A Novel Multi-Mode Adaptive Energy Consumption Minimization Strategy for P1-P2 Hybrid Electric Vehicle Architectures

The presented study aims to propose a new method of driving behavior recognition using a Long Short-Term Memory Recurrent Neural Network (LSTM RNN) in combination with an Energy Consumption Minimization Strategy (ECMS), resulting in a Multi-Mode Adaptive Energy Consumption Minimization Strategy (A-ECMS) for a P1-P2 series parallel Hybrid Electric Vehicle (HEV). Novelty is achieved by focusing on efficient driving mode switching instead of single mode optimization. Therefore, offline optimization was performed over different driving situations to gather different calibrations, which will be utilized in the hybrid propulsion system master controller with the purpose of determining the most fuel-efficient driving mode based on the current driving behavior. A LSTM RNN is used to classify the current driving behavior online based on vehicle speed, acceleration and distance per stop. This paper compares the effect of the proposed method in different driving conditions in order to investigate the benefits and applicability of such a control strategy. Simulations were performed representing a conventional engine-driven vehicle and a hybrid electric vehicle equipped with a P1-P2 series-parallel hybrid propulsion systems. Improvement in fuel consumption against a conventional vehicle of around 52% in average over all driving cycles can be achieved through this approach.