Optimal Signal Decomposition-based Multi-Stage Learning for Battery Health Estimation

TL;DR

This work addresses the challenge of accurately estimating battery state of health (SoH) amid nonlinear aging and capacity regeneration. It introduces OSL, which combines a particle swarm optimization (PSO) tuned variational mode decomposition (VMD) to extract frequency-band intrinsic mode functions (IMFs) with a two-stage learning pipeline (CNN for spatial features followed by LSTM for temporal dynamics) to predict SoH from capacity-degradation signals. On the NASA Ames battery aging dataset, OSL achieves a mean absolute error of about $0.26\%$ and consistently outperforms baselines, including non-decomposed models and other decomposition-based methods, demonstrating the value of optimized signal decomposition and spatiotemporal learning. The results suggest OSL's practical potential for real-world battery management systems, offering improved monitoring and optimization of battery health, with avenues for future enhancement in decomposition strategies and forecasting horizons.

Abstract

Battery health estimation is fundamental to ensure battery safety and reduce cost. However, achieving accurate estimation has been challenging due to the batteries' complex nonlinear aging patterns and capacity regeneration phenomena. In this paper, we propose OSL, an optimal signal decomposition-based multi-stage machine learning for battery health estimation. OSL treats battery signals optimally. It uses optimized variational mode decomposition to extract decomposed signals capturing different frequency bands of the original battery signals. It also incorporates a multi-stage learning process to analyze both spatial and temporal battery features effectively. An experimental study is conducted with a public battery aging dataset. OSL demonstrates exceptional performance with a mean error of just 0.26%. It significantly outperforms comparison algorithms, both those without and those with suboptimal signal decomposition and analysis. OSL considers practical battery challenges and can be integrated into real-world battery management systems, offering a good impact on battery monitoring and optimization.

Figures (4)

• Figure 1: An illustration of the proposed OSL. The time-based battery signals are decomposed into the signals for different frequency bands by VMD. The decomposed signals are processed by CNN followed by LSTM for effective battery data analysis and accurate battery health estimation.
• Figure 2: Capacity degradation curves of the four batteries in the NASA dataset. Capacity decreases in general but is not monotonic.
• Figure 3: The convergence of the fitness value based on envelop entropy with different algorithms for battery B0005. All algorithms converge to the same optimal fitness value and PSO converges the fastest.
• Figure 4: Signal decomposition with VMD in Fig. \ref{['fig:imf-vmd']} and EMD in Fig. \ref{['fig:imf-emd']} for battery B0005 in Ah. VMD decomposes the original capacity curve into three IMFs for different frequency bands. EMD captures the generic aging trend with a residual curve and three IMFs without explicit frequency control.