Data-Enhanced Battery Simulator for Testing Electric Powertrains

Battery simulations are an essential tool for the efficient development of electric powertrains, which require accurate models and reliable hardware. Surprisingly, today’s massively collected measurements are not yet used for realistic and virtual development environments. Among other reasons, handling the large and heterogeneous datasets of automotive batteries still prevents a consequent application. Hence, a data-enhanced electric model of the battery is presented, which relies on a recurrent neural network to compensate the error of a physically-motivated model. Such a hybrid model introduces the high accuracy of machine learning to simulations. Ultimately, it allows a replacement of real batteries with model-based simulators at test benches. The approach is validated based on a comparison of a real battery with a simulator and its different model variants. It is shown, that the data-driven enhancement of existing simulations increases the accuracy while maintaining the robustness of the original model.