MI-SOH: a multi-indicator feature dependency model for lithium-ion battery state-of-health Estimation

Area of Science:

• Battery Technology
• Artificial Intelligence
• Data Science

Background:

• Accurate state-of-health (SOH) estimation is critical for lithium-ion battery safety and lifespan.
• Existing methods struggle with dynamic feature correlations during battery degradation.
• Non-stationary aging patterns in batteries reduce the accuracy of static SOH estimation models.

Purpose of the Study:

• To develop a multi-indicator SOH estimation model that dynamically adapts to evolving feature importance.
• To improve SOH estimation accuracy for lithium-ion batteries under complex aging conditions.
• To provide a practical SOH monitoring framework for intelligent battery management systems (BMS).

Main Methods:

• Proposes MI-SOH, a model with four components: Multi-indicator Feature Weighting, Temporal Pattern Extraction, Cross-Variable Dependency Modeling, and Adaptive Hyperparameter Optimization.
• Employs dual-correlation analysis and dilated convolutions for adaptive feature prioritization and degradation dynamics capture.
• Utilizes inverted transformers to model interdependencies among health indicators throughout battery aging.

Main Results:

• MI-SOH demonstrates superior performance compared to mainstream prediction approaches on NASA and CALCE datasets.
• Achieved average Root Mean Squared Errors (RMSE) of 0.00312 and 0.01126 on benchmark datasets.
• Outperforms existing methods across diverse battery chemistries and lifecycles, indicating robustness.

Conclusions:

• MI-SOH offers a significant advancement in intelligent battery management systems.
• The dynamic SOH estimation framework is crucial for electric vehicle safety and energy storage reliability.
• This adaptive approach addresses limitations of static feature fusion in current SOH estimation models.