Representative battery load profile synthesis leveraging multi-objective optimization heuristics

Abstract

Automotive high-voltage batteries show distinct reactions depending on their concurrent states of demanded power, temperature and $SoC$. To aid development, representative load profiles are frequently derived. Besides velocity-based cycles, literature also proposes the generation of electrical power trajectories. However, current methods fail to represent simultaneous thermo-electrical usage dynamics. Moreover, fitness functions based on highly aggregated parameters do not account for complex battery dynamics. This work presents a methodology to synthesize coupled $P$, $T$, and $SoC$ trajectories. First, MCMC simulation derives an optimal $SoC$ discharge stroke chain. Next, multiple stroke realizations are obtained by concatenating sequentially constrained micro-trips. A genetic algorithm then discovers feasible solutions to the related combinatorial optimization problem. Representativity is measured using the earth mover’s distance between signal distributions. Final profiles are selected from a Pareto front, allowing for the prioritization of Markov- or signal-related fitness. We conclude that applying the scale reduction factor with a threshold of $\hat{R}\le 1.01$ yields suitable length estimations of $SoC$ stroke chains. The general introduction of an optimization step enables mean fitness improvement of up to $40\text{\%}$ compared to sole MCMC sampling. 1D and 2D error function designs yield similar average fitness, while the latter demonstrates to deliver a broader solution variety. Our framework serves as a versatile base for individual battery applications.