Autonomous AI framework detects and restores grid faults in smart distribution networks

Researchers have developed an autonomous artificial intelligence framework that can detect, isolate and restore faults in a smart distribution network with minimal human intervention, combining machine learning with a distributed multi-agent system. The work, published in Energy Informatics by Vishakh K Hariharan and Manjula G. Nair with Fabrizio Granelli, addresses a persistent gap in self-healing protection schemes, which the authors say still struggle to achieve scalable, autonomous fault detection and coordinated service restoration in active distribution networks.

The framework integrates supervised and unsupervised machine learning with a multi-agent system, or MAS, to coordinate the three stages of grid self-healing: detecting the fault, isolating it, and restoring service. Supervised machine learning models handle faults the system has seen before, while an autoencoder-based anomaly detection model is tasked with identifying previously unseen faults. The overall control first detects the fault using machine learning techniques, then isolates it and restores the affected section using the multi-agent system.

Performance figures reported in the study are striking. The proposed autoencoder achieved 94.5% accuracy for detecting previously unseen faults and 99.9% accuracy for known faults. By contrast, conventional supervised machine learning models showed substantially lower performance when confronted with fault scenarios they had not encountered during training, underscoring the value of pairing supervised detection with an unsupervised anomaly detector.

The test bed for the study was a nine-bus system, used as the active distribution system under study to validate the proposed control.

Beyond simulation, the team also built a hardware prototype to examine the impact of the multi-agent system.

The authors note that their approach can increase the robustness and dependability of the power grid even in situations such as high renewable energy penetration and complex energy source scenarios.

Existing self-healing protection schemes, according to the paper, face challenges in achieving scalable, autonomous fault detection and coordinated service restoration in active distribution networks. The proposed framework is presented as a response to those limitations: a scalable, decentralized and intelligent self-healing solution capable of improving fault resilience and service restoration in future smart distribution networks with high penetration of distributed energy resources.

Decentralization is a key design choice. Rather than relying on a single central controller, the multi-agent system distributes the tasks of fault isolation and service restoration across cooperating agents, an architecture the authors associate with scalability. Combined with the layered machine learning approach, the framework aims to keep the grid operating even when fault types fall outside the training data of its supervised models.

The study is a simulation and prototype validation on a nine-bus benchmark system, and the reported accuracy figures apply to that setting; the article is published as an open-access, peer-reviewed accepted manuscript. The authors position the framework as a step toward self-healing protection for the smart grids of the future, where machine learning handles detection and multi-agent coordination handles the physical response to faults on the network.

Subject of Research: Machine learning and multi-agent systems for autonomous fault detection, isolation and service restoration in smart distribution networks

Article Title: An autonomous AI framework for automated fault handling and system restoration in smart distribution networks

Article References: Hariharan, V. K., Nair, M. G., & Granelli, F. (2026). An autonomous AI framework for automated fault handling and system restoration in smart distribution networks. Energy Informatics. https://doi.org/10.1186/s42162-026-00686-x

Image Credits: AI Generated

DOI: 10.1186/s42162-026-00686-x

Keywords: smart distribution networks, machine learning, autoencoder, anomaly detection, multi-agent systems, fault detection, self-healing grid, service restoration, distributed energy resources, power grid resilience, autonomous, framework