[Paper Review] State of health estimation for lithium-ion battery by combining incremental capacity analysis with Gaussian process regression

Xiaoyu Li, Zhenpo Wang|arXiv (Cornell University)|Mar 15, 2019

Advanced Battery Technologies Research31 references4 citations

TL;DR

This paper proposes a novel Gaussian process regression model that combines incremental capacity analysis with partial incremental capacity curves to estimate lithium-ion battery state of health (SOH). By using an advanced Gaussian filter to smooth capacity curves and extracting health indexes from them, the method achieves accurate, robust SOH prediction with uncertainty quantification across diverse aging conditions and initial states.

ABSTRACT

The state of health for lithium battery is necessary to ensure the reliability and safety for battery energy storage system. Accurate prediction battery state of health plays an extremely important role in guaranteeing safety and minimizing the maintenance costs. However, the complex physicochemical characteristics of battery degradation cannot be obtained directly. Here a novel Gaussian process regression model based on partial incremental capacity curve is proposed. First, an advanced Gaussian filter method is applied to obtain the smoothing incremental capacity curves. The health indexes are then extracted from the partial incremental capacity curves as the input features of the proposed model. Otherwise, the mean and the covariance function of the proposed method are applied to predict battery state of health and the model uncertainty, respectively. Four aging datasets from NASA data repository are employed for demonstrating the predictive capability and efficacy of the degradation model using the proposed method. Besides, different initial health conditions of the tested batteries are used to verify the robustness and reliability of the proposed method. Results show that the proposed method can provide accurate and robust state of health estimation.

Motivation & Objective

• To improve the accuracy and robustness of lithium-ion battery state of health (SOH) estimation in energy storage systems.
• To address the challenge of indirect measurement of complex physicochemical degradation processes in batteries.
• To develop a model that provides both SOH predictions and quantified uncertainty for reliability and maintenance optimization.
• To validate the method across diverse aging datasets with varying initial health conditions.

Proposed method

• An advanced Gaussian filter is applied to smooth incremental capacity curves, reducing noise and enhancing feature extraction.
• Health indexes are extracted from partial incremental capacity curves as input features for the Gaussian process regression model.
• The mean function of the Gaussian process model predicts the battery's state of health over time.
• The covariance function models the uncertainty in SOH predictions, enabling confidence estimation.
• The model is trained and validated using four aging datasets from the NASA battery data repository.
• The approach is tested under different initial health conditions to assess robustness and reliability.

Experimental results

Research questions

• RQ1 Can incremental capacity analysis effectively extract health-related features from battery degradation data for SOH estimation?
• RQ2 How accurately can a Gaussian process regression model predict SOH using partial incremental capacity curves as inputs?
• RQ3 To what extent does the proposed method maintain accuracy and robustness across batteries with different initial health states?
• RQ4 Can the model provide reliable uncertainty estimates alongside SOH predictions?
• RQ5 How does the integration of Gaussian filtering with incremental capacity analysis improve prediction performance compared to conventional methods?

Key findings

• The proposed method achieves highly accurate state of health estimation across all four NASA aging datasets.
• The model demonstrates robust performance under varying initial health conditions, confirming its reliability in real-world scenarios.
• Uncertainty quantification via the covariance function provides confidence intervals for SOH predictions, enhancing system safety.
• The use of partial incremental capacity curves as features improves model generalization and noise resilience.
• The advanced Gaussian filter significantly enhances the quality of incremental capacity curves, enabling more precise feature extraction.
• The method outperforms conventional approaches in terms of prediction accuracy and robustness to initial state variability.