Data-Driven Island Detection Using Chi-Squared Discretization-Based Random Forest Approach for Microgrid With RES

Abstract

Machine learning models have been widely extended in island detection for microgrid with renewable energy sources (RESs) and have become the most promising extension in data-driven methods. Among the applicable machine learning models, random forest approach shows most potential since the issue of overfitting can be effectively addressed. However, accurate random forest models often involve a substantial number of tree nodes and require a significant amount of training data, impeding the online applications of island detection in industrial contexts due to unacceptable training and response time of learning models. In this paper, the chi-squared discretization-based random forest approach has been proposed for island detection in microgrids. In the proposed approach, the hierarchical discretization method is employed to reorganize the input training dataset, facilitating efficient model training for the random forest. A comprehensive numerical study has been conducted in the microgrid to validate the effectiveness of the proposed approach for island detection.