Research On SOC Estimation Of Power Lithium Battery Based On Improved LSTM Neural Network

Under the general trend of new energy development,the deepening energy crisis and the growing awareness of environmental protection have led to the rise of the new energy industry.Due to the advantages of high energy density,low power consumption,long cycle life,and high-cost performance,Li-ion batteries are gradually being used in the electric vehicle sector.Accurate State of Charge(SOC)estimation is the key to safeguarding the safety hazard and extending the life of Li-ion batteries.In this paper,we take ternary Li-ion batteries as the experimental object and carry out research based on deep learning and model fusion methods for their high accuracy SOC estimation requirements,as follows:(1)Deep analysis of the working mechanism and key parameter characteristics of lithium-ion batteries.To extract effective input features of neural network models and construct accurate equivalent circuit models,the internal structure and working mechanism of lithium batteries were analyzed,and key parameters of lithium batteries,including different temperature characteristics,capacity characteristics,charge-discharge characteristics at different rates,and open circuit voltage characteristics and their characterization rules,were explored.The degree of coupling with lithium-ion battery SOC was explored,indicating the direction for subsequent research.(2)Construction of a neural network-based dynamic model for batteries and exploration of hyperparameter search optimization methods.To avoid the problem of long-term dependence on traditional neural networks,a Bidirectional Long Short-Term Memory(BiLSTM)neural network dynamic model sensitive to time series information was constructed to estimate the SOC values of powered lithium batteries.To address the problems of numerous and complex hyperparameters and time efficiency of the BiLSTM model training,Bayesian Optimization(BO)is used to optimize the hyperparameters of the BiLSTM model,which improves the training effect and performance of the model.(3)Research on the noise correction strategy of neural network models incorporating the Kalman filter.To address the problem of poor stability of BO-BiLSTM model estimation,a second-order Davinan equivalent model is constructed,and the equivalent model is accurately identified online by the rectangular window recursive least squares method.On this basis,the traceless Kalman filtering(UKF)algorithm is chosen for the study,and the UKF filtering is incorporated into the network model for optimization to build the BO-BiLSTM-UKF fusion algorithm model.Given the influence of the system time-varying noise and other factors on the data sampling process,the coupling relationship between different temperatures and SOC is considered,and a filtering adaptive strategy is adopted to smooth and reduce the noise of the model,which enhances the stability and accuracy of the network.(4)Experimental results of the SOC prediction model for power Li-ion batteries under complex operating conditions are profiled.To verify the performance of the proposed dynamic prediction model and the fusion of filtering algorithms,the SOC estimation effect is experimentally verified by three complex test conditions under different temperature environments,and the model accuracy is verified in a phased simulation of the changing conditions.The results show that the BO-BiLSTM-UKF model can obtain high-accuracy SOC estimation under different temperatures and complex operating conditions,and the best performance is obtained at room temperature(25°C).The optimized model greatly improves the battery SOC prediction performance and verifies the high generalization and robustness of the proposed method.This paper constructs a network dynamic prediction fusion method based on an equivalent model through an in-depth analysis of the functional relationship between the operating characteristics and SOC of lithium-ion batteries and adopts an adaptive algorithm to smooth and reduce the noise of the model.Experimental validation is carried out through complex operating conditions at different temperatures and simulated operating conditions in stages.The results show that the optimized algorithm can effectively improve the accuracy and robustness of the SOC estimation of lithium-ion batteries,further enhancing the management of data from electric vehicle-powered lithium-ion batteries under different environmental temperatures.