Research On SOC Estimation Of On-board Battery Operation And Maintenance Management System For Rail Transit

In recent years,the scale of China’s rail transit network has been expanding,and a large number of rail transit vehicles have been put into operation,which makes the issue of the vehicle safety to be highly regarded.In order to ensure the reliability of power supply,the rail transit vehicles are equipped with the backup power systems,which take the Valve-Regulated Lead-Acid batteries(VRLA)as the core.Under emergency conditions,the on-board VRLA batteries become the final power barrier for the important facilities of the vehicles,and the operation status of the batteries is directly related to the vehicle operation safety.Therefore,it is of great significance to improve the operation and maintenance level of the backup batteries with the help of battery management system(BMS).State of Charge(SOC)is a key monitoring parameter of BMS.Accurate estimation of SOC can effectively reduce the damage to the battery caused by over-charge and over-discharge.In this paper,considering the operating characteristics of the VRLA batteries for rail transit vehicles,the SOC estimation problem was deeply studied.A battery SOC estimation model based on Gated Recurrent Unit(GRU)was designed,and the accuracy of the model was improved by using data augmentation and attention mechanism.Moreover,the model was deployed on the hardware platform with STM32.The specific works are as follows:Firstly,this paper describes the working principle and characteristic parameters of the on-board VRLA batteries,and introduces the architecture of the on-board battery operation and maintenance management system for rail transit.The important role of the edge embedded device integrating SOC estimation function in the system is explained.Secondly,a VRLA battery experiment platform was built,and the VRLA battery data was collected by experiments.A GRU-based SOC estimation model was developed in Python environment.The experiments showed that the root mean square error(RMSE)and mean absolute error(MAE)of the GRU-based SOC estimation model were within 2.5%.In addition,the comparison experiment proved that the GRU-based SOC estimation model has advantages in terms of estimation accuracy and training efficiency.Then,the accuracy of the SOC estimation model was improved by data augmentation and attention mechanism.After using Gaussian noise for battery data augmentation,the average RMSE of the SOC estimation results decreased from 1.78%to 0.87%,and the average MSE decreased from 1.54% to 0.69% under different constant power discharge conditions.The GRU-Attention model reduced the average RMSE of the SOC estimation results from 1.78% to 1.32%,and reduced the average MAE from 1.54% to 1.09% under different constant power discharge conditions.Finally,the main control circuit,the sampling circuit of voltage and current,as well as the communication circuit of the edge embedded device were designed.The GRU and GRU-Attention were deployed and tested on the device,respectively.The results showed that the RMSE and MAE of the battery SOC estimation on the device were within 4%,which had a good accuracy.