Battery Test Data Mining
Feature extraction and anomaly detection on large-scale battery test time-series during an internship at EVE Energy.
- Category
- Energy AI
- Published
- Aug 2024
- Stack
- Python
- Pandas
- NumPy
- SQL
- scikit-learn
Context
Battery Testing & Data Mining Intern at EVE Energy Co., Ltd · Jul – Aug 2024.
What I did
- Processed multi-dimensional test time-series with Python (Pandas/NumPy) and SQL to extract charge/discharge-loss features across temperatures from −20 °C to 60 °C and multiple power levels.
- Cleaned roughly 120 GB of test logs and lifted anomaly-detection accuracy by about 15% over a fixed-threshold baseline.
- Built Isolation Forest and Random Forest models to flag battery degradation and thermal-runaway precursors.
- Automated four pipeline steps — retrieval, conversion, comparison and visualisation — cutting daily manual review time by about 30%.