State of Power Prediction for Battery Packs Based on Particle Swarm Optimization SVM

Hui Yan, Qingbin Wang, Yun Yang, Xianzhong Zhao, Zudi Huang, Yuxi Wang, Shi Liu, Yi Bin, G. S. Huang, Jianfeng YangBatteries 2025 · DOI

TL;DR

This paper introduces a novel approach for predicting the State of Power (SOP) degradation in lithium-ion battery packs, a crucial metric for operational efficiency and safety. By combining Particle Swarm Optimization (PSO) with Support Vector Machines (SVM), the researchers developed a model that effectively captures pack-level degradation patterns and achieves high prediction accuracy, outperforming existing methods.

What Problem Does This Paper Address?

Lithium-ion batteries are ubiquitous, but their performance degrades over time, impacting system efficiency and safety. While much research has focused on individual cell health indicators like State of Health (SOH) and State of Charge (SOC), larger battery packs, which are more common in applications like electric vehicles, present a more complex challenge. The State of Power (SOP) is a critical indicator for battery packs, directly influencing performance aspects like acceleration and energy recovery. However, predicting SOP degradation in battery packs has received less attention. Existing methods often struggle with accuracy, generalization to different operating conditions, and computational efficiency. Furthermore, traditional end-of-life definitions based on fixed capacity thresholds are becoming less relevant with the rise of secondary battery applications.

Methods

The core of this study is a hybrid approach that leverages the strengths of two powerful techniques:

  • Support Vector Machine (SVM): SVM is a robust machine learning algorithm known for its strong generalization ability, particularly in pattern recognition and prediction tasks. It excels at handling complex, non-linear relationships and is less susceptible to noise and outliers. The paper utilizes SVM for its ability to model the intricate degradation patterns of battery packs.
  • Particle Swarm Optimization (PSO): PSO is an optimization algorithm inspired by the social behavior of bird flocks or fish schools. It's used here to automatically tune the critical hyperparameters of the SVM model: the kernel parameter g (determining feature width) and the penalty coefficient c (controlling tolerance for errors). By simulating the collaborative search of particles, PSO efficiently finds the optimal combination of g and c, avoiding the inefficiencies and local optima issues associated with manual tuning or grid search. This optimization significantly enhances the SVM's generalization ability and prediction accuracy.

The study also employs a sliding window approach to construct input features for the SVM. Historical power data within a defined window (50 time steps) is used to predict the value at the next time step, transforming the time-series problem into a standard supervised learning task.

Datasets

The study utilized experimental data from 32 lithium-ion cells, each with a rated capacity of 0.8 Ah.

  • Training Data: Data from eight individual cells were used to train the PSO-SVM prediction model.
  • Testing Data: Data from six battery packs were used for testing and evaluating the model's generalization ability and prediction accuracy. These packs were constructed by connecting cells in parallel: three packs with three cells each (2.4 Ah theoretical capacity) and three packs with five cells each (4.0 Ah theoretical capacity).

The experiments were conducted under baseline testing protocols, which involved standardized charge-discharge procedures in a temperature-controlled chamber to ensure consistency.

Key Insights and Ideas

  • SOP as a Superior Indicator for Pack Health: The paper strongly advocates for SOP as a more comprehensive indicator of battery pack health than traditional SOC or SOH, as it directly reflects operational performance.
  • PSO-SVM Synergy for Robust Prediction: The combination of PSO for hyperparameter optimization and SVM for modeling proves highly effective. PSO's global search capability overcomes SVM's sensitivity to parameter selection, leading to more accurate and reliable predictions.
  • Transfer Learning from Cells to Packs: The study demonstrates that knowledge gained from individual cell degradation can be effectively transferred to predict the behavior of entire battery packs, even with the complexities introduced by parallel configurations.
  • Dynamic Failure Thresholds: The authors acknowledge the limitations of fixed failure thresholds and suggest the need for multiple-threshold schemes, especially considering the growing trend of "secondary utilization" for batteries.

Key Results

The PSO-SVM model demonstrated significant improvements in SOP prediction accuracy:

  • High Precision: Most prediction errors were below 5%, with the maximum error reaching 10.91%.
  • Superior Performance: The PSO-SVM model consistently achieved the lowest prediction errors compared to physics-based models, Differential Model Decomposition (DMD), and Long Short-Term Memory (LSTM) networks. The lowest error achieved by other methods was 13.22% (LSTM), which still exceeded the PSO-SVM's maximum error.
  • Stable Performance: The model showed stable predictive performance across different aging stages and initial conditions, and across different battery pack configurations.
  • Visual Validation: Figure 3 visually confirms that the predicted degradation trajectories closely overlap with the actual trajectories for all tested battery packs. Figure 4 further illustrates the compact and central error distribution of the PSO-SVM model compared to other methods.

Limitations and Future Work

  • Controlled Environment: The experiments were conducted under controlled baseline conditions. The authors acknowledge that the model's generalization capability might be limited under more complex real-world operating conditions, such as varying temperature gradients, different C-rates, partial cycling, dynamic loads, and temperature variations.
  • Future Work: The paper suggests that future research should incorporate multi-dimensional aging indicators, including cumulative throughput, SOP parameters, and electrochemical signal-derived features, to enhance the model's ability to represent battery performance degradation under complex operating conditions and improve its generalization and reliability.

Summary updated

Abstract

Lithium-ion batteries are widely used in energy storage systems and consumer electronics; however, long-term usage leads to capacity degradation, which affects system efficiency and safety. Existing studies have largely focused on individual cells and their SOH and SOC metrics, with less attention paid to larger battery packs, while the SOP more effectively reflects the overall operational characteristics of a battery pack. Therefore, this study proposes a parameter-optimized prediction method based on PSO and SVM, using SOP as a key health indicator for life prediction. In this method, the global optimal solution is obtained by simulating the collaborative search behavior of the particle swarm, dynamically updating particle positions and velocities; this solution determines the SVM’s critical parameters, namely the kernel parameter g and the penalty coefficient c, which are then used to train the SVM model to enhance its generalization ability and prediction accuracy. The results indicate that the PSO-SVM model can effectively capture the degradation characteristics of battery packs, achieving high-precision SOP prediction, with most prediction errors below 5% and the maximum error reaching 10.91%.
BibTeX
@inproceedings{yan2025state,
  author = {Hui Yan and Qingbin Wang and Yun Yang and Xianzhong Zhao and Zudi Huang and Yuxi Wang and Shi Liu and Yi Bin and G. S. Huang and Jianfeng Yang},
  title = {{State of Power Prediction for Battery Packs Based on Particle Swarm Optimization SVM}},
  booktitle = {Batteries},
  year = {2025},
  doi = {10.3390/batteries11120448},
}
Figure