Speed up of data-driven state estimation using low-complexity indexing method

Traditional power system state estimation methods lack the ability to track and manage increasing uncertainties inherent in the new technologies, such as recent and ongoing massive penetration of renewable energy, distribution intelligence, and plug-in electric vehicles. To deal with the inability, a recent work proposes to utilize the unused historical data for power system state estimation. Although able to achieve much higher accuracy, the new approach is slow due to the burden by sequential similarity check over large volumes of high dimensional historical measurements, making it unsuitable for online services. This calls for a general approach to preprocess the historical data. In this paper, we propose to achieve such a goal with three steps. First, because the power systems are with periodic patterns, which create clustered measurement data, dimension reduction is proposed to remove redundancy, but still able to retrieve similar measurements. To further reduce the computational time, the k-dimensional tree indexing approach is employed in step two to group the clustered power system data into a tree structure, resulting in a log-reduction over searching time. Finally, we verify the obtained historical power system states via AC power system model and the current measurements to filter out bad historical data. Simulation results show that the new method can dramatically reduce the necessary computational time for online data-driven state estimation, while producing a highly accurate state estimate.