abderrahmane802/battery-soh-dataset

这是一个用于训练电池健康状态(SoH)预测模型的合成电池退化数据集，模拟了来自NASA PCOE、CALCE和BatteryLife数据集的真实模式。数据集包含两种配置：raw_cycles（每循环测量数据，适用于探索、可视化和特征工程）和windowed（LSTM就绪序列，适用于直接LSTM模型训练）。数据生成细节包括80个电池，每个电池200-400个循环，具有电池间差异，退化模型采用指数+线性+拐点加速，容量范围为1.85-2.15 Ah（典型的18650锂离子电池），SoH范围为0.50-1.00，并按电池ID分为70%训练/15%验证/15%测试集（无数据泄漏）。

Synthetic battery degradation dataset for training SoH prediction models, mimicking real patterns from NASA PCOE, CALCE, and BatteryLife datasets. Contains two configurations: raw_cycles (per-cycle measurements, best for exploration, visualization, and feature engineering) and windowed (LSTM-ready sequences for direct model training). Data generation includes 80 batteries with cell-to-cell variation, 200-400 cycles per battery, exponential + linear + knee-point acceleration degradation model, capacity range 1.85-2.15 Ah (typical 18650 Li-ion), SoH range 0.50-1.00, and 70% train / 15% validation / 15% test split by battery ID (no data leakage).