A Neural Network Based ATC Assessment Incorporating Novel Feature Selection and Extraction Methods

Abstract

This article presents a neural network based method for assessing the inter-area available transfer capability (ATC) of power systems aimed at providing a fast and accurate ATC. In the neural network implementation, the Levenberg-Marquardt modified back propagation algorithm is used in the training of the neural network so as to improve the speed and the convergence in the training process. One of the important considerations in applying neural network to transfer capability assessment is the proper selection and extraction of neural network input features. To achieve this, a hybrid method consisting of both the sensitivity and discrete Fourier transform methods are used in which the sensitivity analysis is first used in selecting the input features and then followed by the discrete Fourier transform method for extracting the meaningful features. To illustrate the effectiveness of the proposed methods, ATC simulations have been performed on the Malaysian power system. Neural network results shows that better ATC assessment accuracy can be obtained by selecting and extracting features using the proposed methods as compared to using only the discrete Fourier transform or the sensitivity methods. Results have also shown that the method is capable of reflecting accurate variations in load levels and effect of contingencies such as line outages. Computational time can be greatly reduced by using the neural network based ATC assessment method and, therefore, it can be used to provide a real time market signal of the capability of a transmission system in delivering power.

Keywords:

• neural network
• transfer capability
• transmission systems