Synthetic Time-Series Dataset for Machine Learning-Based Early Detection of Grid Collapse

This dataset is a synthetic time-series dataset designed to simulate power grid operations with the goal of training machine learning models for early detection of grid collapses. It captures multi-dimensional features that influence grid stability over time.

Key Characteristics:
Size: 527,040 records (likely representing 1-minute intervals over a full year)

Type: Synthetic data mimicking real-world grid behavior patterns

Purpose: Train ML models to predict grid collapse events

Features:
Temporal Marker:

timestamp: Date and time (minute-level precision)

Grid Operational Metrics:

frequency_hz: Grid frequency (nominally 50Hz)

load_MW: Total power demand (in Megawatts)

gen_gas_MW: Gas-powered generation output

gen_hydro_MW: Hydroelectric generation output

voltage_pu: Voltage in per-unit values (1.0 = nominal)

Environmental Factor:

weather_index: Numeric indicator of weather conditions (negative values suggest severe weather)

Event Flags (Binary):

line_trip: Transmission line failure (0/1)

gen_outage_collapse: Generator outage leading to collapse (target variable)

Observed Patterns:
Shows gradual load fluctuations with corresponding generation adjustments

Frequency deviations from 50Hz suggest grid stress

Voltage variations (0.94-1.06 pu range visible) may indicate instability

Weather index correlates with some operational changes

Machine Learning Relevance:
Enables supervised learning for binary classification (collapse prediction) and regression prediction.

Time-series nature allows for sequence modeling (RNNs, Transformers)

Feature correlations can reveal precursor patterns to collapse

Synthetic nature ensures availability of rare event data (collapses)