A multi-scale time-series dataset with benchmark for machine learning in decarbonized energy grids.

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• The transition to carbon neutrality faces challenges in maintaining electric grid reliability due to increased renewable energy integration.
• Data-driven machine learning (ML) approaches are crucial for addressing these challenges in future electric grids.

Purpose of the Study:

• Introduce PSML, an open-access, multi-scale time-series dataset designed to support ML-based solutions for electric grid reliability.
• Facilitate research in critical areas like disturbance detection, energy forecasting, and synthetic data generation for power systems.

Main Methods:

• Synthesized a joint transmission and distribution electric grid dataset capturing complex grid dynamics.
• Included power, voltage, and current measurements across multiple spatio-temporal scales.
• Established ML benchmarks for key use cases using the PSML dataset.

Main Results:

• Demonstrated state-of-the-art ML performance on disturbance detection, classification, and localization.
• Achieved robust hierarchical forecasting for load and renewable energy generation.
• Enabled realistic synthetic generation of physical-law-constrained grid measurements.

Conclusions:

• The PSML dataset is a valuable resource for advancing ML research in safety-critical systems for electric grids.
• This resource will accelerate ML contributions towards the decarbonization of energy sectors.