On forecasting demand for electric power with application of artificial neural networks by energy systems of the regions of the russian federation

The calculation of the forecast demand for electric energy by energy systems and complexes of the constituent entities of the Russian Federation is an urgent task. The use of deterministic methods for objects of a similar scale is practically excluded due to the absence or significant incompleteness of the source data. Statistical data available in official sources in an unchanged format is usually presented for a period of 3 - 5 years, which is insufficient for the use of artificial neural networks. The article attempts to study the properties of similar energy systems and complexes. Modern power systems and complexes belong to closed subsystems, the set of elements and connections of which is equivalent to the set of elements of local subsystems of a higher level energy system. This means the inadmissibility of drawing up predictive rules of functioning without taking into account heterogeneous external influences. The system and subsystems are presented as a "black box". Interactions between the system and the external environment and within the system are carried out by the transmission of signals, which are described by a finite set of factors available for analysis and forecasting. The analysis of the possibility of supplementing the general population with statistical data on other objects with a similar structure is carried out. The property of heteromorphism of energy systems and complexes is confirmed. The example of energy systems in the regions of the Russian Federation shows the possibility of a similar approach if non-collinear groups of factors are applied to the analysis. The results of 15 calculations of the most energy-intensive entities of the country are presented, in 28 % of cases the accuracy of forecasted power consumption accuracy is less than 5 %. A further increase in the accuracy of the forecast should develop in the direction of increasing the number of input factors, subject to the condition of the absence of their collinearity and multicollinearity. It is shown that energy systems and complexes of various scales can be described by non-Gaussian stable distributions with infinite dispersion of non-Gaussian distributions, which makes incorrect the use of such methods as the simple extrapolation method, as well as statistical methods based on the assumption that the random distribution law is normal.