| With the aggravating of energy crisis and environmental problems,electric vehicles have become the focus of attention.As one of the energy storage components of mainstream electric vehicles,lithium-ion batteries still have a series of problems in use,among which,accurate estimation of the lithium-ion battery power has become one of the difficulties in the practical use of electric vehicles.As a form of external battery characteristics,the battery model plays a role not only in the management of lithium-ion batteries,but also in the estimation of lithium-ion batteries.In this paper,the most widely used ternary lithium-ion batteries are taken as the research object,and the modeling scheme and SOC estimation of lithium-ion batteries are studied.Starting with the working principle of lithium-ion batteries,this paper introduces the electrochemical reactions during their operation and the internal structure.The advantages and disadvantages of the three kinds of equivalent circuit models are analyzed,and the second-order Thevenin model with lower complexity and higher accuracy is taken as the research object of this paper finally.The internal resistance characteristics,capacity characteristics and open-circuit voltage characteristics of lithium-ion batteries are analyzed,and the OCV-SOC curve of lithium-ion batteries is obtained.It is of great significance for further research on SOC of lithium-ion batteries.In order to model the battery more effectively,the principle of RLS and its application in battery modeling are analyzed.The experiments show that RLS cannot distinguish the two RC links of the second-order Thevenin model and the battery model identified by RLS has problems with accuracy and applicability.Aiming at the problems of parameter identification in RLS,a joint modeling scheme based on the combination of model fusion and RLS is proposed.By analyzing the ensemble learning algorithm in machine learning,the model fusion method of AdaBoost algorithm in ensemble learning is used to fuse one of the RC links effectively,and then the other RC link is identified by RLS specifically.The experiments show that the model built by the scheme has high accuracy under different working conditions.In order to obtain the accurate SOC of the battery,the principle of UKF is analyzed in this paper,and UKF is applied to the estimation of battery SOC.The experiments show that UKF is too depend on the battery model and it is difficult to guarantee the performance with reduced accuracy of battery.In order to reduce the impact of battery model on SOC estimation,the data-based BP neural network is analyzed in this paper.Aiming at the problem that the two-input BP neural network has a poor training effect when the amount of data is low,the four-input BP neural network is proposed using feature engineering to optimize the training effect.The experiments show that four-input BP neural network has a higher accuracy when estimating SOC,but relying too much on data is also a problem with BP neural network.In order to overcome the problems of UKF and BP neural network under limited conditions at the same time,BP-UKF algorithm is proposed by combing UKF and BP neural network.UKF is used to estimation preliminary and then BP neural network is used to compensate in this algorithm.The experiments show that the algorithm can estimate SOC accurately under the premise that the accuracy of the battery model and training data are difficult to further optimize,which greatly improves the accuracy and adaptability of the SOC estimation algorithm.The calculation process of UKF is researched during SOC estimation,and the relationship between the SOC compensation value and the terminal voltage of the battery model is obtained.The battery is then corrected by using a more accurate SOC.The experiments show that the voltage error value of the battery model terminal after the correction using the SOC compensation value at the previous moment is reduced significantly,which is of research significance. |
PDF Full Text Request