Lithium-ion Battery State Of Health And Remaining Useful Life Prediction Based On Artificial Intelligence

The global issues of environmental pollution and energy depletion have made the development of sustainable energy a global consensus.As a key focus area of China’s development strategy,new energy vehicles(EVs)have attracted lots of attention.Lithium-ion batteries(LIBs)are widely used as the power sources in EVs due to their merits of high specific energy,long life,environmental friendliness and no memory effect,etc...However,the nonlinear degradation of LIBs severely limits their efficiency and lifespan,and even threatens the operation safety.As two critical indicators of LIBs safety,the accurate estimation of the State of Health(SOH)and prediction of Remaining Useful Life(RUL)is essential for their efficient production,safe operation,and secondary recycling.This thesis has studied two models to perform the battery state prediction based on data-driven methods,and explored the estimation accuracy of the models about batteries under different discharge strategies and temperatures,which provides strongly practical support for online prediction of batteries.The main contents of this thesis including the following results:(a)Batteries data under different discharge strategies and ambient temperatures from the National Aeronautics and Space Administration(NASA)dataset were preprocessed;furthermore,the health indicators(HIs)for battery aging were determined by considering Pearson correlation coefficient(PCCs)and types of parameters.The multi-dimensional evaluation metrics were then established to measure prediction accuracy and model performance.(b)The time-series based Long Short-Term Memory(LSTM)neural network was proposed to predict the battery SOH or available capacity at first.The effects of different training-set proportions and timestep inputs on model performance were studied separately.By using this algorithm,the influence of different discharge strategies and experimental temperatures on the battery SOH prediction was investigated.The results confirm that model accuracy can be improved with increasing training-set proportion.However,the optimal timestep input varies with battery cycling conditions,and the model exhibits a poor estimation performance for batteries at extremely low temperatures.(c)A two-layer ensemble prediction model(Stacking Regression,SR)that combines the benefits of multiple machine learning methods was established.Based on the same HIs,SOH and RUL of the batteries were predicted by SR model with different training set proportions,and the effects of discharge strategies and temperatures on the prediction accuracy of SR model were also investigated.The superiority of the SR model and its stability at low temperatures is confirmed by comparing the prediction performance of the SR model with that of other machine learning methods.An online prediction with dynamic forward progression using the SR model was proposed and compared with the LSTM model,the corresponding benefits,drawbacks,and application scenarios were also investigated.