Neural Network-Based Online Energy Management for Multi-Mode Power Split Hybrid Vehicles

Abstract

Hybrid electric vehicles (HEVs) are equipped with a traditional internal combustion engine (ICE) and one or more electrical motors (EMs). HEV multi-mode power-split powertrain architecture improves fuel consumption, battery life, and vehicle emissions. However, this architecture is known for its control complexity due to the involvement of several modes of operation. Global optimal control strategies are commonly utilized as a benchmark in HEVs however they cannot be implemented on the electronic control unit (ECU) due to their extensive computational load. In this paper, a neural network (NN) -based energy management system (EMS) is proposed to control the mode and the power split of an HEV. Firstly, dynamic programming (DP), a global optimal control strategy, is utilized to achieve optimal fuel consumption using drive cycles at a wide range of conditions. Then, the proposed NN-based EMS is trained and tested using the data collected offline from the DP. The results show that the proposed NN-based EMS is able to predict the mode and power split of an HEV with only 2% higher than the optimal fuel consumption obtained by the DP.