Offline–online efficient solver parameter selection for accelerating unit commitment calculation

Unit commitment (UC) is a large-scale discrete optimization problem widely applied in power system operation, whose significant computational burden motivates recent advances in acceleration UC computation. A practical challenge arises that it is computationally expensive to tune specific parameters in the integration of different algorithms, resulting in the limitation of solution efficiency, such as in commercial discrete optimization solvers. Therefore, this paper proposes an offline–online efficient solver parameter selection method, where offline tuning history has been exploited to inform online efficient selection. First, in the offline tuning process, a representative instance is carefully tuned to construct a dataset of parameter configurations and their corresponding effectiveness. Second, a small set of efficient parameters and their specific values is strategically identified based on the data-driven analysis. Finally, in the online selection process, those limited parameter configurations are evaluated in parallel to accelerate the computation of other instances. 100 test cases from RTS-GMLC demonstrate that the proposed method reduces computational time by 17.07% on average, without introducing additional parameter tuning cost. • Propose an efficient solver parameter selection method for unit commitment problems. • Offline tuning history is exploited to inform online efficient selection. • A representative instance is tuned to construct a dataset in the offline stage. • A small set of efficient parameters is identified in the online stage. • Test cases from RTS-GMLC demonstrate the effectiveness of proposed method.