Short-Term Forecast of Electricity Load for LLC 'Omsk Energy Retail Company' Using Neural Network

摘要

According to rules of interaction between the subject of the Electricity Market and JSC ATS, subjects of the Electricity Market are obliged to implement the daily hourly forecast in the mode 'for a days ahead'. To ensure of high-quality prediction of the electricity loads, subjects of electricity market need to prepare the regulatory base, to develop a technique of creation of the forecast of the electricity loads, and also to count the risks connected to the accuracy of the used models. On the one hand, the complexity of the problem being solved is characterized by the availability of data on the supply points, since not always the subject of the electricity market has the opportunity to collect data on the consumption of individual power facilities in the hourly mode. From the other hand, the introduction of commercial accounting systems can solve this problem with the investment of a large investment in the installation automatic system for commercial measurement of the electricity loads (ascme), but as a rule subject of electricity market goes for such long-term payback costs. The work can be useful both to specialists of power sales companies who are engaged in building forecast models, as well as to specialists of the electricity market entities, who carry out forecasts for the electricity market for the day-ahead. The main aim of the study is applying methodology forecasting using neural network for building predictive models for LLC 'Omsk Energy Retail Company'. The methods used in the study: Holt-Winters model, the ARIMA, neural networks, temperature and wind index. The results. It considered methods of construction of predictive models, the path of their evolution since the launch of Electricity market. Method of constructing the forecast of 'Omsk Energy Retail Company' was developed using neural network, taking into account the temperature and wind index and allocation of common types of days by electricity load.