Degradation Adaptive Observation For Temperature Distribution Of Solid Oxide Fuel Cell Stack Based On Machine Learning

Environmental pollution and the increasing lack of non-renewable energy are bottlenecks of the sustainable development of human society.For theses reasons,countries all over the world are looking for a new,clean and renewable energy utilization method.Solid oxide fuel cell(Solid Oxide Fuel Cell,SOFC)can directly convert the chemical energy in hydrocarbons into electrical energy through electrochemical reaction at 600～900℃ and has the advantages of high efficiency and pollution-free.Therefore,it has been widely concerned and becomes a research hotspot in recent years.SOFC stacks have strict requirements on operating temperature,and the stack temperature control is the key to the safe and stable operation of SOFC.However,due to the extremely high tightness requirement of the stack,it is impossible to install the temperature measuring devices to obtain and control the stack internal temperature within an acceptable cost.Therefore,the use of soft-sensing methods to estimate the internal temperature of the stack is the key to the safe operation of SOFC.For this reason,this thesis takes the Steam Reforming Solid Oxide Fuel Cell(SR-SOFC)system as the research object,and designs a temperature observer for engineering application that can quickly and accurately estimate the stack temperature distribution.Aiming at the above research goals,this thesis mainly does the following research work:(1)Based on the physical prototype system,the mechanism model of the SR-SOFC system is built on Matlab/Simulink,and the SR-SOFC system model is comprehensively verified by the particle swarm optimization algorithm to ensure the accuracy of the model,laying a solid foundation for follow-up research.(2)The evolution analysis of the stack temperature distribution at the static space level and the dynamic time level is carried out,and the temporal and spatial characteristics of the stack temperature distribution are discovered,that is,the stack temperature distribution and the input parameters present a temporal correlation;the stack temperature distribution and the mutual heat transfer between each temperature node present a spatial correlation.(3)Subsequently,this thesis uses the multivariable linear regression algorithm to establish the temporal-spatial characteristic model of the stack temperature distribution,and uses the least square support vector machine optimized by the particle swarm algorithm to estimate the temperature of the central node,thereby constructing a temperature distribution observer based on machine learning,and then carries out dynamic and static performance evaluation of the observer.(4)In order to overcome the observer estimation error caused by the degradation of the stack,this thesis designs the degradation adaptive calibration strategy of the stack temperature distribution observer based on the stochastic gradient descent algorithm,and uses the measured temperature data to identify the degradation and calibrate the observer.The simulation results show that: in the case of no degradation of the stack,the average RMSE and MAPE of the observer’s estimation results are 0.4109 K and 0.032%,respectively;in the case of stack degradation,the average RMSE and MAPE of the observer’s estimation results after calibration are 0.4855 K,0.031%,respectively.Therefore,the stack temperature distribution observer designed in this thesis can quickly and accurately obtain the SR-SOFC stack temperature distribution,which lays a solid foundation for the safe and efficient operation of the SR-SOFC system.