Fault Diagnosis And Fault Tolerant Control Of Proton Exchange Membrane Fuel Cells

Proton exchange membrane fuel cell(PEMFC)is an environmentally friendly power generation device using proton conducting polymer as electrolyte membrane,which has broad application prospect in transportation power and new power system.It has a broad application prospect in transportation and distributed power generation.However,the high failure rate,short life and other limitations lead to its commercialization process is affected,how to timely and accurate fault diagnosis of PEMFC in the process of operation and take the corresponding fault tolerant control measures,is a very necessary research topic.At the same time,with the development of artificial intelligence technology,data driven device health management research is getting more and more attention.Therefore,from the perspective of data-driven,this paper studies fault diagnosis and fault-tolerant control of PEMFC.The main research contents include:(1)Considering the spatio-temporal characteristics of PEMFC system operation data set,a Stacking ensemble learning fault diagnosis method is proposed.Objective weight method(Critic)is adopted to reflect the operating state of PEMFC voltage,current,temperature and pressure variables for weight,and then extract the spatial and temporal characteristics of variables and feature fusion,the construction of spatio-temporal feature set.A Stacking ensemble learning framework with CNN,RF,KNN and XGBoost as the base classifier and XGBoost as the metadata classifier was established to diagnose PEMFC system in four operating states: normal,flooded,membrane dry and hydrogen leakage.The calculation results show that the average diagnosis accuracy of the Stacking model is improved by 2.83%,and the diagnosis accuracy of the Stacking model is 99.99%,and the calculation time of CNN fault diagnosis is reduced by 28 s.Therefore,the proposed method can realize fast and accurate fault diagnosis of PEMFC system.(2)Based on GA-LSTM,the data driven models of normal,flooded,membrane dry and hydrogen leakage of PEMFC system were established.Taking Mean Square Error(MSE)of LSTM model as the objective function of Genetic Algorithm(GA),the hyperparameter combination of LSTM hidden layer node number,minimum training sample number,iteration times and learning rate was optimized,and the LSTM prediction model of PEMFC system was established.The simulation results show that the RMSE index values of GA-LSTM model in training set and test set are 0.0489 and 0.0558 respectively,which can achieve better fitting effect and avoid the increase of model complexity.(3)A fault-tolerant control scheme for PEMFC is proposed.The prediction model based on LSTM neural network and the optimization controller based on BP neural network were established,and the model predictive control scheme with the input gas pressure of anode and cathode of PEMFC system as the control quantity and the stack voltage as the output quantity was constructed.The simulation results show that the PEMFC system with fault-tolerant control link can ensure the output voltage of PEMFC system to meet the requirements under the conditions of flooding,membrane drying and hydrogen leakage,and obtain faster response speed and better control effect.