Impact of Driving-Cycle Characteristics on Electric-Vehicle Battery Performance: A Bayesian-Optimization Framework

The energy consumption of electric vehicles (EVs) is significantly influenced by driving behavior, yet battery packs are often sized against uniform standard test cycles. This work presents a dual contribution: an empirical analysis of battery sizing sensitivity across transient driving conditions and a critical diagnostic evaluation of composite objective formulations in multi-parameter battery optimization. We couple a MATLAB/Simulink electro-thermal EV powertrain model with a Bayesian optimizer using a Gaussian-process surrogate to size cell capacity (Ccell), series cell count (Ns), and parallel string count (Np) across four standard MATLAB-sourced drive cycles (Urban, Highway, Mixed, and Aggressive) matched for total distance (16.3–17.8 km). Empirically, specific energy consumption increased from 89.60 Wh/km (Urban) to 166.97 Wh/km (Aggressive), an 86.3% rise driven by transient peak power demands rather than mean speed. Methodologically, we demonstrate that standard composite sizing objectives suffer from structural degeneracy: while total pack energy (NsNpCcell) is uniquely identified, the individual series–parallel topology (Ns vs. Np) remains underdetermined when SOC, thermal, and current penalty constraints remain inactive at convergence. Identifying and isolating this objective non-uniqueness provides a methodological framework for EV powertrain optimization, preventing unsupported claims of global optimality in under-constrained battery-pack designs.