Research On Cell Battery Fault Diagnosis Of Electric Vehicle Based On Deep Neural Network

The failure of lithium batteries in electric vehicles hinders the vigorous development of the electric vehicle industry.The lithium battery fault diagnosis method based on the lithium battery electrochemical mechanism model is difficult to be applied to the fault conditions of electric vehicles in complex and uncertain traffic environments,resulting in insufficient universality of the fault diagnosis model and low diagnostic accuracy.In this paper,based on the voltage,current,temperature and capacity data in the process of lithium battery charge and discharge,the use of wavelet packet decomposition and deep neural network theory as a method for lithium battery overcharge failure,overdischarge failure and aging fault diagnosis research.Details as follow:(1)Aiming at the problem of non-stationary noise interference in the process of lithium battery fault data collection,a feature extraction method of lithium battery fault data based on wavelet packet decomposition is proposed.Through wavelet packet layer-by-layer decomposition,the optimal wavelet tree is selected with Shannon entropy as the cost function,the optimal wavelet tree node signal is reconstructed,data noise is removed,and feature extraction is completed.(2)Aiming at the time-dependence of lithium battery fault data,a lithium battery fault diagnosis method based on long-short term memory(LSTM)neural network is proposed.Maintain the long-term dependence of lithium battery fault data through memory cells and gating functions,under the constraint of the target loss function,the LSTM neural network learns the mapping relationship between fault data and fault types,and uses the attention mechanism at all time steps to select the relevant network hidden state to achieve accurate classification of lithium battery fault data.The experimental results show that,driven by the data features of wavelet packet decomposition,the average diagnosis accuracy of the three faults based on the fault diagnosis method of LSTM neural network is 94.13%.(3)Aiming at the characteristics of spatial non-linear distribution of lithium battery fault data,a lithium battery fault diagnosis method based on convolutional neural network(CNN)is proposed.Use continuous convolutional layers with weight sharing and sparse interaction to transfer fault data layer by layer in multiple convolutionallayers to complete the conversion of low-level features of fault data to high-level features,and use global average pooling to realize data features from high-dimensional to the low-dimensional structured output,under the supervised learning mechanism,the mapping of input parameters to fault types is realized.The experimental results show that,driven by the data features of wavelet packet decomposition,the average diagnosis accuracy of the three faults based on the fault diagnosis method of CNN is 97.09%.(4)Aiming at the multi-classification problem of lithium battery fault data under time and space constraints,a lithium battery fault diagnosis method based on long-short term memory-convolution neural network(LSTM-CNN)is proposed.By integrating the LSTM neural network and the CNN in parallel,the two network structures are used to fully analyze the input data,and the output vectors of the two networks are combined to complete the classification of lithium battery failure under the joint action of the two neural networks.The experimental results show that,driven by the data features of wavelet packet decomposition,the average diagnosis accuracy of the three faults based on the fault diagnosis method of LSTM-CNN is 99.40%.