Physics-aligned causal inference of driving styles on electric vehicles battery aging

Abstract

Abstract Lithium-ion batteries (LIBs) remain the core service life bottleneck of electric vehicles (EVs), and driving style is a critical human factor dominating EV battery degradation as assessed via aging proxy indicators and semi-empirical model predictions. Existing studies mostly focus on correlation analysis between driving behavior and aging, suffering from unphysical black-box models and unreliable causal identification due to uncontrolled confounding factors. This paper proposes a physics-aligned causal inference framework to quantify driving styles’ impact on EV battery aging. It embeds the second-order Thevenin equivalent circuit into a generative model for physics-constrained representation, realizes confounder disentanglement via a gradient reversal layer, and quantifies the estimated causal effect through counterfactual inference. By bridging physics-based battery modeling and data-driven causal inference, this framework provides a principled approach for quantifying driving behavior’s causal impact on battery aging, with potential applications in battery life-extension strategies and personalized driver guidance. Validated on real-world EV operation data, the method achieves 0.55% MAPE in terminal voltage reconstruction, and reveals that aggressive driving is associated with over 2.5 times higher model-predicted capacity fade than eco-driving under the estimated causal association framework.