Neural network based model predictive control of voltage for a polymer electrolyte fuel cell system with constraints
Xiufei Li, Miao Yang, Yuanxin Qi, Miao Zhang
TL;DR
This work addresses the challenge of maintaining a stable PEFC output voltage under workload disturbances while respecting safety constraints on hydrogen pressure and input-rate changes. It introduces a data-driven neural network model to capture voltage and pressure dynamics, and integrates it into a model predictive control framework by linearizing the NN and solving a quadratic program with slack for infeasibility. Compared to a physics-based MPC, the NN-MPC achieves comparable voltage regulation while largely satisfying the safety constraints, though occasional minor hydrogen pressure violations occur under certain step changes due to model and linearization errors. The approach demonstrates the viability of data-driven MPC for constrained PEFC voltage control in practical, disturbance-rich environments.
Abstract
A fuel cell system must output a steady voltage as a power source in practical use. A neural network (NN) based model predictive control (MPC) approach is developed in this work to regulate the fuel cell output voltage with safety constraints. The developed NN MPC controller stabilizes the polymer electrolyte fuel cell system's output voltage by controlling the hydrogen and air flow rates at the same time. The safety constraints regarding the hydrogen pressure limit and input change rate limit are considered. The neural network model is built to describe the system voltage and hydrogen pressure behavior. Simulation results show that the NN MPC can control the voltage at the desired value while satisfying the safety constraints under workload disturbance. The NN MPC shows a comparable performance of the MPC based on the detailed underlying system physical model.