Hybrid Al-MAS approach for forecasting and fault-detection in a microgrid

Abstract

This report presents the design and implementation of an intelligent system for forecasting and fault detection in renewable energy microgrids. The proposed architecture is based on a Multi-Agent System, where each agent is responsible for a specifi cfunctio nrelate dt osola rgeneration ,battery storage or electrical load. Artificia lintelligence model nlsare integrated with inthe agent st oenabl eaccurate, prediction and classification tasks.Solar power forecasting is performed using a recurrent neural network trained on environmental and temporal features. Battery state of charge is estimated using a tree-based regression model, and the load demand forecasting is performed using a similar model trained on temporal and historical consumption features, and fault classification is handled using an artificial neur alne two rktrained on structured electrical measurements. Each model maintained high accuracy metrics, with R 2 (Coefficien to fDetermination )value sexceedin g0.9 4fo rbot hSO Can dloa dforecastin gtasks ,closely matching actual trends. The fault classificatio nmode lachieve da naccurac yabov e98% ,ensuring reliable identificatio no ffaul tconditions. The agents operate asynchronously and exchange information through coordinated message queues to maintain consistency across scenarios. The system applies rule-based logic to combine the outputs of all agents and take appropriate decisions under different operating conditions. Evaluation results demonstrate reliable forecasting performance, accurate fault detection, and coherent agent coordination. These outcomes support the potential of AI-driven agent architectures to improve microgrid intelligence and resilience in realistic operating environments.