Causal Discovery for Rechargeable Battery Modeling Considering Group-Level DAG Constraints

Rechargeable battery has attracted great research popularity in recent years, and numerous data-driven solutions have been proposed for its modeling. While most research works in this domain focus on the statistical correlations among extracted variables, understanding their causal relationships is equally important to provide useful insights, thereby improving the operation in real-world scenarios. In order to reach this goal, causal discovery has been utilized, which learns a directed acyclic graph (DAG) structure, referred to as causal graph, from observation data to represent the causal relationships among variables with specific physical interpretations. However, existing approaches only consider the DAG characteristics between individual variables. It may easily lead to group-level causal conflicts if multiple variables are created from the same source factor, which is a common practice for modeling the degradation of rechargeable batteries. In order to tackle this challenge, a novel group-level DAG constraint is proposed in this paper as a continuous regularization term that discourages cycles among blocks as well as individual elements in the weighted adjacent matrix corresponding to the causal graph. It is efficiently solved using an augmented Lagrangian-based method.