Explaining Deep Reinforcement Learning-Based Methods for Control of Building HVAC Systems.

Deep reinforcement learning (DRL) has emerged as a powerful tool for controlling complex systems, by combining deep neural networks with reinforcement learning techniques. However, due to the black-box nature of these algorithms, the resulting control policies can be difficult to understand from a human perspective. This limitation is particularly relevant in real-world scenarios, where an understanding of the controller is required for reliability and safety reasons. In this paper we investigate the application of DRL methods for controlling the heating, ventilation and air-conditioning (HVAC) system of a building, and we propose an Explainable Artificial Intelligence (XAI) approach to provide interpretability to these models. This is accomplished by combining different XAI methods including surrogate models, Shapley values, and counterfactual examples. We show the results of the DRL-based controller in terms of energy consumption and thermal comfort and provide insights and explainability to the underlying control strategy using this XAI layer.