Deep Learning-Based Transmission Line Screening for Unit Commitment

Solving the transmission-constrained unit commitment (TC-UC) problem efficiently for high-quality solutions is one of the biggest challenges faced by independent system operators today. One way to achieve this is by removing superfluous network constraints from the problem. Several approaches have been used to identify such constraints. However, these methods are either too conservative or fail to maintain solution feasibility. In this paper, a deep learning-based approach is developed to identify the subset of transmission lines that can be safely removed from the TC-UC problem. The idea is to capture the temporal relationship between past line loading levels, nodal demands, and future line loading levels to identify superfluous network constraints. To achieve this, a novel regression-based classification approach is developed, where the regression model is used to predict line loading levels, and different thresholds are applied to classify transmission line capacity constraints as necessary or not for the TC-UC problem. The major advantage of this approach is that once trained, the model can be used under different classification thresholds. Numerical results show that the proposed approach significantly improves computational efficiency without compromising the solution quality.