Evaluating and comparing machine learning approaches for effective decision making in renewable microgrid systems

Abstract

This study addresses the critical issue of energy management in micro-grid (MG) systems incorporating renewable energy sources and hydrogen storage. The research introduces an innovative approach by conducting a comparative analysis of two machine learning methods, namely k-Nearest Neighbors (k-NN) and Random Forest (RF), to optimize micro-grid decision-making. The investigation reveals the consistent superiority of Random Forest, particularly in precision and F1-scores, across key micro-grid components such as the fuel cell relay, battery relay, super-capacitor relay, and grid system relay. The results demonstrate that the RF method consistently achieves high macro average precision factors (90%, 86%, 84%, 82%) and impressive macro average F1-scores (90%, 87%, 88%, 85%), surpassing the performance of k-NN, which yields notably lower precision factors (30%, 15%, 14%, 28%) and F1-scores (41%, 23%, 26%, 34%). This superior performance positions RF as a robust machine learning method for micro-grid decision-making, specifically in the realm of energy storage and renewable sources.The novelty of this work lies in establishing Random Forest as a reliable tool capable of handling the intricacies of micro-grid decision-making, thereby enhancing sustainable energy management processes. Additionally, the resilience of RF to imbalanced data adds to its effectiveness in diverse operational scenarios. This research sheds light on the potential of RF to contribute significantly to the advancement of sustainable energy solutions in micro-grid systems.

Keywords

• Machine learning
• Renewable microgrid
• Hydrogen storage management
• Underground hydrogen storage
• Cavity saline