IN-24-C007 - Explainable AI Framework for Model Predictive Control in Energy Management Systems

Optimizing energy consumption in buildings stands as a pivotal concern in the face of escalating environmental challenges. To address this, the integration of Artificial Intelligence (AI) techniques, specifically Model Predictive Control (MPC), has emerged as a promising approach. However, the lack of interpretability in AI-controlled systems hinders their widespread acceptance and trust. Addressing this challenge, Explainable AI (XAI) methodologies offer insights into AI decision processes, fostering comprehension and confidence. This research focuses on bridging this gap by proposing an Explainable AI framework for MPC systems for optimizing the battery charge state for residential buildings equipped with photovoltaic generation and onsite battery storage from the NeurIPS 2022 CityLearn Challenge dataset. The AI-based MPC aims to dynamically adjust the charge state of the battery to minimize the cost and carbon emissions from predicted future energy demands and solar generation, optimizing the battery's charge level accordingly. This study investigates the reliability of AI systems in building controls, aiming to align AI decision-making with user expectations. Through the application of XAI techniques, the study confirms that the AI model comprehends system features, enabling objective improvements and enhancing user confidence. This allows the MPC to not only optimize building energy consumption but also explain the decision-making process behind AI-controlled systems. The integration of Model Predictive Control with Explainable AI paves the way for sustainable and transparent energy management practices in the built environment.