Related Experiment Video
Updated: Aug 14, 2026
10:36
Optimization of An Air-Based Heat Management System for Dusty Particulate Matter-Covered Lithium-Ion Battery Packs
Published on: November 3, 2023
A Hybrid Ridge Regression-Convolutional Bidirectional Long Short-Term Memory Framework with Dual-Level Transfer
Chengwei Ge1,2, Chunling Wu1,2, Zhen Zhang1,2
1School of Energy and Electrical Engineering, Chang'an University, Xi'an 710064, China.
Materials (Basel, Switzerland)
|August 13, 2026
Summary
This study introduces a hybrid Ridge regression-convolutional bidirectional long short-term memory model for accurate lithium-ion battery state-of-health estimation at high temperatures. The dual-level transfer learning approach significantly improves performance across different battery datasets.
Area of Science:
- Battery Technology
- Machine Learning
- Degradation Modeling
Background:
- Accurate state-of-health (SOH) estimation for lithium-ion batteries (LIBs) is critical for reliable operation, especially under high-temperature conditions (40-50 °C).
- Accelerated electrochemical degradation and complex nonlinear aging patterns at elevated temperatures pose significant challenges to existing SOH estimation methods.
- Existing models often struggle with cross-battery generalization and adapting to varying degradation behaviors.
Purpose of the Study:
- To develop a robust and accurate SOH estimation framework for LIBs operating at high temperatures.
- To address the challenges of accelerated degradation and nonlinear aging patterns in LIBs.
- To enhance the cross-battery adaptation capability of SOH estimation models using transfer learning.
Main Methods:
- A hybrid framework combining Ridge regression for global trend and convolutional bidirectional long short-term memory (BiLSTM) for nonlinear residuals.
- Implementation of a dual-level transfer learning strategy for cross-battery adaptation, involving prior-regularized regression and fine-tuning of the BiLSTM network.
- Extraction and adaptive denoising of four health-related features using locally weighted scatterplot smoothing.
Main Results:
- The proposed hybrid model achieved a root mean square error (RMSE) as low as 0.0009 on a single battery at 50 °C, outperforming other benchmark models significantly.
- The dual-level transfer learning strategy reduced RMSE from ~0.009 to 0.0028, demonstrating its effectiveness in cross-battery adaptation.
- Joint adaptation of the Ridge prior and residual network yielded a mean RMSE of 0.004325, showing superior performance compared to individual adaptation methods.
Conclusions:
- The hybrid Ridge-CNN-BiLSTM framework with dual-level transfer learning provides a highly accurate and robust solution for LIB SOH estimation under high-temperature conditions.
- Transfer learning is crucial for adapting SOH estimation models to different batteries and conditions, significantly improving generalization.
- The proposed method offers a promising approach for enhancing the safety and lifespan management of lithium-ion batteries in demanding environments.
