Learning to Solve Large-Scale Deterministic and Stochastic Security Constrained Unit Commitment Problems

Machine learning has been used in various areas in the energy sector, e.g., forecasting and detection. In this talk, we focus on using machine learning techniques to improve the computational performance of fundamental mixed-integer programming optimization problems in power systems: security-constrained unit commitment and its stochastic version. Machine learning is used to extract information from historical instances (e.g., input data, solutions, solver information) to improve the computational performance of MILP solvers when solving similar instances in the future. Computational results show that the proposed machine learning approaches can effectively improve performance.