Evaluation of critical outage duration for PV/BES and PV/BES/H₂ systems with machine learning models

This study presents a novel framework for evaluating the reliability of hybrid renewable energy systems (HRES) under grid outage conditions by introducing Critical Outage Duration (COD) decision-making metric that identifies the threshold at which system reliability, measured by the Loss of Power Supply Probability (LPSP), begins to degrade. Unlike traditional reliability metrics, COD directly relates to grid interruption tolerance, offering planners a practical tool for system design. The techno-economic performance of two configurations namely PV/Battery Energy Storage (PV/BES) and PV/BES/Hydrogen (PV/BES/H₂) was evaluated for a residential setting in Malaysia using HOMER simulations. To generalize insights and eliminate the need for repeated simulation, five machine learning (ML) models of MLPNN, KNN, DT, RBNN, and SVM were developed to predict COD, LPSP, and Cost of Electricity (COE) based on component sizes and load levels. Among them, MLPNN achieved the highest accuracy in predicting LPSP and COE, while showing moderate effectiveness in COD prediction. This integration of a new reliability metric with predictive ML models not only improves system adaptability and design precision but also supports scalable, resilient energy planning across diverse outage scenarios.