How to Evaluate Battery Storage Health with Big Data and LightGBM

Project Background

Developing advanced storage technologies is crucial for achieving green energy transition and carbon‑neutral goals. In 2024, new‑type storage (mainly lithium‑ion batteries) accounts for nearly 50% of installed capacity, but its lifespan and safety need systematic evaluation.

Specific Requirements

The project focuses on two operational sites—Datang Leizhou and a sodium‑ion plant in Hubei. Tasks include data cleaning, analysis, lifespan and status prediction, and producing a final assessment report.

Overall Framework

The raw data are processed and split into two modules: (1) operational state assessment to detect anomalies, and (2) SOC/SOH prediction for remaining useful life. The workflow is illustrated in the diagram below.

Consistency Evaluation

Consistency is judged by SOC ranging from 5% to 95% and SOH generally above 80%. Temperature and voltage are the primary indicators; derived metrics include average voltage, voltage range, standard deviation, and temperature statistics.

DBSCAN clustering isolates abnormal individual cells—identified as cells 13, 44, and 67 in the example. Visual inspection shows voltage dips in the faulty cells, confirming the anomalies.

SOC/SOH Prediction

Traditional electrochemical models are costly and require precise parameter identification. LSTM models need large datasets and are slow. The team selected a LightGBM model, achieving an R² score of 0.7, though some high‑variance regions remain under‑captured.

Effect Demonstration

The implemented system accepts Excel uploads, performs battery cluster consistency evaluation, isolates abnormal cells, and predicts SOC for individual cells.

Conclusion

The project successfully delivered a battery‑cluster consistency assessment, abnormal cell detection, and SOC/SOH remaining‑life prediction using big‑data techniques.