LSTM Based Small Samples Research On The Life Prediction Method Of Lithium Battery

In the era of information warfare,lithium-ion batteries are widely used in military drones and various types of advanced weapons and equipment.However,in the course of use,problems such as improper storage and use can lead to equipment failure and huge losses of financial and material resources.Therefore,it is very important to predict the remaining useful life of lithium batteries.Considering that the life cycle of lithium batteries used in military weapons is relatively long,it is difficult to obtain data on more performance degradation.Therefore,this paper needs to carry out RUL prediction analysis of lithium batteries under small sample conditions.To address the problem of low accuracy of RUL prediction results under small sample conditions,this paper improves on the basis of Long Short-Term Memory neural network,and the main research work is as follows:Firstly,the public ageing datasets of Li-ion batteries selected in this paper are NASA B05 and CACLE CS2＿35.The battery parameters in these two datasets are analyzed and feature factors are extracted,and the five indirect health factors extracted are used as inputs to the prediction model.The degree of correlation between them and the battery capacity is analyzed according to the grey correlation method to verify the validity of the indirect health factors selected in this paper.Secondly,according to the extracted indirect health factors,the curve relationship between them and the capacity of Li-ion battery is plotted to obtain the a priori knowledge,and then the monotonic a priori knowledge is added to the performance function of the LSTM neural network in the form of a mathematical expression,and the coefficients of the penalty factor are set,and then a fused constrained Li-ion battery RUL prediction model,called the MC-LSTM model,is established.The experimental simulation results show that the MC-LSTM algorithm model proposed in this paper has a higher prediction accuracy compared to the prediction effect of the standard LSTM.Thirdly,to further optimize the prediction accuracy of the MC-LSTM model compared with the actual capacity of the lithium battery,this paper improves the MCLSTM into a bi-directional constrained network Bi-MC-LSTM;because the Bi-MCLSTM neural network has more hyperparameters,and the artificial settings will have an impact on the model accuracy.Therefore,the Whale Optimization Algorithm is chosen to optimize the hyperparameters of the network,and the hyperparameters after optimization are reassigned to the network to construct the WOA-Bi-MC-LSTM prediction model;finally,after experimental verification,the model proposed in this paper is closer to the actual capacity of lithium batteries,and has higher prediction accuracy and smaller error compared with other algorithms.Finally,a set of software system is designed to evaluate the remaining life prediction of lithium batteries.The software can implement functions such as feature extraction and pre-processing of lithium battery data,and also display the algorithm model and lithium battery RUL prediction results proposed in this paper one by one,and then users can make the remaining life prediction of lithium batteries based on their own data,and the system will give corresponding suggestions according to the prediction results.