Data-driven Optimal Scheduling of Isolated Microgrid Using Random Forest Regressor

Abstract

Design, operation, and planning of Microgrids (MGs) have been enriched by advances in machine learning (ML) techniques and availability of real data. These advancements have significantly improved the accuracy of predictions related to energy demand and renewable generation within MGs. ML-based algorithms can analyze complex patterns in data to make precise forecasts about future energy demands, renewable energy output, and even potential system failures. In this context, this paper utilizes real open-access wind speed dataset from Nottingham to create a forecast model based on a random forest regressor (RFR). This forecasted data, along with actual wind speed for wind power generation, serves as the input for a mixed-integer linear programming model designed for the optimal scheduling of standalone microgrids (OSSM). The results demonstrate that the discrepancies in the total cost, voltage deviation, and power loss of the MG, when comparing forecasted data to actual data, are less than 5%, 1%, and 2%, respectively. These results validate the effectiveness of ML-based models, specifically RFR, in the development of OSSM.