Research On Remaining Useful Life Prediction Of Lithium-Ion Batteries Based On Improved UNet And Hybrid Attention Mechanism Method

Lithium-ion batteries are one of the most advanced batteries available today,and their high energy density and high efficiency make them an indispensable source of energy in industrial production and consumer electronics,and other fields.However,non-linear capacity degradation is an inevitable problems during the usage of lithium-ion batteries.This problem may affect the stability and service life of lithium-ion batteries,thus negatively affecting the normal operation of equipment.Therefore,it is crucial to detect the lithium-ion battery reaching its service life in time and to replace the battery in advance.Aiming at the remaining useful life(RUL)prediction of ithium-ion batteries,this paper proposes a data-driven RUL prediction method for lithium-ion batteries,and the main contributions of the paper are as follows:Firstly,this paper proposes a lithium-ion battery feature extraction method based on improved UNet(IUNet).The current lithium-ion battery feature extraction suffers from the problem of information loss when mapping high-dimensional feature space to low-dimensional feature space.which leads to lower RUL prediction accuracy.The UNet network can effectively compress data and extract features,so it becomes a feasible lithium-ion batteries feature extraction method.For the uneven training of model parameters in the UNet training process,this paper proposes a layer-based constrained loss function IUNet method.Finally,the lUNet feature extraction method is combined with the Support Vector Regression(SVR)algorithm to achieve an accurate prediction of the RUL of lithium-ion batteries.Secondly,a RUL prediction method for lithium-ion batteries based on hybrid attention(HA)mechanism is proposed.The hybrid attention method can enhance the model’s attention to key information while retaining the global information,thus improving the discriminative nature of the model in the feature space.Specifically,the method establishes semantic relationships between different batteries’ observed features and RUL prediction results and between different time-step features and RUL prediction results.By weighting the features with end-to-end training,the model can quantify the contribution of different features to the RUL prediction,and thus obtain accurate RUL prediction results.Finally,the effectiveness of the method in this paper is demonstrated by comparing multiple sets of experiments.In order to verify the effectiveness of the IUNet feature extraction method and the hybrid attention RUL prediction method proposed in this paper,feature extraction and lithium-ion batteries RUL prediction experiments are conducted on the lithiumion batteries dataset provided by NASA in this paper.The experiments verify the effectiveness of the feature extraction method and lithium-ion batteries RUL prediction method proposed in this paper.This paper extracts the key features of lithium-ion batteries observation data based on IUNet,and then establishes the semantic relationship between lithium-ion batteries features and RUL prediction by the hybrid attention mechanism.The lithium-ion batteries RUL prediction results will provide important information for battery management system,so the research content of this paper will provide key technical support for lithium-ion batteries application,and has important research significance and value in improving the safety of lithium-ion battery.