The selforganizing neural network approach to load forecasting in the power system

The paper presents the neural network approach to the prediction of 24-hour load consumption in the power system. Two different structures are studied: the conventional Kohonen network and the extended form of self-organizing network taking into account the activity of both winner and neighbouring neurons (so called fuzzy self-organizing network). These methods have been combined with the multilayer perceptron approach to the prediction of mean and variance of the load. Thanks to such solution the average accuracy of the power consumption prediction for the power system may be greatly improved. The method based on self-organization is universal, flexible and easy for use in any power system.