Neural network algorithm for short-term forecasting of electricity consumption of agricultural producers

Abstract

Under the conditions of a market electric power industry, the problem of short-term forecasting of hourly electricity consumption becomes by much urgent. The purchase of electricity by guaranteeing suppliers on the wholesale electricity and capacity market involves forecasting their hourly electricity consumption for the next day for all groups of electricity supply points. Both the financial results of the guaranteeing supplier and the price of electricity for the end consumers depend on the accuracy of this forecast. The time series of hourly electricity consumption of a group of delivery points of a guarantee supplier, including agricultural producers, is a multifactorial functional dependence. Short-term forecasting of a given time series is a complex, poorly formalized task. Currently, when solving practical problems of predicting electrical loads, preference is given to methods based on deep convolution networks, recurrent neural networks, as well as ensembles consisting of several neural networks. The most common way to obtain the final output signal is an ensemble averaging block. This paper discusses the problem of increasing the accuracy of short-term forecasting of electricity consumption of the electrical complex of regional electrical networks using deep machine learning tools. The effectiveness of using an adaptive algorithm for training deep neural networks in short-term forecasting of power consumption of a given electrical complex was studied. The issues related to the use of convolutional and recurrent neural networks to solve the problem of predicting electrical loads are investigated. A comparative analysis of the accuracy of the short-term forecast of electricity consumption of gas-turbine gas production units, including large agricultural producers, obtained using the ensemble neural network method and single neural networks was made.