Using machine learning to achieve highly efficient solutions of frequency-constrained unit commitment models

The decreasing rotational inertia of electric power systems induced by the increasing penetration of renewable energy resources degrades system frequency stability, which represents a significant challenge that must be addressed in frequency-constrained unit commitment (FCUC) models. However, the computation time required to solve FCUC models increases exponentially with increasing system scale. The present work addresses this issue by proposing a machine learning–based approach to increasing the computational efficiency with which the FCUC model is solved by reducing the number of binary variables employed in the FCUC model via accurate predictions of the mapping relationships between power system loads, the maximum output of renewable energy units, and the commitment patterns of thermal generation units. The evaluation outcomes of the methodology introduced in this paper, when applied to the IEEE-118 bus system, reveal a remarkable performance: it enhances the solution speed by a factor ranging from 2.98 to 9.79 times, while maintaining an objective function value that deviates by less than 0.1% from those obtained through conventional solution approaches. To ascertain its scalability, we further tested the refined IEEE-2383 bus system. Results showed a significant speed-up of 2.29 to 3.94 times, with objective function deviations within 0.3%.