Research On Lithium-ion Battery Model Parameter Identification And State Of Charge Estimation Method

With the rapid growth of electric vehicles,lithium-ion power batteries are becoming increasingly popular.However,without efficient and reliable monitoring of its key working states,it can cause significant issues such as vehicle safety.The State of Charge(SOC)of lithium-ion batteries serves as the foundation for essential battery management systems(BMS)functions,such as energy optimization distribution and balanced management,as well as an important foundation for real-time analysis of battery operation.Accurate estimation of SOC is critical for increasing battery and vehicle performance and safety.This paper investigates the SOC estimation method for lithium iron phosphate batteries.The research content and major conclusions of this article are as follows:(1)The variation law of open circuit voltage(OCV)with temperature was investigated,as well as the connection between charge/discharge voltage and OCV at four different temperatures.The mean value method was used to calculate the corresponding OCV,and thus the OCV curve at 25°C was determined for the study.(2)Using a Forgetting Factor Recursive Least Squares algorithm,the parameters of the Dual Polarization(DP)equivalent circuit model are identified online,and the results show that the identified model parameters of resistance and capacitance are reasonable under Federal Urban Driving Schedule(FUDS)and Dynamic Stress Test(DST)conditions.During90% of the discharge time,the error between the predicted voltage of the model and the experimental terminal voltage was concentrated in the range of ±0.6 m V and ±0.4 m V,showing that the identification method was effective.On this premise,the Extended Kalman Filter algorithm was used to estimate the battery SOC,and the estimation resulted in a close match to the reference SOC.(3)The Unscented Kalman Filter algorithm was employed to predict the battery SOC in order to further enhance the anti-noise adaptability of SOC estimation.The Adaptive Square Root Unscented Kalman Filter(ASRUKF)SOC estimation algorithm was created by integrating the Square Root Filtering algorithm to lessen the algorithm’s reliance on prior knowledge and introducing an adaptive filtering algorithm to achieve adaptive processing and measurement noise.The SOC was estimated based on two algorithms under FUDS and DST conditions.Under FUDS and DST conditions,the SOC was calculated using two algorithms.The findings demonstrated that there was often ±1% error between the two methods and the reference SOC.The ASRUKF algorithm’s Root Mean Square Error(RMSE)under the two dynamic situations was less than 0.5%.Lastly,the two methods’ initial error correction and convergence performance were confirmed.The ASRUKF algorithm converged almost two times as quickly as the Unscented Kalman Filter technique with a 50%initial error set.The enhanced method is more accurate and greater tolerant of initial errors.(4)A Bayesian Optimization-Long and Short-term Memory neural network(BO-LSTM)technique for SOC estimation was developed using big data and machine learning frameworks.The Bayesian Optimization(BO)approach was created to increase the stability and accuracy of the LSTM model and avoid choosing hyperparameters based on empirical data.The SOC was calculated under three different temperatures and two different dynamic conditions.According to the estimation results,the minimum target value predicted during the BO iteration process was basically stable after 20 iterations.Furthermore,the RMSE of the BO-LSTM algorithm for SOC estimation was within 0.3% under both dynamic conditions.Finally,the BO-LSTM algorithm was compared to the ASRUKF algorithm,and the results show that the BO-LSTM model provides more accurate SOC estimation.Furthermore,the accuracy of the ASRUKF algorithm is influenced by the structure of the equivalent circuit model and the accuracy of model parameter identification,whereas the accuracy of the BO-LSTM algorithm is influenced by data quality and network initialization settings;when the model and model parameters are clear and the amount of data is small,the ASRUKF algorithm can be chosen to be used;when enough data are available and the accuracy requirement is high,the BO-LSTM algorithm when enough data are available and the accuracy requirement is high.