PEMFC Imitated Water-drop Block Channel Structure Optimization And Performance Prediction Based On Machine Learning

The megatrend of low-carbon energy and industrial development is already in place.A new generation of proton exchange membrane fuel cells(PEMFC)with clean and low-carbon characteristics is being progressively developed.The channel structure affects the fluid flow,gas diffusion,and electrochemical reactions for the proton exchange membrane fuel cell,which directly limits the final output performance.Based on the above issues,an innovative PEMFC channel structure is designed with imitated water-drop block for improving the mass transfer and output performance of PEMFC.Then,a performance prediction model and a structural optimization framework based on machine learning methods are framed for enhancing the PEMFC performance calculation efficiency and structural optimization speed.Firstly,for the PEMFC channel structure,inspired by falling water drops,four different imitated water-drop(IWD)blocks are designed to investigate the effect on the fuel cell performance based on the low resistance and high speed of the streamlined shape.Results show that the designed fourth imitated water-drop block(IWD IV)channel has the highest PEMFC current density and power density.Since the design of the streamlined block and the gap between the block and the electrode ribs,the lower pumping power caused by the pressure drop in the channel effectively increases the PEMFC net power,and the IWD IV channel is the optimal design.Secondly,the block height,width,and arrangement spacing are optimized based on the IWD IV channel for obtaining the optimal imitated water-drop block channel.The effect of different structural parameters on the PEMFC performance based on the optimal imitated water-drop block is researched.Results show that with the optimal channel structure,the PEMFC has a more uniform reactant concentration distribution inside the PEMFC.The mass transfer and electrochemical reactions are effectively improved.The higher power density and net power are obtained.Using 642 sets of numerical simulation results,a multi-physical structure model database is constructed to lay the groundwork for the subsequent training of neural network prediction models.Again,for solving the problem of long computation time and high resource consumption of numerical simulation of fuel cells,an AdaBoost ensemble neural network prediction model is constructed based on machine learning methods to predict the PEMFC output performance under different block structure parameters.The multi-physics structure model database is pre-processed.The hyper-parameters of the base learners in the prediction model are fine-tuned to obtain the optimal prediction model.The prediction results show that compared with the Bagging ensemble prediction model and back-propagation(BP)neural network,the AdaBoost ensemble prediction model can predict the PEMFC polarization curve with extremely high accuracy within 1 second.Consequently,this ensemble performance prediction model can effectively improve the efficiency of PEMFC performance calculation.Finally,the ensemble prediction model is coupled with the improved gray wolf optimizer(IGWO)algorithm.A channel structure optimization framework is designed to maximize power density for the PEMFC single-objective optimization.The ensemble prediction model is utilized as a surrogate model to calculate the fitness function of IGWO.The optimization results show that the optimization framework takes less than 2 minutes to predict the optimal solution for the channel structure parameters and power density.Then,the predicted optimal solution is returned to the physical model for verification with an error of only 3.96%.Simultaneously,the optimal channel structure can effectively improve the PEMFC performance.The peak power density is enhanced by 5.423% over the straight channel.The proposed optimization framework will guide the performance calculation and channel structure multivariate optimization for PEMFC.