Hierarchical reinforcement learning based operational optimization for compressed air system

Abstract

The high proportion energy consumption of compressed air system (CAS) have drawn more and more attentions, and its energy conservation and emission reduction is of great significance to achieve the green transformation in an industrial park. In this study, a novel operational optimization method for CAS by hierarchical reinforcement learning (HRL) strategy is proposed, which involves two modules in terms of the operational optimization by combining the selectable of start–stop devices and operational mode set, the control optimization by adaptive parameter controller. Moreover, the close interaction between operational optimization and tracking control by using the hierarchical structure with Q-learning process. Besides, a feedback correction strategy is also proposed for reducing the negative effect of prediction output error and compressor physical limitation. Furthermore, a series of typical cases in an industrial park are selected for verifying the effectiveness of the proposed HRL, which shows that the proposed method achieves the best trade-off between complexity and time consuming.