A data-driven framework for microgrid design integrating machine learning model with economic-energy-environmental parameters

This study proposes a data-driven framework for designing community microgrids that integrate photovoltaic systems, wind turbines, diesel generators, and battery storage. The framework optimizes microgrid configurations based on economic, energy, and environmental (3E) sustainability performance indicators (3E-SPI). To achieve these objectives, we developed a data-driven model that combines Homer-Pro with a custom Python tool integrating extreme gradient boosting (XGBoost) machine learning algorithm and thirteen 3E-SPI calculations for community microgrid systems. Subsequently, a multi-objective optimization model with a two-layer multi-criteria decision-making (MCDM) approach was employed to evaluate microgrid configurations based on thirteen 3E-SPI to support stakeholders in the decision-making process. In the first layer, Best Worst Method (BWM) determines the weights of the 3E-SPI, whereas in the second layer, Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) and VIˇsekriterijumsko Kompromisno Rangiranje (VIKOR) methods are used to rank microgrid alternatives. The predictive performance of XGBoost was compared with that of random forest (RF), support vector regression (SVR), and deep neural network (DNN). The analysis revealed that XGBoost outperformed other models, achieving superior predictive performance, with a coefficient of determination (R²) exceeding 0.95. The MCDM results indicate that hybrid photovoltaic/wind/battery/diesel microgrid is the optimal solution for the studied community, yielding a total net present cost of approximately $1.3 million, a levelized cost of energy of $0.29/kWh, and annual CO₂ emissions of 169.11 kg. Overall, the proposed framework provides a practical tool for policymakers and energy planners to design cost-effective, reliable, and sustainable microgrids.