Smart monitoring for a greener future: New AI-driven model predicts lithium-ion battery health with unprecedented accuracy

Researchers develop a hybrid framework combining biological optimization with deep learning to overcome data noise and aging complexities, ensuring safer and more reliable battery operations.

As the world shifts toward carbon neutrality, lithium-ion batteries (LIBs) have become the backbone of green energy storage and electric transportation. However, accurately predicting a battery's "State of Health" (SOH) remains a major technical hurdle due to the "noise" generated during years of cyclic aging. In a study published in the international journal  ENGINEERING Energy (formerly Frontiers in Energy), a team of Chinese researchers has unveiled a novel estimation model that significantly improves prediction accuracy and reliability.

The research, led by Juqiang Feng from the Anhui University of Science and Technology and colleagues, introduces a fused algorithm that integrates advanced signal processing with optimized machine learning strategies. This breakthrough offers a robust decision-making tool for battery safety management systems (BSMS), potentially preventing hazardous failures such as fires or explosions in aging battery packs.

The Challenge of Aging Data

A battery's SOH is not a parameter that can be measured directly; it must be inferred from indicators like voltage, current, and temperature. As batteries age, internal chemical reactions become increasingly complex, and sensor data often carries significant noise or outliers, which can "confuse" traditional estimation models and lead to dangerous inaccuracies.

"An accurate assessment of SOH is the cornerstone for guaranteeing the long-term stable operation of electrical equipment," the researchers state. "However, the noise carried during cyclic aging poses a severe challenge to both accuracy and the model's ability to work across different battery types."

Whales and Neural Networks: A Hybrid Solution

To solve this, the team developed a multi-stage approach. First, they utilized the Whale Optimization Algorithm (WOA)—a technique inspired by the foraging behavior of humpback whales—to find the perfect parameters for Variational Modal Decomposition (VMD). This process allows the system to "clean" the raw battery data, accurately separating the useful health signals from background noise.

Once the data is refined, a Convolutional Neural Network (CNN)—the same technology used in advanced image recognition—extracts high-level features of the battery's health. Finally, a Support Vector Machine (SVM) acts as a high-precision regressor to deliver the final SOH prediction.

Proven Across Diverse Battery Types

What sets this study apart is its rigorous validation. The researchers tested the model on two very different datasets:

• High-Capacity Mining Batteries: 228 Ah LiFePO₄ batteries used in coal mine transport vehicles under harsh conditions.
• Small-Scale Consumer Batteries: 1.1 Ah LFP/graphite batteries subjected to various discharge protocols.

The results demonstrated that the proposed model effectively solves the interference of data noise and consistently outperforms traditional techniques in accuracy and generalization. Whether at different temperatures or discharge rates, the algorithm provided a stable and precise "health check" for the batteries.

Impact on Sustainable Development

By providing a more reliable way to monitor battery degradation, this technology supports UN Sustainable Development Goals related to clean energy and climate action. It enables battery operators to optimize maintenance schedules, prolong battery lifespans, and ensure the safety of large-scale energy storage systems essential for a decarbonized global power grid.

JOURNAL: ENGINEERING Energy (formerly Frontiers in Energy)

DOI: https://doi.org/10.1007/s11708-024-0969-x

Article Link: https://link.springer.com/article/10.1007/s11708-024-0969-x

Cite this article: Zhang X, Feng J, Cai F, Huang K, Wang S. A novel state of health estimation model for lithium-ion batteries incorporating signal processing and optimized machine learning methods. Front. Energy, 2025, 19(3): 348–364. https://doi.org/10.1007/s11708-024-0969-x