On-line identification method of state sensitivity of a power system based on target regularization

Abstract

The invention discloses an on-line identification method of state sensitivity of a power system based on target regularization, which comprises the following steps: collecting system parameters and system observation quantities, performing spatial regularization on the collected data to obtain a regression model; calculating Euclidean distance of other run points and current run points, and selecting a batch of samples with small Euclidean distance to calculate weight value after sorting, and then stretching weight space to obtain regression model variant, and calculating a sensitivity vectoraccording to cost function, regression model and regression model variant. A system parameter and a system observation quantity are collected by that invention, and the data space is regularized, Distance calculation, sample selection, weight value calculation, weight space stretching, regression calculation and sensitivity vector calculation can solve the error problem caused by ill-conditioned matrix inversion under collinear condition, characterize the system characteristics of power systems in different regions, stabilize numerically, improve the accuracy of sensitivity identification on-line, and can be widely used.

Description

technical field [0001] The invention relates to the intersecting research field of power system identification and data processing, in particular to an on-line identification method of power system state sensitivity based on target regularization. Background technique [0002] The state sensitivity of the power system (hereinafter referred to as sensitivity) refers to the sensitivity between the state variables of the power system, usually expressed as the first-order partial derivative relationship between the power system observations and system parameters, and the observations can be the eigenvalue damping ratio of the system , bus voltage and other variables, the system parameters can be the injected power of each bus and other variables. [0003] Traditionally, the calculation and analysis methods of sensitivity need to rely on the offline mathematical model of the system. Under the condition of strong uncertainty in the future power system, this type of method has prob...

AI Technical Summary

Problems solved by technology

Data collinearity makes the distribution of samples in the high-dimensional data space concentrated in local areas, which cannot fully describe the system characteristics, resulting in the matrix X of the existing locally weighted linear regression method T Numerical instability problems such as WX inverse ill-conditioning, which directly limit the accuracy and application value of online sensitivity identification

Images