# A Probabilistic Approach to Adaptive Protection in the Smart Grid

## TL;DR

This paper introduces a probabilistic, data-driven protection strategy for smart grids using Gaussian Discriminant Analysis, enhancing fault detection, interpretability, and adaptability across different network configurations.

## Contribution

It develops a novel, interpretable protection method that reduces communication needs and improves fault isolation in smart grids during various operational modes.

## Key findings

- Effective fault detection in simulated smart grid scenarios
- Reduced communication requirements for protection relays
- Successful adaptation across different network topologies

## Abstract

Smart grids are critical cyber-physical systems that are vital to our energy future. Smart grids' fault resilience is dependent on the use of advanced protection systems that can reliably adapt to changing conditions within the grid. The vast amount of operational data generated and collected in smart grids can be used to develop these protection systems. However, given the safety-criticality of protection, the algorithms used to analyze this data must be stable, transparent, and easily interpretable to ensure the reliability of the protection decisions. Additionally, the protection decisions must be fast, selective, simple, and reliable. To address these challenges, this paper proposes a data-driven protection strategy, based on Gaussian Discriminant Analysis, for fault detection and isolation. This strategy minimizes the communication requirements for time-inverse relays, facilitates their coordination, and optimizes their settings. The interpretability of the protection decisions is a key focus of this paper. The method is demonstrated by showing how it can protect the medium-voltage CIGRE network as it transitions between islanded and grid-connected modes, and radial and mesh topologies.