Deep learning-supported multi-objective power system scheduling in an energy internet optimal scheduling environment

Abstract

In the context of energy internet, new energy access makes the power system present multi-objective coupling, difficult security constraints, etc., and the traditional model-driven methods have limitations in scheduling efficiency and adaptability. Based on this, this paper proposes a PPO-DDPG model that combines deep reinforcement learning and proximal policy optimization algorithm. The model acquires the dynamic timing data of the power system with deep reinforcement learning and designs constraints on its states, actions, rewards and so on. The PPO algorithm is then introduced to update the network parameters as a way to improve the dynamic scheduling effect on multi-objective power systems. Experiments show that the new energy consumption rate of the PPO-DDPG algorithm after convergence is as high as about 97.5%, which is about 30% higher than that of the PPO algorithm, and the average survival time as well as the reward value are significantly better than that of the existing methods. Therefore, relying on deep learning technology to empower multi-objective power system scheduling is more economical, ensuring that the power system can obtain optimal economic benefits when accessing different types of energy under high uncertainty.