Research On Fault Diagnosis Of Battery Pack Based On Deep Learning

Along with the environmental pollution and energy crisis in the world day by day,the problem of energy and environment protection has been attached more and more importance by the world.It is the common understanding of all mankind to push for a green and low carbon transition.It is also the duty of China as a responsible nation.In order to promote energy transition,the development of electric transport has become a strategic option,and the electric car will have the opportunity to develop.This has also promoted the fast development of the electric battery system.Battery pack is the main source of power forelectric vehicles.The operation of a battery pack has a direct impact on the performance indicators of EVs and the safety and stability of driving.In the case of failure of the battery,not only will the service life of the battery be affected,but also the vehicle performance and driving experience will be decreased,and the safety of the vehicle will be seriously endangered.Therefore,it is necessary to solve the problem of fault diagnosis of battery pack.Due to the complexity of the non-linearizing system,the conventional approach is prone to disturbance due to uncertainty.In order to decrease the impact of failure on the battery pack,this thesis uses the Deep Learning Method,which is important to improve the performance of the battery pack.Taking lithium battery as the research object,this thesis proposes a battery pack fault diagnosis method based on Transformer architecture by analyzing the related technologies of battery pack fault diagnosis and applying deep learning.Aiming at the micro-short circuit and overcharge-discharge faults of lithium battery cells,the design uses the Gram angle field to preprocess the battery data,and converts it from time series to images,and uses the convolutional autoencoder for feature extraction.Input the Transformer deep learning network,use each feature of the battery as an attention head,use the attention mechanism to calculate the degree of correlation,and use the obtained results for battery fault diagnosis.The accuracy rate of this method for judging the micro-short circuit and overcharge-discharge faults of lithium battery cells is 96.59%.Aiming at the SOC inconsistency fault of lithium battery packs,a denoising autoencoder is designed to extract features from the preprocessed battery data,and the obtained results are input into the Transformer network for SOC estimation,and the inconsistency faults of the battery packs are judged.At the same time,the model parameters are compared experimentally,and the optimal parameters are obtained.The mean absolute error is 0.0072.Compared with other fault classification methods based on deep learning algorithms,the experimental results show that the method in this thesis has better real-time performance and accuracy.The excellent performance of the Transformer-based deep learning model in the fault diagnosis of lithium-ion battery packs is verified.Finally,by analyzing the requirements of the battery pack fault diagnosis system,the overall design principles,architecture and process of the system are determined.Based on this,the hardware for battery data collection and management is designed,and the Alibaba Cloud platform is used to build a battery pack fault diagnosis system.The web management platform,together to form a battery test platform,by simulating the occurrence of battery failures and obtaining data of lithium batteries under different conditions,realizes the battery pack fault diagnosis analysis and related processing operations based on Transformer deep learning,and verifies the feasibility of the method.to ensure the safe operation of the battery pack.