Intelligent Parameter Identification Of PEMFC Based On Bayesian Neural Networ

For a long time,the global economy has been developing rapidly,and fossil fuels have been the main power source of the world economic development.However,the long-term overdevelopment and use of fossil fuels have made the global crisis such as resource shortage,environmental pollution and ecological destruction increasingly emerge.As a result,more attention is being paid to the efficient use of clean and low-carbon energy sources,such as hydrogen,solar,wind and tidal power.Hydrogen is the simplest,lightest and most abundant element in the known universe,and is considered to be one of the most important energy carriers.In addition,hydrogen plays a key role in the integrated development of renewable energy applications.For example,in the fuel cell power generation technology,hydrogen can be injected into the fuel cell to provide power to the system.This electrochemical based equipment can realize the efficient and reliable conversion of chemical energy to electric energy,which fully meets the requirements of environmentally friendly energy supply.In addition,fuel cells have become the fourth generation of power generation technology after hydroelectric,thermal and nuclear power.Fuel cell technology is capable of converting chemical energy into electrical energy in an efficient and non-polluting manner,with unlimited fuel supply and almost zero pollutant generation.As a result,fuel cell power generation technology has become quite popular in various industrial applications.In addition,fuel cells can be generally classified into five categories based on different electrolyte types,namely,alkaline fuel cells(AFC),proton exchange membrane fuel cells(PEMFC),phosphoric acid fuel cells(PAFC),molten carbonate fuel cells(MCFC),and solid oxide fuel cells(SOFC).Among them,PEMFC as the most outstanding performance member of fuel cells system,not only can rely on PEMFC with high conversion efficiency and low cost as a satellite power source,but also PEMFC with high power density,small size,low operating temperature and no noise,can be used as a technical power source for urban transportation vehicles and underwater submarines.Moreover,the performance of PEMFC as a promising environment-friendly power conversion device is degraded,even leading to a serious reduction of its lifetime.Therefore,an accurate and effective PEMFC model is essential for its simulation analysis,optimal operation and control,and lifetime prediction.In recent years,various PEMFC models have been developed,which are usually divided into analytical and electrochemical models.In particular,the feasibility and efficiency of electrochemical models have been verified in many studies.Therefore,in this paper,electrochemical models are used to study the parameter extraction of PEMFC,where several unknown parameters,such as air pressure,fuel flow rate,and cell temperature,are extremely critical to ensure the accuracy of modeling.However,the inherent multivariate,multi-peak and nonlinear characteristics of the PEMFC severely increase the difficulty and complexity of its parameter estimation.In addition,the inevitable noisy data under various operating conditions usually hinders meta-heuristic algorithms from obtaining high-quality parameters for PEMFC.To address these obstacles and prevent the data overfitting phenomenon,this paper proposes a Bayesian regularized neural network based meta-heuristic algorithms(BRNN-Mh As)for PEMFC.BRNN-Mh As approach is used to filter data noise and obtain more accurate fitting curves for electrochemical models under various operating conditions.To verify more precisely the accuracy of the algorithm for enhancing model parameter estimation,the performance of the algorithm is comprehensively evaluated and analyzed in this paper through a comprehensive comparison with typical meta-heuristic algorithms under various operating conditions.Three case studies are also conducted for three experimental conditions:(a)low pressure-high temperature,(b)medium pressure-medium temperature and(c)high pressure-low temperature.The results verify that BRNN-Mh As can extract PEMFC parameters more accurately with high accuracy,speed and stability.In addition,the performance of the proposed method is comprehensively evaluated and analyzed by a comprehensive comparison with several advanced meta-heuristic algorithms under different operating conditions.