Controle preditivo não linear para o gerenciamento de energia de um veículo elétrico híbrido a célula a combustível

Abstract

This work proposes an energy management system (EMS) for a fuel cell hybrid electric vehicle (FCHEV). This system is based on nonlinear model predictive control (NMPC) and employs a recurrent neural network (RNN) to model a proton exchange membrane (PEM) fuel cell (FC). With NMPC, it is possible to formulate control objectives that would not be possible with linear model predictive control (MPC), such as maximum e ciency point tracking (MEPT) of the FC. In addition, compared to traditional electrochemical models, the RNN can predict the FC's nonlinear dynamics with better accuracy. The EMS was implemented on a low-cost single board computer, and the experiments for controller validation were performed on a hardware-in-the-loop (HiL) test bench equipped with a real 3 kW FC stack. The experimental results demonstrate that the proposed EMS can meet the vehicle's energy demand, where it performs the battery's charge sustaining battery, and can operate the FC in its most e cient region. In addition, a comparative study was also carried out between the proposed NMPC, a linear MPC and a hysteresis band control. The results of this comparative study demonstrate that the NMPC provides a better fuel economy and can reduce the FC degradation.