An Artificial Intelligence Driven Method for Power System Control Based on the Cloud-Edge Collaboration Architecture

The new-type power system with the high penetration of renewable energy accessed is of strong uncertainty and complexity, which can be challenging for the traditional methods to control. It's significant to introduce artificial intelligence to meet the challenge. This paper proposes a cloud-edge collaborative framework based on multi-agent deep reinforcement learning for power system regulation. Using an unsupervised clustering algorithm, the power grid is decomposed into several sub-networks according to the geographical relationship. Then, edge computing platforms are set up on each sub-network, where agents are deployed. The distributed control problem of each subnetwork can be modeled as a Markov decision model. The global observation information of the system is delivered to each edge platform through the cloud computing center, and all agents are trained to learn the best regulation strategy according to the global information. The proposed method can effectively decompose the centralized tasks and transfer them to the edge side, alleviating the pressure on the cloud center and enhancing the robustness of system operation.

I. Introduction

Under the background of carbon peaking and carbon integration, many countries have vigorously developed access to renewable energy. However, the access to a high penetration of renewable energy and a high penetration of power electronic equipment will bring randomness and volatility to the power system. California, due to the sudden drop of wind power and solar power units, there was a turn power outage accident. It can be seen that the high penetration of renewable energy grids has the characteristics of complex operating scenarios and many emergencies. The traditional centralized control methods may not work effectively. To maintain the stable operation of the power grid and promote the consumption of renewable energy, it is of great significance to study the active power optimization scheduling method for a large-scale power grid with a high penetration of renewable energy accessed.