Skip to content

Glossary

This glossary defines key terms used in BatteryML documentation.

A

Arrhenius Factor: Temperature-dependent factor exp(-Ea/RT) used to model temperature effects on degradation.

B

BaseModel: Abstract base class for all neural network models in BatteryML.

BasePipeline: Abstract base class for all feature extraction pipelines.

C

Cache: Hash-based caching system for expensive computations (especially ICA).

Cell ID: Identifier for individual battery cells (e.g., 'A', 'B', 'C').

C-rate: Charge/discharge rate normalized by capacity (1C = full capacity in 1 hour).

D

dQ/dV: Incremental capacity analysis metric, derivative of capacity with respect to voltage.

Degradation: Gradual loss of battery capacity and performance over time.

DualTracker: Experiment tracker that logs to both local files and MLflow simultaneously.

E

Experiment ID: Identifier for experiments (1-5) in the LG M50T dataset.

Early Stopping: Training technique that stops training when validation loss stops improving.

F

Feature Dimension: Number of features in a sample (e.g., 15 for summary features).

Fit/Transform: Pattern used by pipelines: fit on training data, transform on test data.

G

GPU: Graphics Processing Unit, used for accelerated neural network training.

H

Hash-Based Caching: Caching system that uses hash of parameters as cache key.

Hydra: Configuration management framework used in BatteryML.

I

ICA: Incremental Capacity Analysis, technique for analyzing battery degradation.

Impedance: Internal resistance of battery.

L

LAM: Loss of Active Material, degradation mechanism.

Latent Dimension: Dimension of latent state in Neural ODE models.

LightGBM: Gradient boosting framework, fastest model in BatteryML.

LLI: Loss of Lithium Inventory, degradation mechanism.

LOCO: Leave-One-Cell-Out, cross-validation strategy.

LossRegistry: Registry system for managing loss functions (MSE, Huber, Physics-Informed, etc.).

M

MAPE: Mean Absolute Percentage Error, evaluation metric.

MAE: Mean Absolute Error, evaluation metric.

MLflow: Experiment tracking and model management platform.

MLP: Multi-Layer Perceptron, simple neural network model.

N

Neural ODE: Neural Ordinary Differential Equation, continuous-time model.

Normalization: Scaling features to zero mean and unit variance.

O

ODE: Ordinary Differential Equation.

Output Dimension: Number of model outputs (typically 1 for SOH prediction).

P

Pipeline: Feature extraction component that transforms raw data to Samples.

PipelineRegistry: Registry system for managing pipelines.

R

R²: Coefficient of Determination, evaluation metric.

Registry Pattern: Design pattern for extensible component registration.

RMSE: Root Mean Squared Error, evaluation metric.

RPT: Reference Performance Test, periodic capacity measurement.

S

Sample: Universal dataclass format for data in BatteryML.

Savitzky-Golay: Smoothing filter used in ICA analysis.

SHAP: SHapley Additive exPlanations, model interpretability method.

SOH: State of Health, remaining capacity as fraction of initial capacity.

Split Strategy: Method for splitting data into train/validation/test sets.

T

TensorBoard: Visualization tool for training metrics.

Temperature Holdout: Split strategy that trains on extreme temperatures, validates on intermediate.

U

Unit Normalization: Automatic conversion of units (e.g., mAh → Ah, °C → K).

V

Validation Set: Data used to evaluate model during training.

Voltage Curve: Voltage vs. capacity curve from discharge test.

Next Steps