An Operating Condition Adjustment Method for Power Grid Using Multi-DRL-Agent Architecture

Abstract

As the planning and dispatching of new types of power systems become increasingly complicated， dynamic and dubious， it is very challenging to lay down operating mode for power grid that satisfies both security and economic constraints. This process typically requires intensive manual tuning and simulation calculations to search for feasible solutions in a high-dimensional action space that fulfills the requirements on safety and economy of the grid in ground state and fault conditions. Therefore， an approach based on multi-DRL-agent architecture is proposed to describe the problem as a Markov decision process （MDP）. The approach automatically adjusts different types of controllable grid resources by training centralized and distributed multi-DRL-agent to control the power of transmission lines， thus complying with various safety indicators of grid operation. The method is proved to be effective in a power grid model.

Keywords

• artificial intelligence
• grid dispatch and control
• deep reinforcement learning
• multi-agent