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

Vijay Babu Pamshetti, Wei Zhang ∗, King Jet Tseng, Bor Kiat Ng, Qingyu Yan · 2025

Read Optimal Signal Decomposition-based Multi-Stage Learning for Battery Health Estimation by Vijay Babu Pamshetti, Wei Zhang ∗, King Jet Tseng, Bor Kiat Ng, Qingyu Yan (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-16377.txt.

Introduction

Pamshetti and co-authors present in 2025 a method to gauge battery health with very low error. Sound health checks keep batteries safe and cut cost, but aging is uneven and lost charge can partly return. Their method first splits the battery signal into bands by pitch like layers.

A tuned split gives clean layers that each hold a slice of the aging pattern. Learning then runs in stages to read both space and time traits in those layers. On a shared aging dataset, mean error was just a fraction of one percent. The method beat rivals with and without signal splits, and it fits real battery control systems.

Catalog record

CREATORVijay Babu Pamshetti, Wei Zhang ∗, King Jet Tseng, Bor Kiat Ng, Qingyu Yan
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE30 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-16377

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Optimal Signal Decomposition-based Multi-Stage Learning for Battery Health Estimation Thanks: This work was supported in part by A*STAR under its MTC Programmatic (Award M23L9b0052), SIT’s Ignition Grant (STEM) (Grant ID: IG (S) 2/2023 – 792), MTC Individual Research Grants (IRG) (Award M23M6c0113), and the National Research Foundation, Singapore and Infocomm Media Development Authority under its Future Communications Research & Development Programme (Grant FCP-SIT-TG-2022-007). Vijay Babu Pamshetti, Wei Zhang ∗, King Jet Tseng, Bor Kiat Ng, Qingyu Yan 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 o ptimal s ignal decomposition-based multi-stage machine l earning 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. I Introduction Batteries have revolutionized modern technology with applications spanning portable electronics, electric vehicles, energy storage systems, etc. Li-ion battery is one of the most popular battery types thanks to its rapid charging capabilities, low self-discharge rates, and longevity. The demand for Li-ion batteries is expected to reach 4.7 TWh by 2030 with an annual growth of over 30% [ 1 ] . Despite the surging demand, batteries’ complex electrochemical systems present significant operational challenges, and battery health shall be well monitored and predicted to ensure battery safety, reduce costs, and extend lifespan [ 2 , 3 ] . A common battery health indicator is the state of health (SoH). It measures the ratio of the battery’s remaining capacity to its rated capacity, e.g., when the battery is brand-new. Research efforts have been made to estimate and predict SoH accurately, yet challenges remain. Most recent research works in this domain are machine learning (ML) based. For example, Zhang et al. [ 2 ] introduce a self-attention graph pooling convolutional network (SAGPCN), which optimizes network structure with node characteristics and graph topology for SoH estimation. Jiang et al. [ 4 ] utilize bidirectional long short-term memory (LSTM) networks enhanced by attention mechanisms for complex temporal feature extraction. Ansari et al. [ 5 ] and Ma et al. [ 6 ] propose SoH neural networks with hyper-parameters optimized by meta-heuristics. A common feature of these existing solutions is that battery measurements are processed as time-series data which is often the default data format for battery instrumentation sensors. However, time-series analysis alone faces significant challenges in handling complex battery data characteristics, particularly capacity regeneration and measurement noises. Recently, researchers have proposed to complement time-dimension battery data with frequency-relevant signals. Given the time-based battery signals, existing signal decomposition methods can decompose such signals into several signals for different frequency bands. One of the early methods is empirical mode decomposition (EMD). Liu et al. [ 7 ] apply EMD to decompose battery signals and process the decomposed signals with LSTM for battery capacity estimation. EMD and its variants face limitations with complex nonlinear sequences, often resulting in suboptimal decomposition. Thus, variational mode decomposition (VMD) [ 8 ] has emerged as a new and superior alternative to EMD in battery health estimation with its enhanced resilience to noise and robust mathematical foundation. Ding et al. [ 9 ] and Chen et al. [ 10 ] integrate VMD with gated recurrent units and Transformer, respectively. VMD’s effectiveness in signal decomposition relies on a few parameters, which are often optimized with meta-heuristics, e.g., cuckoo search algorithm (CSA) [ 9 ] . Despite these advances, challenges persist in optimizing the integration of signal decomposition and ML. It seems challenging to capture the battery dynamics complexity sufficiently with a single ML. In this paper, we propose OSL, an o ptimal s ignal decomposition-based multi-stage l earning for estimating battery SoH. The optimity of OSL lies in its usage of VMD with parameters optimized by particle swarm optimization (PSO) [ 11 ] , as well as two-stage MLs with convolutional neural network (CNN) and LSTM. Given the collected battery signals, our PSO customizes the optimal parameters of VMD for the battery. VMD with its battery-specific optimal parameters decomposes the battery signals into several intrinsic mode functions (IMFs), each corresponding to a frequency band. We expect the high-frequency band to capture battery noises and rapid variations and the low-frequency band to represent trends such as long-term capacity decline. IMFs together offer a comprehensive understanding of the battery behavior and improve the effectiveness of the following learning process. CNN as the first stage of learning aims to extract spatial features from the IMFs and LSTM in the second stage handles temporal information. Finally, OSL leverages VMD’s effective signal decomposition and CNN-LSTM’s in-depth signal processing to estimate battery SoH accurately. Specifically, we have the following main contributions in this paper. • We proposed OSL which integrates signal decomposition and multi-stage ML for optimal battery health estimation. • We optimized OSL and observed that PSO converges to the optimal fast for VMD parameter optimization. • We conducted an experimental study and demonstrated that OSL achieves just 0.26% mean errors and performs the best compared to existing algorithms. The rest of the paper is organized as follows. We present OSL for battery health estimation in Section II . In Section III , we conduct an experimental study and present results and discussions. Finally, we conclude this paper in Section IV . II OSL Methodology We present the methodology and technical details of our proposed OSL for battery health estimation in this section. II-A OSL Input and Output An illustration of the system architecture of OSL is shown in Fig. 1 . OSL aims to correlate battery signals with battery health. Let f(\cdot) be the input battery signals. Battery dynamics in practice can be measured for several aspects, e.g., capacity, current, voltage, and temperature. In this paper, we aim to understand the strengths and limitations of signal processing and ML for batteries, and we prefer a simple model with capacity as the only input to facilitate the understanding. Thus, f(\cdot) in this paper is the battery capacity degradation curve in the time dimension. OSL processes and analyzes the signal f(\cdot) up to time slot t and estimates the battery health at time t . OSL can be easily configured to forecast battery health in the future and this can be discussed in the extended version of this paper. Specifically, let f(t-T^{\text{in}}),\ldots,f(t-1) be the input of OSL at time t where T^{\text{in}} is the length of the input window. The output of OSL is the estimated SoH \hat{y}^{\text{SoH}}(t) for time t and let y^{\text{SoH}}(t) be the ground-truth. OSL aims to approximate such a ground-truth with \hat{y}^{\text{SoH}}(t) as close as possible. Fig. 1: An illustration of the proposed OSL.

Read the complete text