Online Monitoring Of Lithium Battery Health State Based On A Variant Long Short Term Memory Neural Network

In recent years,in order to effectively cope with the depletion of traditional fossil fuels and the deterioration of the environment,lithium batteries have been widely used.During the operation of lithium batteries,side reactions such as lithium precipitation and electrolyte oxidation will occur.The side reaction of the battery will directly lead to the decline and attenuation of the battery performance,which is manifested in the macroscopically as a decrease in capacity and an increase in internal resistance,thereby reducing the service life of the battery.Therefore,accurately predicting the health state of the lithium-ion battery and its remaining useful life will not only help ensure the safe and reliable operation of the system,but also maximize the comprehensive utilization of the remaining value of the battery.This thesis proposes a technical framework for online monitoring of lithium battery health based on improved long-short-term memory neural networks.Based on this framework,a hybrid neural network is designed to capture the temporal and spatial characteristics of multiple variables that affect battery degradation,and an automatic hyperparameter optimization algorithm is introduced to automatically select the hyperparameters of the network.The main work of this thesis is as follows:1.Created active-state-tracking long-short-term memory neural network for lithium battery state of health estimation and remaining useful life prediction.Active state tracking long-short-term memory neural network first determines the discarding of old information and the retention of new data at the same time through a fixed connection.Secondly,the new input data and the historical unit state are multiplied by element to filter out more useful information.Finally,the peephole connection from the cell state is added to the output gate to shield unwanted error signals.The Pearson correlation coefficient is used to analyze the correlation of the battery discharge process variables such as voltage,current,temperature,sampling time,etc.,to filter out the input of the model,and 10-fold cross-validation is used to evaluate the best network structure and hyperparameters.Realize the state of health estimation and remaining useful life prediction of the lithium battery.2.A hybrid neural network combining convolutional neural network and active-state-tracking long-short-term memory neural network is proposed.The hybrid neural network captures the temporal and spatial characteristics of multiple variables that affect battery degradation in a hierarchical manner.And actively learn the long-short-term dependence embedded in these characteristics.The Kolmogorov-Smirnov test method is used to establish the prior distribution of hyperparameters in the hybrid neural network,which serves as an agent for evaluating the influence of battery data on the modeling of the hybrid neural network.Bayesian optimizes the hyperparameters of the neural network to realize the state of health estimation and remaining useful life prediction of the lithium battery.Experimental results show that compared with other neural network methods,the online monitoring method of lithium battery health based on hybrid neural network improves the accuracy of model prediction.