Research on Frequency Stability Emergency Control Strategy Based on Deep Reinforcement Learning

In order to achieve the goal of peak carbon dioxide emissions and carbon neutrality, and build a clean, low-carbon, safe, controllable, flexible, efficient, intelligent, friendly, open and interactive new-type power system, a large number of conventional generation units have been replaced by renewable energy sources, resulting in the reduction of the inertia coefficient of the power system and the deterioration of the transient frequency stability. After the HVDC blocking of a power grid, taking the sending end power grid as an example, a large amount of active power surplus leads to the increase of frequency, which may lead to the transient frequency instability of the power grid. However, the operation modes of the power system with high proportion of renewable energy sources are complex and changeable, and the manual establishment of different emergency control strategies for massive operation modes leads to a tremendous increase in workload and low efficiency. Therefore, this paper proposes a method, which in turn establishes a fast prediction model of frequency emergency control strategies based on deep reinforcement learning (DRL). From the perspective of data-driven, aiming at the frequency instability issues of HVDC blocking, a fast prediction model of frequency emergency control strategies is established to improve the adaptability of emergency control strategy to complex and variable operation modes. Finally, based on a regional power grid model, the application of this method is explained, and its effectiveness is verified.