abderrahmane802/battery-soh-dataset

Battery State of Health (SoH) Dataset

Synthetic battery degradation dataset for training SoH prediction models, mimicking real patterns from NASA PCOE, CALCE, and BatteryLife datasets.

Dataset Configs

1. raw_cycles — Per-cycle measurements

Flat table with one row per battery cycle. Best for exploration, visualization, and feature engineering.

Column	Description
battery_id	Unique battery identifier (0-79)
cycle_number	Cycle index within the battery
initial_capacity_Ah	Initial rated capacity (Ah)
current_capacity_Ah	Current capacity at this cycle (Ah)
SoH	Target: State of Health (0.5-1.0)
avg_discharge_voltage_V	Average discharge voltage (V)
charge_capacity_Ah	Measured charge capacity (Ah)
avg_temperature_C	Average cell temperature (°C)
internal_resistance_Ohm	Estimated internal resistance (Ω)
charge_time_ratio	Charge time ratio (increases with degradation)
voltage_drop_V	Voltage drop at end of discharge (V)
coulombic_efficiency	Coulombic efficiency (ratio)
cycle_number_norm	Normalized cycle number (0-1)
charge_rate_C	Charge C-rate (0.5, 1.0, 1.5, 2.0)
discharge_rate_C	Discharge C-rate (0.5, 1.0, 1.5, 2.0)
knee_point_cycle	Cycle where accelerated degradation begins
base_temperature_C	Ambient temperature (°C)

2. windowed — LSTM-ready sequences

Sliding window sequences of 20 consecutive cycles. Ready for direct LSTM model training.

Column	Description
features	20×8 matrix: 20 cycles × 8 features per cycle
SoH	Target: SoH at the cycle following the 20-cycle window
battery_id	Battery identifier
cycle	Cycle number of the target SoH
initial_capacity_Ah	Initial capacity
charge_rate_C	Charge protocol C-rate
discharge_rate_C	Discharge protocol C-rate

Feature order in the 8-feature vector:

1. avg_discharge_voltage_V
2. charge_capacity_Ah
3. avg_temperature_C
4. internal_resistance_Ohm
5. charge_time_ratio
6. voltage_drop_V
7. coulombic_efficiency
8. cycle_number_norm

Data Generation Details

• 80 batteries with cell-to-cell variation
• 200-400 cycles per battery
• Degradation model: exponential + linear + knee-point acceleration
• Capacity range: 1.85-2.15 Ah (typical 18650 Li-ion)
• Knee point: random between cycle 200-400
• SoH range: 0.50-1.00
• Split: 70% train / 15% validation / 15% test (by battery ID, no data leakage)
• Seed: 42 (fully reproducible)

Usage

from datasets import load_dataset

# Load raw per-cycle data
raw = load_dataset("abderrahmane802/battery-soh-dataset", "raw_cycles")

# Load LSTM-ready windowed data  
windowed = load_dataset("abderrahmane802/battery-soh-dataset", "windowed")

# Quick peek
print(raw["train"][0])
print(windowed["train"].features)

Associated Model

Trained LSTM model: abderrahmane802/battery-soh-lstm

References

• BatteryLife (arxiv 2502.18807) — degradation patterns
• NASA PCOE Battery Dataset — capacity fade curves
• CALCE Battery Research Group — cycling protocols