The Load Model Composition Method in Power Systems Using Artificial Neural Network

Abstract

An accurate load model is necessary to improve the accuracy of power systems dynamic stability analysis. Recent studies to estimate accurate load model have mainly been focused on the measurement-based load modeling and the method to estimate a single representative load model called load model composition or load representation. The current load model used for the dynamic stability analysis in South Korea power system was aggregated with the measurement data from only three distribution stations. The proposed algorithm is the load model composition method based on artificial neural network technique using more measurement data than the algorithm to estimate the current load model. The proposed load model composition method using the artificial neural network uses the load composition ratio as the input value and the ZIP model parameter, which is the estimation result at each distribution line, as the output value. The measurement data to estimate the parameter were collected from 105 distribution lines of 9 substations. The performance of proposed algorithm was verified for three perspectives in the case studies. The accuracy of the proposed algorithm has been improved by comparing the measurement data with the calculated result by using the proposed model. The mean absolute percentage error (MAPE) between the measured data and the proposed model is 1.7%. The proposed algorithm could be applied to estimate the representative load model using data measured at multiple measurement points.