| Batteries serve as the fundamental component of electric vehicles,yet they are also the bottleneck that hinders the development of electric vehicles.By implementing a battery SOC(State of Charge)estimation algorithm to monitor and analyze the charge and discharge states of batteries,it is possible to accurately ascertain their real-time operation and optimize their usage efficiency.This study explores methods for SOC estimation utilizing equivalent circuit models and neural network models with the aim of achieving precise SOC estimation and facilitating more efficient and judicious use of batteries.The main focus of my research is as follows:(1)The chemical principles of lithium-ion batteries are introduced and an experimental platform is established to conduct charging and discharging tests.An experimental plan is designed to calibrate the actual battery capacity and obtain performance data.(2)The second-order RC equivalent circuit model was chosen for its balance between accuracy and complexity.The battery’s state and observation equations were constructed and the relationship between open circuit voltage(OCV)and state of charge was determined by fitting experimental data.To address the issue of data saturation and biases caused by colored noise in recursive least squares(RLS)method,bias compensated RLS can be used to identify model parameters under dynamic working condition.By analyzing the terminal voltage error of the battery model,it is proved that the BCRLS algorithm has good identification accuracy.(3)The gray wolf algorithm was improved using an enhanced attenuation factor,dynamic weight,and adaptive position update strategy to create the improved gray wolf algorithm(IGWO).In the later stage of particle filtering,due to resampling,a large number of particles are concentrated in a certain area,so the IGWO algorithm is introduced to optimize the distribution of particle sets.The state of charge(SOC)was estimated using both the particle filter(PF)and IGWO-PF algorithms based on battery charge and discharge data under dynamic conditions.Results showed that the IGWO-PF algorithm had higher accuracy for SOC estimation and maintained good performance under different initial SOC errors and noise interference.(4)This thesis introduces a bidirectional learning strategy using a Bidirectional LSTM(Bi LSTM)network to address the lack of correlation between state of charge(SOC)and future battery data in LSTM networks.Using Genetic Algorithm(GA)to optimize the number of hidden layers and the number of neurons in Bi LSTM,and build a Bi LSTM network according to the results.After verification,the prediction result of this network is better than that of pure LSTM network,and it can estimate SOC well under various dynamic conditions.To verify the applicability of the algorithm to different batteries,we used the CACLE public dataset to conduct SOC estimation studies at different temperatures.The results show that the GABi LSTM algorithm can be applied to different types of batteries and has good predictive performance at different temperatures. |
