Electric Power Harmonic Sources Current Estimation Based On Support Vector Machine

With the development of GPS technology, data storage and processing, and communications technologies, power system harmonics measurement system has obtained rapid development and application in China. Harmonic State Estimation Techniques is the use of limited and configured harmonic measurement to estimate the distribution of harmonics in the power grid, so that the harmonics can be monitored in real time. However, due to the high cost of harmonic measurement equipments, large quantity of harmonic information allows an increase in the cost of storage and computing. People hope to use less harmonic measurement configuration to estimate the harmonic current source in power grid. Especially in the case that harmonic measurement is configured for underdetermined, the effect of traditional state-estimation techniques for harmonic source current is poor. To solve this problem, with a previous knowledge of the obtainment of harmonic source position, we use support vector machine to estimate harmonic source current in power system, in order to save cost ensuring the estimated effect.Firstly, this article describes the definition, causes, hazards and influence arising from the impact of power system harmonics. Then the principles and components of the harmonic measurement system in power system are introduced. Following is the description of mathematical model of harmonic state estimation based on the least squares method.Also it talks about the harmonic modeling method of equivalent circuit for each component in the power system.Secondly, it tells the principle of support vector machine. The selection of kernel function and model parameters is analyzed and studied. After selecting the Gaussian radial basis function as kernel function, this article respectively use a grid search method and genetic algorithm based on cross-validation to carry out a parameter optimization among the penalty coefficient C, width parameters σ of Gaussian radial basis kernel function and the parameters ε loss of function, which the model needs to determine.Thirdly, this article establishes a IEEE-14node harmonic network by using MATLAB simulation. In the underdetermined case of the Harmonic measurement configuration, this article based on network topology analysis, by calculating sensitivity factor corresponding to the branch harmonic current versus harmonic injected current, determine the location of harmonic monitoring points and then configure the harmonic measurement. Then it takes advantage of the established IEEE-14nodes harmonic network to obtain a sample data required to modeling. The sample data are normalized, as the relationship between the magnitude and phase angle of the branch harmonic currents in monitoring points and those two of the injected current in the harmonic source is established by support vector machine. In the process of support vector machine modeling, a grid search method and genetic algorithms are respectively used for parameter optimization.Finally, in the case of measurable noise and immeasurable noise, it carries out4harmonic source current estimation modelings by respectively using radial basis function neural network, the support vector machine based on grid search method, the support vector machine based on genetic algorithm and least squares. And four kinds of results are finally analyzed and compared. The result proves that when the harmonic measurement is configured for underdetermined, the support vector machines is effective to the harmonic source current estimation.