Near Real-Time Economic Dispatch Forecast of Islanded Water–Energy Microgrid Systems Using LSTM Neural Networks

The growing demand for power and water necessitates economic cooperation between water and energy infrastructure. Recent studies considered the economic dispatch of the integrated water–energy systems using numerical optimization frameworks, which are computationally demanding and limit their real-world usage. To address this, a novel, near real-time, economic dispatch forecast framework for an islanded water–energy microgrid system is proposed. This framework employs a long short-term memory (LSTM) neural network approach to forecast the power dispatch of the conventional generators, renewable energy units, batteries, and power consumption of the water pumping station. Initially, a mixed-integer optimization model is solved numerically to obtain hourly optimal dispatch outputs using real-world load and weather datasets for all four seasons. The resulting dispatch is interpolated in a minute resolution, feeding the LSTM model for training, and validation. A significant runtime reduction (above 99%) is observed with the proposed forecasting framework, demonstrating its effectiveness for large-scale systems.