Research On Inlet Humidity Control And Remaining Useful Life Prediction Of PEMFC Based On Machine Learning

Proton exchange membrane fuel cell(PEMFC)is widely regarded as a new power generation technology with broad application prospects.At present,the main bottleneck that hinders its large-scale commercial promotion lies in the difficulty of its moisture and heat control,and its service life is not long enough.In view of this,a set of flexible and compatible PEMFC test bench is built in this thesis.The mechanism analysis and machine learning methods are used to study and discuss the PEMFC inlet humidity control method and life prediction method.The main work is as follows:(1)A PEMFC measurement and control experimental platform is built.Based on the analysis of the working principle and measurement and control requirements of PEMFC,a PEMFC measurement and control experimental platform with the architecture of "Upper machine-Signal acquisition-Equipment" is designed and built with the single PEMFC as the core.The upper machine software including human-computer interaction,data processing and other functions is developed by using Lab VIEW software and MODBUS communication protocol.The communication between the upper computer and the instrument layer is realized by using routers and data acquisition cards,and the detection and control of the single PEMFC is realized by using related instruments.(2)This thesis presents a neural network-based control method for inlet humidity of PEMFC.Aiming at the external humidification scheme of PEMFC using bubbling humidifier,a mathematical model reflecting the thermodynamic relationship between inlet humidity and bubbling humidifier temperature is constructed,so that the complex inlet humidity control is simplified to a relatively simple temperature control.According to the non-linear and time-varying characteristics of the bubbling humidifier,an adaptive controller combining neural network and PID control law is proposed in this thesis.The controller can adaptively adjust the control parameters during operation.The comparative experimental results show that the comprehensive performance of the controller is better than that of fuzzy PID and classical PID controllers.(3)A life prediction model of PEMFC based on attention mechanism and neural network is proposed.In this study,the aging characterization index of PEMFC is studied by using the correlation analysis method.According to the common working characteristics of PEMFC,a deep learning model based on attention mechanism is proposed.Then,in order to verify the effectiveness of the model,this investigation compares the model with five other mainstream machine learning algorithms based on a large PEMFC open data set.The results show that the proposed model has the highest accuracy.Finally,based on a constant working condition,the remaining life of PEMFC is predicted,and the validity of the prediction results is discussed.55 Figures,11 Tables,93 References...