Machine learning models for the economic dispatch of islanded micro water-energy systems

The economic operation of interdependent water-energy infrastructure is crucial for accommodating an increasing population and enhancing operational and economic stability. Real-time resolution of this issue is challenging due to numerous decision variables and constraints, making it a nonlinear and non-convex problem. Traditional numerical solutions require significant computational resources, especially for large-scale systems. This paper introduces a novel machine learning-based approach to solve the combined economic dispatch problem of an islanded water-energy microgrid. Three machine learning models are proposed: multilayer perceptron, random forest, and support vector machines. These models are trained using datasets obtained by solving the mixed integer nonlinear programming framework of integrated water-energy systems. Findings reveal that the random forest model excels in accuracy, with mean squared errors in the order of $O\left(10^{-3}\right)$. Additionally, the model significantly outperforms traditional methods in computational efficiency, achieving a runtime less than a microsecond. The fastest model, multilayer perceptron, achieves a runtime reduction of 99.99%, showcasing substantial efficiency gains. These results demonstrate the feasibility of using machine learning models for minute-based economic dispatch in real-time operations, marking a significant advancement in managing islanded micro water-energy systems.