User-Induced Heuristics for Security-Constrained Unit Commitment: Variable Influence Diving and Variable Significance Neighborhood Search

The heuristics in mixed-integer linear programming (MILP) solvers are mostly general-purpose, which can only recognize some common structures. It sometimes may be challenging for them to obtain a high-quality feasible solution, thus reducing the overall efficiency of the MILP solver for solving large-scale security-constrained unit commitment (SCUC) problems. Based on the knowledge of the SCUC problem structure, this paper proposes two user-induced heuristics: variable influence diving (VID) and variable significance neighborhood search (VSNS). VID can iteratively select and fix fractional binary variables based on system parameters and the characteristics of SCUC, leading to the generation of a high-quality initial solution. In VSNS, a novel neighborhood-defining method is proposed, which identifies and fixes some principal binary variables by calculating a significance indicator. The proposed two heuristics can be built into any MILP solver and used for any node in branching, exploiting the input information after the presolve phase. The two heuristics are implemented in the CBC solver and tested on the IEEE 118-bus system, the Polish 2383-bus system, and the French system with 1888 buses, 1951 buses, and 2848 buses. The results demonstrate that the proposed approach can greatly enhance the computational efficiency for solving the SCUC problem.