Research On Life Prediction Of Supercapacitors Based On Improved Long And Short-term Memory Neural Network

Due to the rapid consumption of fossil fuels and environmental problems caused by large amounts of greenhouse gases,people are eagerly seeking green alternative energy sources and energy storage equipment.The development of high-performance energy storage equipment is crucial to the development of environmental friendly society and renewable energy.Because of the high power density,supercapacitors have great potential for development in the energy field and are the core devices in energy storage systems.Therefore,the health status of supercapacitors seriously affects the safe operation of the entire energy storage system,which has aroused great concern.Long short-term memory neural network optimized based on hybrid genetic algorithm is proposed to predict the remaining useful life of supercapacitors.First,determine the input variables of the deep neural network.By analyzing the structural characteristics between the electrode and the solution,and the characteristics and laws of the movement of internal charges,the energy storage mechanism of the supercapacitor is studied,and the influencing factors of the supercapacitor performance aging are obtained.The input variables of the neural network model are determined according to the influencing factors of the performance aging of the supercapacitor,in order to achieve accurate prediction of the remaining useful life of the supercapacitor.Secondly,two tests are designed to obtain supercapacitor aging data.A large amount of data is the basis of high precision for life prediction,to fully reflect the aging condition of the power supply under different running environment,this paper uses different charge and discharge strategies based on different temperatures and voltages to perform a steady-state cycle life test and Hybrid Pulse Power Characteristic test on the supercapacitors to make the measured data more real and effective.After determining the input variables of the neural network and obtaining a large amount of experimental data,a prediction model of the long short-term memory neural network optimized based on the hybrid genetic algorithm is constructed.The model uses a hybrid genetic algorithm to automatically find the optimal dropout probability of the neural network and the number of hidden layer units.The hybrid genetic algorithm uses the root mean square error of remaining useful life of the supercapacitor between which predicted by the long short-term memory neural network and the groundtruth as its fitness function.During the optimization process,the genetic algorithm converges to the vicinity of the local optimal solution,the sequential quadratic programming algorithm is utilized for further local search,the dropout probability and the number of hidden layer units are optimized quickly and accurately,and the obtained optimal parameters are input into the long short-term memory neural network to predict the remaining useful life of the supercapacitors.After the deep learning model is designed,the cycle life prediction experiment is conducted on the aging data of trained and untrained supercapacitors,and compared with other prediction models,the results prove that the prediction model proposed in this paper has higher prediction accuracy and strong generalization ability.In addition,a life prediction experiment of supercapacitors under dynamic test conditions was conducted to further confirm the versatility of the prediction model.