MTSBV Algorithm Design for Real-time Battery Data Segmentation

With the increasing number of new energy vehicles, the safety of power batteries has gained significant attention. A prerequisite for accurate estimation of battery state is effective segmentation of battery data. In response to this issue, this paper develops a frequency-domain feature-based real-time segmentation methodology for battery data analysis, specifically proposing a Multi-Threshold Spectral Band Variance (MTSBV) approach. To verify the effectiveness and robustness of the approach in real-world driving applications, we used two cycles of data, the Federal Urban Driving Program (FUDS) and US06, from publicly available battery datasets at the University of Maryland. In addition to analyzing the performance of the algorithm under ideal noise-free conditions, we also evaluated its performance when the signal is corrupted by additive white Gaussian noise (AWGN), salt pepper noise (SPN), and colored noise (CN). The performance of the algorithm was analyzed from two perspectives: the algorithm’s performance under different Signal-to-Noise Ratios (SNRs) and its comparison with other commonly used segmentation algorithms. Experimental results show that MTSBV algorithm has a good effect on solving the problem of battery data segmentation under low SNR conditions. Under FUDS and UD06 conditions, when SNR=15dB, Root Mean Square Error (RMSE) is reduced by 32.1% and 21.8% compared with other optimal conditions, respectively.