Research On Lithium Battery Life Prediction Algorithm Based On Model And Data-Driven Fusion

With the gradual transformation of the global energy development trend to a green direction,the new energy vehicle industry has gradually become an important part of my country’s strategic development.Lithium-ion batteries are widely used in new energy vehicles due to their advantages of long life,large specific energy,stability and safety.As important indicators reflecting the safety of lithium-ion batteries,State of Health（SOH）and Remain Useful Life（RUL）need to be effectively and accurately predicted.This thesis combines model-based and data-driven approaches to study the prediction of available capacity and remaining life of Li-ion batteries under different usage conditions.（1）A prediction algorithm based on time series Long Short-Term Memory（LSTM）neural network is proposed,which can be used to predict the current available capacity when there is historical available capacity data of lithium-ion batteries.Based on this algorithm,the influence of different timestep inputs on the capacity prediction results is studied,and a multi-dimensional evaluation index is established to measure the pros and cons of the results,so as to obtain the best prediction time step.Using the Whale Optimization Algorithm（WOA）to optimize the LSTM neural network,the results show that the optimized design can greatly improve the prediction effect.（2）For the situation that the prediction method using the available capacity input is relatively simple and difficult to obtain in actual working conditions,a reasonable and accessible health factor is proposed to predict the SOH.The correlation analysis method was used to calculate the correlation.Based on the prediction results of the WOALSTM algorithm,the constant current charging time and the gradient discharge time were selected as the two best health factors.The results showed that the prediction error was within 1%.Aiming at the problem that the WOA optimization algorithm is easy to fall into the local minimum point,three improvement points are designed: nonlinear weight factor,differential variation disturbance term and adaptive adjustment search strategy.The overall prediction result of the improved algorithm is slightly improved,which proves that the Improved Whale Optimization Algorithm（IWOA）has stronger global search ability.（3）In view of the instability in the actual working conditions of lithium-ion batteries,which makes it difficult to determine the health factor,or the incomplete historical data records lead to discontinuous time series input,research a method that does not rely on historical data and can instantly predict SOH and RUL regression algorithm.The equivalent circuit model,the IWOA and the RF（Random Forest,RF）algorithm were combined to achieve the SOH and RUL prediction of lithium-ion batteries through the fitting parameters of the equivalent circuit components,and the prediction errors were within 1% and 4% respectively.The prediction of RUL using SOH is also realized based on the IWOA-RF algorithm.The results show that the prediction effect is good,and the average error is within 3%.