Machine Learning Based Design of Optimal Energy Management Strategy for Hydrogen-Fueled Hybrid Vehicle Powertrain

Abstract

<div class="section abstract"><div class="htmlview paragraph">The topic of decarbonisation involves improvements of hybrid vehicles powertrains design, from fuel type, powertrain components sizing and configuration up to control strategies. To reduce the emission of pollutants due to the combustion of traditional fuels, manufacturers are moving towards the use of “green fuels”, such as green hydrogen. In this context, the series hybrid vehicles demonstrate excellent potential: they can be equipped with hydrogen-fuelled combustion engines as range extenders, which can operate at optimal conditions without suffering from extreme transient manoeuvres. A suitable design of the control strategy of vehicle powertrain is mandatory to optimally manage the power split between range extender and battery, considering features and operating limits of both components according to power constraints.</div><div class="htmlview paragraph">This paper proposes an Energy Management Strategy (EMS), derived from an optimal approach suitable for online applications, which accounts for the key points mentioned above. The analyses are carried out on a L6e class quadricycle vehicle, whose powertrain is powered by a battery and a hydrogen-fuelled internal combustion engine as range extender. The internal combustion engine considered for this work features hydrogen lean charge operations, which may not guarantee the requested torque response under fast transients.</div><div class="htmlview paragraph">For the purpose of the study a modelling framework has been exploited to represent the main powertrain components. The proposed optimal EMS algorithm is based on Pontryagin’s Minimum Principle (PMP) method whose results are applied in simulated environment based on a vehicle dynamics model and designed following a backward approach. The simulation results are then exploited to design a Neural Network (NN) based EMS, suitable for onboard control applications. The NN training is accomplished by using the PMP-based EMS results, which provide information linking the control variable to the battery State-of-Charge (SOC) and road load.</div></div>