Thermal Management of Air-Cooled PEMFC: Machine Learning-Based Warm Starting of Active Set Methods in Model Predictive Control

This paper proposes a method that speeds up the Model Predictive Control (MPC) algorithm in the thermal management system of air-cooled Proton Exchange Membrane Fuel Cell (PEMFC), with an integration of machine learning and Active Set Method (ASM) of quadratic programming. Firstly, the parameters of the electrochemical model and mass transfer model of PEMFC are identified by swarm intelligence algorithms such as particle swarm algorithm and bat algorithm, and a semi-empirical model that can simulate actual dynamics is established. Based on this, a model predictive controller based on Active Set Method (ASM) is designed, and the optimization solution algorithm is optimized to solve the problem of slow and poor real-time performance. Combined with machine learning methods such as K-nearest neighbor algorithm and support vector machine, the warm start of the optimization solution algorithm is realized to improve the solution efficiency. The results show that using the warm-start MPC algorithm, the average number of iterations required for each optimization step can be reduced to 1/2~1/3 of the number of iterations required for cold start, indicating that the warm-start MPC algorithm combined with Machine Learning can effectively improve the solution efficiency and control performance of the air-cooled PEMFC thermal management system.