Load profile choices greatly influence voltage, current, and temperature behavior during battery simulation. Until now, there has been no framework to derive on-demand representative load profiles for simulations at any system level.

We propose a methodology to synthesize such profiles. High-dimensional vehicle history information is compressed into significantly shorter, multivariate time series data. During the synthetization of profiles, signal consistency, sequence, and representativeness are ensured.

Existing synthetization techniques are adapted and extended. Micro-trip (mTr) representations serve as basic construction units, that are concatenated under the regime of a global cost function matching the base population’s signal probability distributions. The choice of each subsequent construction unit is constrained by a probabilistic sequence model. A common approach is to run unsupervised classification algorithms on abstract representations of each mTr. Class label probabilities are subsequently used to define a Markow process.

Finally, representativeness is defined on two levels: statistical distribution and simulative match between synthesis and base population. This requires performant metrics and complex simulation models that cover all relevant dimensions. Existing simulations, such as the previously published HV-battery optimization tool (Epp et al., 2022, https://doi.org/10.1016/j.est.2022.104854) optimize High-Voltage Batteries (HVB) on multiple dimensions, while optimal trade-offs need to be found among the component dimensioning. Representative profiles are thus utilized to test system robustness and to support the decision-making process.

Given our proposed framework, Pareto-optimal battery design configurations, covering the requirements of all relevant usage variants, are enabled. To overcome the disadvantages of classical approaches, we focus our future research on the utilization of sophisticated Deep Neural Network architectures.