Learning-Integrated Unit Commitment Optimization Based on GCN-GRU and Load Scenario Clustering

Unit commitment (UC) is a fundamental problem in power systems, typically formulated as a mixed-integer linear programming (MILP) model. As the scale of the system expands, numerous variables and constraints, especially binary variables, impose a significant computational burden on solving the UC problem. This paper proposes a learning-integrated optimization framework to accelerate the solution of the UC problem. First, a graph convolutional network (GCN)-gated recurrent unit (GRU) based model is employed to predict commitment decisions. Next, multiple prediction models are established by clustering load scenarios to reduce learning complexity. Finally, the UC problem is solved by fixing high-confidence predictions through confidence filtering. Case studies conducted on the IEEE 118-bus system using data generated from real-world load scenario characteristics demonstrate that the proposed framework effectively accelerates the solution of the UC problem with only a slight loss in solution quality.