Hybrid Deep Learning for Dynamic Total Transfer Capability Control

Abstract

This letter proposes a data-driven hybrid deep learning method for dynamic total transfer capability (TTC) control. It leverages deep learning (DL) to achieve fast prediction of TTC and reduce the problem complexity, while the deep reinforcement learning (DRL) method, e.g., proximal policy optimization (PPO), is enhanced by competitive learning (CL) to obtain a better generalization of the DRL agents. This also allows us to deal with system stochasticity. Comparison results with other model-based alternatives on the IEEE 39-bus system highlight the advantages of the proposed method for variable unseen and insecure scenarios.

Topics & Concepts