Research on Optimal Dispatch of Thermal Power Units Integrated with Security Constraint Identification

Under the national promotion of the “Dual Carbon” goals, as power systems continue to develop with increasing renewable energy integration, traditional thermal unit scheduling methods face challenges such as large-scale optimization models caused by numerous security constraints in actual grid operations and difficulties in rapid solving. To address this, this paper constructs a deep learning-based day-ahead optimal scheduling model for thermal units considering security constraints.First, we propose an active constraint identification method based on Gated Recurrent Unit (GRU). This approach efficiently identifies the binding constraint set in day-ahead generation scheduling optimization models by analyzing characteristic parameters and deep neural network processing results. Subsequently, we integrate the identified effective constraints with security constraints to establish a security-constrained reward mechanism.Second, the effectiveness of the proposed constraint identification model is validated on the IEEE 30-bus system using YALMIP and the CPLEX solver, and the thermal generator model incorporating constraint identification is then formulated and solved. Finally, comparative experiments between the constraint identification-based model and iterative solution methods demonstrate that this approach effectively reduces model complexity and computational burden while improving computational efficiency.