Accurate state of charge (SOC) estimation is of great significance to promote the development of new energy vehicles. And a battery model’s accuracy is of great significance for the battery SOC estimation accuracy. To this end, a method based on back propagation neural network-Ant Lion Optimizer (BPNN-ALO) and unscented Kalman filter (UKF) is proposed. First, a 2-RC battery model is established and BPNN is used to fit the OCV-SOC corresponding relationship. Second, the BPNN and ALO are combined to complete the battery model’s parameter identification, and UKF is used to complete SOC estimation. Finally, verify the BPNN-ALO-UKF under two working conditions, and compare it with the other two methods. The test results show that the proposed method has higher SOC estimation accuracy, the minimum values of Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) are only 0.51% and 0.40%, respectively. And under different working conditions, it also has a better generalization and robustness.
Summary
Obtaining an accurate mapping relationship between the state‐of‐charge (SOC) and open‐circuit voltage (OCV) of lithium‐ion batteries at different ambient temperatures is of great significance for realizing accurate lithium‐ion battery SOC estimation considering the ambient temperature influence. However, the desired OCV‐SOC relationship is highly nonlinear, and the conventional polynomial fitting method is likely to result in relatively large fitting errors. To solve this problem, a method based on a backpropagation neural network (BPNN) to improve the OCV‐SOC fitting accuracy is proposed, and the SOC estimation of lithium‐ion batteries considering ambient temperature influence is completed. First, the relationship between the SOC and OCV of a lithium‐ion battery at different ambient temperatures is obtained by establishing a BPNN fitting model. Second, by optimizing the covariance decomposition process, a diagonalization of matrix unscented Kalman filtering (UKF) is proposed, which improves the accuracy and stability of the filtering algorithm. Then, the forgetting factor recursive least squares algorithm is combined to accomplish the online update of battery model parameters. Finally, under three working conditions, the effectiveness and robustness of the proposed method are verified. The simulation results show that the proposed method can obtain the most accurate SOC estimation results at each temperature, and the root mean square error (RMSE) and mean absolute error (MAE) under all three working conditions are less than 1.1%. Even if there is a certain error in the initial SOC, the proposed method can ensure that the RMSE and MAE of the SOC estimation results at each temperature do not exceed 1.5%.
To address the problem of the accuracy decrease of state of charge estimation caused by sudden high current impact, this paper proposes a lithium-ion battery SOC estimation method based on optimized deep convolutional neural network. Firstly, the 18650 battery was tested under actual driving conditions to obtain experimental data, and the experimental data was preprocessed by moving window to fit the two dimensional convolutional neural networks. Secondly, the proposed method was trained and tested, and the model parameters were further optimized. Thirdly, the proposed method is compared with sequence-tosequence methods such as long short term memory and gated recurrent unit, and the results verify the superiority of the proposed method. This article provides a method to the battery SOC estimation, which is more conducive to practical applications.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.