Robust Learning-assisted Data-driven Congestion Management via Sparse Sensitivity Estimation

Congestion management based on locational marginal pricing (LMP) and financial transmission right (FTR) mechanisms is vital to power system secure operation. This paper proposes a novel robust learning-assisted data-driven congestion management strategy against cyber-physical uncertainties. This is achieved by embedding a novel sparse estimation method of distribution factors (DFs), which allows for learning the inherent sparsity information of DFs. This method consists of a robust sparse estimator and an efficient online algorithm, integrating robust, adaptive, and sparse estimation techniques in a fast recursive parallel computing framework. The sparsified DFs promote constraint reduction to build tractable, timely, and robust surrogates of high-dimensional optimization problems against uncertainties. Comparative results on a large-scale system validate that the proposed strategy accurately and fast computes LMPs, congestion patterns, and FTR revenues without relying on exact system models and massive historical data. Without loss of generality, this work yields a new implication for optimization-based power system operation applications: model-less, robust, and efficient operations can be achieved via embedding sparse sensitivity estimation into optimization problems.