Research On Parameter Identification Of Proton Exchange Membrane Fuel Cell Based On LMBP And EL

In the era of fossil fuel shortage and global warming,countries are gradually transitioning their energy structures from non-renewable,inefficient,and high-polluting fossil fuels to renewable,efficient,and low-carbon clean energy.Hydrogen energy,as a secondary energy source with the advantages of wide availability,cleanliness,high heat value,etc.,can help to scale up the consumption of renewable energy and facilitate energy transfer across time and space.It can also promote the balance of energy supply and demand in China,and contribute to the low-carbon development of industries such as industry,construction,and transportation.Hydrogen fuel cells are the principal manifestation of hydrogen energy development and utilization,which are influential carriers for achieving ‘green and low-carbon’ transformation at the energy terminal.Among them,proton exchange membrane fuel cell(PEMFC)has emerged as the mainstream of hydrogen fuel cell promotion and application with several strengths due to its advantages of no noise pollution,sustainability,high conversion efficiency,low carbon,and environmental protection,which has been commonly applied in the fields of aviation,electric vehicles,and distributed power generation,etc.As an emerging green energy technology,it is crucial to establish an accurate PEMFC model for its subsequent research on maximum power point tracking,optimal control,behavior prediction,as well as performance evaluation.Nevertheless,various unknown parameters contained in the model have a considerable impact on the accuracy and reliability of PEMFC.In addition,the highly nonlinear and strongly coupled of PEMFC also seriously hinder to obtain satisfactory parameter identification results by conventional approaches.Hence,this paper focuses on the parameter identification strategy in the modeling process to obtain accurate parameters for precise PEMFC modeling.Based on PEMFC thermodynamics,electrode reaction kinetics along with relevant empirical formulas,an electrochemical steady-state mechanism model is constructed,which contains seven unknown parameters to be identified.For the sake of precise and reliable parameter identification under various operating conditions,Levenberg-Marquardt backpropagation(LMBP)algorithm based on artificial neural network(ANN)is proposed.ANN based on PEMFC(ANN-PEMFC)is devised according to the structure and characteristics of backpropagation network,where the parameters to be identified are set as network weights and biases.Additionally,the neuron weights and biases are optimized by LMBP to complete the parameter identification with excellent accuracy and stability.Meanwhile,the optimization results are comprehensively compared with other five metaheuristic algorithms(Mh As),which indicate that LMBP can exactly simulate polarization characteristic curves under various operating conditions,so as to remarkably enhance the precision and reliability of parameter identification.LMBP is able to accurately identify parameters when sufficient measurement datasets are available,but insufficient or missing measurement datasets may lead to poor parameter identification in actual engineering.Based on this,this paper uses the extreme learning machine(ELM)strategy for data prediction to achieve an appropriate expansion of the dataset,thereby significantly improving the feasibility of subsequent optimization algorithms and the accuracy of optimization results.ELM is combined with five meta-heuristics and LMBP to compare the parameter identification results of each optimization algorithm without prediction and based on ELM prediction.Simulation results demonstrate that ELM can effectively reduce the impact of insufficient or missing datasets,which can guarantee high quality and stability of parameter identification for subsequent optimization algorithms.