Optimising computational efficiency in dynamic modelling of proton exchange membrane fuel cell power systems using NARX network

Abstract

Despite emerging as a green solution for power systems across various fields, fuel cell systems still face challenges that hinder their adoption due to difficulties in accurately characterising subsystems and complex phenomena, as well as the lack of effective computational models. This work utilises advanced AI technology to develop a fuel cell power system dynamic model with significantly enhanced computational speed. Three key milestones are achieved. First, a mechanistic/semi-empirical fuel cell model is established based on parameters with direct physical meaning. This model effectively illustrates the internal mechanisms of the fuel cell, providing deeper insights into its operation. Second, a complete dynamic model of a fuel cell power system is developed, comprising all necessary components and being capable of independently powering an external load or interacting with other systems. Third, by employing a Nonlinear Autoregressive model with External Input (NARX), a metamodel of the fuel cell system is created, achieving significantly improved computational efficiency while retaining essential knowledge of key phenomena. When comparing the simulation results of the NARX metamodel with those from the original mathematical model, the coefficient of determination (R²) exceeds 0.98 in post-startup conditions. Moreover, the computational speed increases at least 90-fold. The resulting metamodel demonstrates substantial potential for resolving the existential obstacles in fuel cell modelling, helping to foster the adoption of the system in real-world decarbonisation.