Research On Remaining Useful Life Prediction Of Lithium Batteries Based On Data-driven Approach

As human society pays more and more attention to energy conservation and environmental protection,new energy plays a more important role in life and industrial structure.With the advantages of long life,strong stability and high specific energy,lithium battery occupies an important position in electric vehicles and other fields,and has become the focus of new energy application research.Battery performance is closely related to battery life,and performance degradation is easy to reduce the practical value of battery,and even lead to safety accidents.Based on the idea of data drive and the method of multi-model fusion,this paper constructs the remaining useful life prediction model of CNN-LSTM-TPA lithium battery,which can effectively predict the RUL of the battery,so as to ensure the safety and reliability of the battery.First of all,this paper introduces the degradation mechanism of lithium battery life decline,and analyzes the battery decay process based on NASA battery data set.Secondly,in order to solve the problem that it is difficult to obtain the capacity characterizing the battery life in real time and difficult to apply to the RUL prediction,the characteristic parameters which can characterize the battery life are extracted from the battery charge and discharge voltage,and the availability is evaluated based on the Pearson correlation coefficient between the parameters and the battery capacity.The parameters with high availability are selected and the characteristic parameters are preprocessed to indirectly predict the RUL of lithium battery.Finally,in order to solve the problem of low prediction accuracy caused by lithium battery capacity regeneration,measurement errors and model performance limitations,this paper adds convolution neural network based on long-term and short-term memory network to improve the network’s ability to extract features from input data,introduces time pattern attention mechanism,adjusts the model’s attention to different data,and adopts the strategy of capacity prediction to RUL prediction to construct a complete RUL prediction process.Finally,the RUL prediction of lithium battery based on CNN-LSTM-TPA model is realized.Several groups of comparative experiments were carried out in the NASA battery open data set,and the proposed model was compared with other RUL prediction models,and the performance was evaluated by using parameters such as MAE,RMSE and R~2.The experimental results show that the RUL prediction model of lithium battery based on CNN-LSTM-TPA can effectively improve the accuracy and stability of RUL prediction,and can effectively solve the problem of large prediction deviation caused by battery data fluctuation,and has strong generalization ability in the whole life cycle of new battery.The average value of R~2 of CNN-LSTM-TPA algorithm reaches 0.986 in battery capacity prediction task and 0.981 in battery RUL prediction task.The degree of fit is significantly improved compared with SVR,CNN and LSTM models,and MAE and other evaluation indicators are also better than other models,which can more accurately predict lithium battery RUL.The research results of this paper can provide theoretical support for the formulation of lithium battery management strategy to some extent.