Causal relationship discovery and causal-oriented approaches for enhanced performance and interpretability in prediction of prosumer behavior and demand flexibility

Abstract

Causal analysis paves the way for more interpretable assessment of complex systems and phenomena, such as human-in-the-loop components of energy systems. This article will pursue novel approaches for causal analysis of prosumers' behavior. The knowledge of this causality is core for multiple smart grid applications including but not limited to the design of demand side management programs, retail electricity market design, development of effective distributed energy resources aggregation strategies, and net load forecasting. The complex nature of human interactions with energy relies on many factors and understanding behavior causality is a core, unsolved challenge. This article presents a probabilistic algorithm for discovering causal relationships between the end users' consumption flexibility and its influencing factors. The obtained causal knowledge is then utilized to boost the precision of demand flexibility prediction. Two causal-oriented approaches are proposed to enhance the performance and interpretability of predictive models, incorporating causal information through causal regularization and data preprocessing. Simulation results demonstrate that the algorithm can effectively identify causal probabilities among different factors and unveil key characteristics of the prosumers' behavior. Additionally, these proposed causal-oriented approaches outperform the non-causal-oriented predictive models in terms of both performance and interpretability, highlighting the advantages of incorporating causal information.