A health state assessment method for new energy vehicles based on an ensemble belief rule base with dual-adaptive feature selection

New energy vehicles (NEVs) are a key technology for achieving the low-carbon transition of the transportation sector. The operational safety and energy efficiency of NEVs are critical for the sustainable development of transportation systems. However, the complex system structure and highly variable operating conditions make health state assessment of NEVs challenging. The belief rule base (BRB) framework combines knowledge-driven and data-driven information to handle uncertainty and has been widely used in health assessment of complex systems. In multi-attribute multi-fault scenarios, BRB models still suffer from combinatorial rule explosion and cognitive bias from a single expert, which lead to increased model complexity and reduced interpretability. To address these issues, this study develops an ensemble BRB method with dual-adaptive feature selection, termed Dual-Adaptive Ensemble BRB (DA-EnBRB). The method employs a dual-adaptive mechanism based on Extreme Gradient Boosting (XGBoost) and mutual information (MI) to dynamically construct compact multi-attribute feature subspaces and to build a BRB expert model in each subspace. A multi-strategy crowned porcupine optimizer (MSCPO) optimizes the parameters of all BRB experts. Class-specific expert weights are then learned and used to fuse sub-model outputs via weighted voting, yielding the final diagnostic result. Experiments on NEV fault diagnosis and NEV energy efficiency assessment demonstrate the effectiveness and generalization capability of DA-EnBRB.