The use of artificial intelligence to predict electric power consumption of a power supply company

Abstract

The existing approaches to using artificial intelligence in training the neural network using the Neurosimulator 5.0 application to predict electricity consumption according to the data of the previous period are analyzed in this article. It is also concluded that it is advisable to develop this direction of calculations for forecasting and designing power supply systems. The article is devoted to the problem of choosing a model for forecasting electricity consumption when solving the problem of operational daily planning of electricity supplies in the wholesale market. The task of forecasting electricity consumption acquired particular relevance after the emergence of the wholesale electricity market: an underestimation of the forecast leads to the need to launch expensive emergency power plants, while an overestimation leads to an increase in the costs of maintaining excess capacity. The choice of artificial neural networks for this purpose is well-founded. The most suitable architecture of an artificial neural network for solving the problem in question is a multilayer perceptron containing several layers of neurons: an input layer, one or more hidden layers and a layer of output neurons. The transmission of information usually takes place in one direction - from the input layer to the output layer. An example of power consumption prediction based on the results of the nearest measurements in the time domain is considered and an approximation error is determined. The results of approximation and prediction of power consumption showed that a root-mean-square relative error did not exceed 6.32 %, but there is an outlier at one point up to 34 %. The reserve for improving the forecast accuracy is to study the influence of additional factors such as an ambient temperature and the day factor which takes into account the load distribution by the days of the week.