control based on new load with time delay characteristics

With the development of modern communication technology and the large number of new controllable loads connected to the power grid, the new controllable loads with flexible regulation characteristics can participate in the emergency frequency stability control. However, the communication state differences and spatial distribution characteristics of controllable load will affect the actual effect of frequency control. In this paper, an emergency frequency control method based on deep reinforcement learning is proposed considering the response time of controllable load shedding. The proposed method evaluates response ability for emergency control of controlled loads through load response time, controllable load amount and controllable load buses. Then, the controllable load with smaller response time is cut out preferentially to ensure rapid control, and the Markov Decision Process (MDP) is used to model the emergency frequency control problem. Finally, Rainbow algorithm of Deep Reinforcement Learning (DRL) is used to optimize the emergency frequency stability control strategy involving controllable load resources. The formation of emergency load shedding instruction is directly driven by high-dimensional operation state data after power grid failure, so that, the aim of minimizing the economic cost is achieved under the constraint of system stability. The effectiveness of the proposed method is verified in the IEEE 39-bus system.