Adaptive Neuro-Fuzzy and Tree-Based Estimation of Lithium-Ion Battery State of Health for Electric-Vehicle Battery-Management Systems

Accurate online estimation of lithium-ion battery state of health (SOH) is a core function of the battery-management system (BMS) that governs range, safety and lifetime in electric vehicles. Because capacity fade is a slow, nonlinear function of operating stress, model-based estimators must capture the coupled effects of cycling, temperature and usage depth without imposing a heavy computational load on embedded hardware. This paper compares an adaptive neuro-fuzzy inference system (ANFIS) with four tree-based regressors for SOH estimation from operating-stress descriptors only: equivalent full cycles, cell temperature, depth of discharge and C-rate. A physics-informed capacity-fade model combining an Arrhenius temperature dependence with a depth-of-discharge power law generated an 1855-point data set covering 0–1600 equivalent cycles and 15–45 °C. The ANFIS model attained a test coefficient of determination of 0.9965 and a root-mean-square error of 0.49 % SOH, substantially outperforming extreme gradient boosting (R²=0.9638) and a single decision tree (R²=0.7909). Random-forest feature importance ranked cycle number highest, followed by depth of discharge and C-rate. The results indicate that a compact neuro-fuzzy estimator offers an accurate, interpretable and embeddable solution for on-board SOH monitoring.