Supervised Machine Learning Approach to Predict the Optimal Equivalence Factor for Predictive Energy Management Strategies of Plug-In Hybrid Electric Vehicles

Nikolai Kimmig, Jan Philipp Schlomann, Daniel Goerke, Stefan Schmiedler, Bernhard Geringer, Peter Hofmann · 2025 · 3 citationsRead the paper

<div class="section abstract"><div class="htmlview paragraph">Achieving minimal fuel consumption in map-based energy management strategies or equivalent consumption minimization strategies (ECMS) for Plug-in Hybrid Electric Vehicles (PHEVs) requires prior knowledge of the optimal equivalence factor (EF). This factor, which weights the fuel consumption of the internal combustion engine (ICE) and electric energy consumption, can be calculated if the exact driving profile is known. However, in real-world scenarios, the exact driving profile and consequently the optimal EF is unknown. This uncertainty motivates the use of predictive information to estimate this factor, aiming to enable fuel optimal control in real-world driving.</div><div class="htmlview paragraph">This paper presents a methodology to predict the optimal EF across various initial battery states of energy and real-world driving profiles using a regression model for a given powertrain configuration. Initially, the optimal EF is determined, and a range of possible input features based on driving profiles are calculated and evaluated through correlation studies. To further assess the importance of these input features, a wrapper-type feature selection is conducted. For this purpose, commonly used supervised machine learning algorithms are used, such as decision trees, support vector machines, neural networks, and Gaussian processes. The study identifies the necessary features and the most suitable machine learning algorithm, followed by a scenario-based sensitivity analysis to understand the impact of incorrect input data, and thus evaluate the robustness of the prediction.</div><div class="htmlview paragraph">The findings provide essential predictive information to forecast the optimal EF for a given powertrain configuration, considering data availability, quality, and granularity. Additionally, the study addresses limitations in prediction accuracy due to incorrect or missing data and proposes suitable handling methods. Thus, this research lays the basis for a predictive energy management strategy for PHEVs, utilizing supervised machine learning to predict the optimal EF.</div></div>

2 ideas Seedlabs derived from this research