System-Level Integration of Artificial Intelligence in SCADA-Based Power Distribution Networks: Architecture and Performance Evaluation

This paper presents a system-level integration framework for embedding artificial intelligence (AI) into SCADA-based power distribution networks to enhance operational performance and decision-making. A four-layer reference architecture maps AI inference across field devices, substation edge systems, control-center, and deterministic safety layers. The framework is evaluated on the IEEE 33-bus feeder using Monte Carlo fault injection (N = 10,000). The simulation results show that edge-resident AI reduces Fault Detection Latency (FDL) by 56.7 %, Restoration Time (RT) by 60.0 %, and False Alarm Rate (FAR) by 62.4% compared to conventional SCADA. The Operational Degradation Index (ODI) decreases from 0.180 to 0.071, indicating improved reliability. A Proximal Policy Optimization (PPO) Reinforcement Learning agent for Volt-VAR Optimization reduces active power losses by 8.6%, and improves voltage compliance to 94.3%. The safety validation layer verifies that AI controls actions in 99.97% of cases to ensure safe execution. Therefore, these results illustrate that the integration of AI as a system-level component can enable SCADA to be used as a real-time decision support platform for critical infrastructure.