Development of a neural network-based energy management system for a plug-in hybrid electric vehicle

The high potential of Artificial Intelligence (AI) techniques for effectively solving complex parameterization tasks also makes them extremely attractive for the design of the Energy Management Systems (EMS) of Hybrid Electric Vehicles (HEVs). In this framework, this paper aims to design an EMS through the exploitation of deep learning techniques, which allow high non-linear relationships among the data characterizing the problem to be described. In particular, the deep learning model was designed employing two different Recurrent Neural Networks (RNNs). First, a previously developed digital twin of a state-of-the-art plug-in HEV was used to generate a wide portfolio of Real Driving Emissions (RDE) compliant vehicle missions and traffic scenarios. Then, the AI models were trained off-line to achieve CO2 emissions minimization providing the optimal solutions given by a global optimization control algorithm, namely Dynamic Programming (DP). The proposed methodology has been tested on a virtual test rig and it has been proven capable of achieving significant improvements in terms of fuel economy for both charge-sustaining and charge-depleting strategies, with reductions of about 4% and 5% respectively if compared to the baseline Rule-Based (RB) strategy.