Machine Learning-Based Parallel Computing Warm-start Method for Bilevel Decomposition for SCUC

The Unit Commitment (UC) problem is a common power dispatching mathematical model handled by independent system operators (ISOs). Security-Constrained Unit Commitment (SCUC) problems arise when network security limits. As the number of units involved in daily operations increases and the constraints become more complex, the SCUC problem becomes larger and more difficult to solve. To overcome the challenge of solving such a large-scale MILP, we propose a bilevel decomposition method based on machine learning to rapidly solve SCUC. This method accomplishes the bilevel decomposition of hydro, thermal power, and time based on the characteristics of the problem and a substantial quantity of data learning. Multiple subproblems can be solved in parallel, and high-quality, feasible solutions for warm start or fixing variables. The results of numerical experiments indicate that our method provides more efficient solutions of higher quality than using CPLEX directly for large-scale instances.