Forecasting Peak Hours for Energy Consumption in Regional Power Systems

Abstract

. Electrical power is the second most important commodity in electrical energy markets. For consumers, the charged amount of “generator” power is determined as the average value of hourly consumption amounts on working days during peak hours in the region. The cost of power in some regions can reach 40 % of the final tariff, so reducing the load during peak hours by 10 % can lead to a decrease in monthly consumer payments by 3 %. However, such a way of saving money is not available to the consumer since the commercial operator of the wholesale market of electricity and capacity publishes the peak hours of the regions after the 10th day of the next month, when this information is no longer relevant. Timely forecasting of peak hours will make it possible, on the one hand, to reduce consumer costs for payments for electric power, and on the other hand, to smooth out the daily schedule of electric load of the power system, thereby optimizing the operation of generating equipment of stations and networks of the system operator. The article presents a study of the effectiveness of machine learning methods in the context of forecasting the peak hour of a regional power system. The study concerns the period from November 2011 to October 2023, covers 76 regions of the Russian Federation, including subjects of price (1st and 2nd) and non-price zones and includes 10 machine-learning methods. The results of the study showed that statistically, the K-nearest neighbors clustering method turns out to be the most accurate, although not universal. Support Vector Classifier and Decision Tree Classifier have demonstrated high efficiency (in terms of accuracy and speed). The study also refuted the assumption that the closest data in terms of time series has the greatest value in predicting peak hours.