Robust Data-driven Sparse Estimation of Power Flow Sensitivities for Smart Grid Monitoring and Operation

Distribution factors (DFs) are sensitivities of active power line flows with respect to system control variables, which play a crucial role in smart grid monitoring and secure operation. This paper proposes a novel data-driven framework for robust sparse DF estimation that can handle the collinearity of high-dimensional synchrophasor measurements in large-scale smart grids. It does not require power flow models, thereby facilitating power flow sensitivity analysis with adaptiveness to operating-point changes and data uncertainties from renewable generations. An Elastic-net (Enet) estimator is used to sparsify DFs, and a new LARS-Enet algorithm is designed to achieve the data-driven estimation. Owing to the
$l_{1,2}$
regularization in Enet, this framework can yield the sparse estimator with robustness to collinearity. Due to the early-stopping property and data-driven settings of LARS-Enet, this framework can solve the estimator fast online. Results carried out on the IEEE 300-bus system indicate the potential of the proposed framework for implementing new grid codes associated with power flow sensitivity analysis and advancing the future development of power-flow security-constrained operations in large-scale smart grids.