DNN-Based Active Constraints Screening to Preprocess SCUC Problem

Security-constraint unit commitment (SCUC) determines which generation units must be on and off-line over a time horizon. The computational burden increases from the increase in system size and various constraints. This paper proposes a method with an integration of a machine learning approach and optimization to solve the SCUC problem. A preprocessing strategy based on the deep neural network by predicting active voltage and branch constraints is applied to reduce the computation time of the SCUC problem. Numerical results of the modified IEEE 30-Bus system and IEEE 118-Bus system suggest that active constraints can be figured out with high probability in a very short time. Moreover, the constraint-reduced SCUC problem can produce competitive results in terms of computational efficiency with almost no loss of solution quality compared with the full constraints SCUC problem. The proposed approach achieves speedups of between 20 and 40% on different testing examples.