Artificial intelligence applied to the operation and control of power grids

Abstract

This work proposes a vision and architecture of artificial intelligence based on deep neural networks for the large-scale and secure control of the transmission grid. The solution integrates SCADA data and phasor measurement unit (PMU) readings with operational context, including network topology, demand, weather, and equipment temperature.

In this decision-making cycle, both supervised learning—using deep neural networks trained with scenario labels derived from human operation—and reinforcement learning—for maneuver policies capable of acting within configurations not yet known to the human operator—are employed.

The system coordinates voltage, current, power, and frequency control, demand management, and equipment operation (transformers, generators, circuit breakers, capacitor banks, etc.), enabling active and proactive responses to contingencies and routine switching. An operational dashboard provides situational awareness and explainable recommendations, with prior validations (load flow and operational constraints) before execution and always keeping the operator in the loop.

Tests with historical events and simulated scenarios show shorter response times, containment of limit violations, and relief of overloads, demonstrating both feasibility and scalability. Looking forward, the proposed architecture enables coordinated distributed control, enhancing dynamic stability and resilience against normal and extreme weather events—representing a concrete step toward the future intelligent, data-governed power grid.