Deep Learning-Based Hydrothermal Scheduling Integrating Wind Power and Pumped-Storage Hydropower for Low-Carbon Economic Dispatch

Abstract

The increasing penetration of variable renewable energy sources into electric power systems requires advanced optimization tools to address the complexity of hybrid hydrothermal scheduling while minimizing generation costs and carbon emissions. This study investigates the application of four deep learning architectures—Kolmogorov–Arnold networks (KANs), long short-term memory (LSTM), gated recurrent unit (GRU), and deep feedforward (DFF)—to solve the hydrothermal scheduling problem in hybrid power systems that incorporate wind power generation and pumped-storage hydropower (PSH) plants. The methods were evaluated on a 10-generator test system over a 24-h planning horizon in three objective-weighting scenarios, considering economic dispatch only, pure emission minimization only, and balanced objectives. All architectures successfully solved the integrated problem and satisfied the system constraints. This study reports the first application of the KAN to the hydrothermal scheduling problem, demonstrating its viability and interpretability potential for future applications in electric power systems.