The Research On Remaining Useful Life Prediction And Faults Diagnosis Of Power Lithium-ion Battery

Lithium-ion batteries have been widely applied on the electric vehicles and smart girds due to the advantages of high energy density and long cycle life.The accurate remaining useful life(RUL)prediction and fault diagnosis are of great significance to the security,durability,and cost of lithium-ion battery energy storage systems.However,the irreversible aging of lithium-ion batteries can lead to the continuous degradation of available capacity and deterioration of electrical performance.Thus,the state monitoring and security control of lithium-ion battery systems are still severely challengeable.In order to ensure the safe efficient operation of the battery pack,this paper embarks on the battery safety management study from two aspects of battery status assessment and system fault diagnosis based on data-driven methods.(1)Three common equivalent circuit models(ECMs)are established and the degradation mechanisms are investigated in detail based on the basic operation principle of lithium-ion battery.Moreover,two types lithium-ion batteries are experimented on the repetitive charge and discharge cycle experiment and the batteries’ aging behavior is characterized via extracting features from the aging experimental data.(2)The long short-term memory recurrent neural network(LSTM RNN)model is constructed to achieve the capacity estimation through the whole cycle life based on the aging factors that are extracted from the battery cycle life experimental data.The LSTM RNN model not only has strong adaptive learning ability and calculation efficiency,but also can effectively learn the long-term dependence in the battery degradation process.The experimental results validate that the proposed LSTM RNN model can precisely estimate the battery capacity in the whole cycle life.(3)Combining with the estimated capacity in the whole cycle life,a prediction method of RUL for the lithium-ion battery based on Box-Cox Transformation(BCT)is proposed.The BCT is exploited to construct a linear model between the transformed capacities and cycles.The battery RUL prediction is realized by extrapolating the linear model and the prediction uncertainty of the model is generated using the Monte Carlo simulation.The experimental results extrude that the presented method can exactly predict the battery RUL in the whole cycle life and at different cycle number positions,highlighting a high prediction reliability.(4)This study proposes a novel fault diagnosis and abnormality detection method for battery packs based on statistical distribution of operation data that are stored in the cloud monitoring platform.A fault diagnosis coefficient is designed to detect the variation of parameters in each state based on the classifying of the operation state and the Gaussian distribution principle.On this basis,the K-means clustering algorithm,the Z-score method and 3σ screening approach are exploited to detect and locate the abnormal cells.Experimental results validate that the presented method can accurately diagnose battery system faults and monitor the status of battery pack,which highlights that the proposed method has great research prospects and practical value.