In the development of automotive high-voltage batteries, simulations and test bench measurements are essential. Batteries are subjected to time-dependent load profiles to assess their response to operational conditions. Typically, the selection of appropriate inputs is based on rough estimates of how the battery will be used, often utilizing the WLTC homologation or short, self-generated profiles. Initial approaches in literature also create custom velocity- or power-based representative drive cycles.

However, experiments indicate that batteries can be highly sensitive to the heterogeneity and sequence of different stress factor combinations. This raises concerns about whether the aforementioned approaches can fully capture battery usage. Currently, research lacks a framework for modeling multivariate time series profiles in the battery domain. Beyond the need for a general synthesis architecture, there is also a need to investigate the sensitivity between modeling assumptions and actual battery behavior.

This study applies a novel synthesis methodology that generates statistically representative battery time series across multiple signal dimensions. A two-stage clustering technique distinguishes between frequently and rarely occurring intensities of battery stress. Subsequently, multivariate battery load profiles are created and optimized. These time series are then used in simulations and test bench measurements to evaluate the correlation between statistical representativeness and induced battery behavior. The derived profiles demonstrate the ability to replicate similar battery behavior compared to their unmodified counterparts.