Look-Ahead Power Grid Dispatch Method Based on A2C Algorithm

Abstract

With the rapid advancement of new power system construction, the uncertainty of power grid operation mode is increasing, the complexity of short-time optimization decision increases rapidly, and the type and number of dispatching objects are growing exponentially. Current power grid dispatching schedules based on physical models have some problems, such as slow calculation speed, long time consumption and insufficient adaptability to cope with multiple uncertain scenes. In this study, we propose to use model-free deep reinforcement learning method to carry out research on look-ahead dispatching of power grids. Firstly, we describe the look-ahead dispatching model and establish a look-ahead economic dispatching model of power grids considering operational safety and operational efficiency, then a neural network is used to parametrically represent the policy of the power grid and the A2C algorithm is used to learn the parameterized policy. The proposed method is validated by using the IEEE 30 bus system with wind farms as an example.