Inactive Constraint Reduction and Search Domain Restriction Based on Machine Learning for Large Scale Unit Commitment

Abstract

As a unit commitment in a power system operation, a generation schedule indicating optimal generations and on/off of generators is created so that the power needed by consumers can be generated and transmitted on transmission lines. In order to determine the huge number of on/off for each generator at each time, under numerous operational constraints for uncertainty generations and the maximum transmission capacity violations due to localized generations of renewable energy, a large-scale combinatorial optimization is required to solve for making a generation schedule. On the other hand, to solve this large-scale optimization problem by only a standard optimization method takes long computation time. Therefore, in order to shorten the computation time, by using machine learning, invalid constraints that do not affect a resultant schedule are eliminated to reduce the optimization scale, and the optimal on/off are estimated to restrict the optimization search domain. To evaluate the performance of the proposed method, it is applied to the IEEE-118 bus power system model. The results show that the proposed method enables up to 82% reduction in computation time while maintaining an optimization accuracy of 0.8%.