Leveraging Machine Learning for Boosting Performance of Power Systems Optimization Problems

With the growing scale and complexity of modern power grids, available data could provide useful information and insights into the underlying power grid, and machine learning methods could be valuable to help understand and reveal the relationship of system parameters and optimal operations, finally addressing optimal operation problems in a more efficient and accurate way. In this talk, we will discuss three examples to marriage the data-driven and machine learning methods to power systems optimization problems for improving the computational performance, including: (i) a data-driven approach by leveraging historically solved unit commitment (UC) instances to promptly solve new UC problems; (ii) a closed-loop predict-and-optimize framework for deriving better forecasts that could lead to an enhanced UC solution quality; and (iii) an adaptive dynamic programming-based approach for conducting effective optimal schedules of energy-limited resources within a limited time horizon while considering their potential future values.