Electrical Energy Consumption Prediction is based on the Recurrent Neural Network

Abstract

The results of forecasting of electric energy consumption based on recurrent neural network model. When developing the model, the following dominant factors were taken into account: data on energy consumption over the forecast period; meteorological factors (air temperature, cloudiness, amount of precipitation, wind speed, length of daylight, etc.); date (day, month); data of production calendars (information on the day of the week: weekday / weekend / holiday / shortened); specificity of the industry in the district under consideration (combining statistical information on major centers of federal districts). The factors were selected on the basis of test runs through the neural network of fixed configuration. The relevance of the study is explained by the practical importance of searching for the most accurate methods for predicting the main parameters of the Russian energy market, when a large error of the forecast subject to more expensive tariffs. The constructed recurrent neural network has yielded more accurate prediction results than the widely used mathematical prediction models based on regression dependencies. The obtained scientific result will help to reduce costs and increase the energy efficiency of the electro-energy subjects in the wholesale electric energy and capacity in Russia.