Selecting the Forecasting Model of Heat Energy Delivery

Abstract

To solve the problems of saving thermal energy, instead of the currently used cumbersome methods or complex mathematical models, for example, in the form of partial differential equations, it is proposed to use simpler models of time series for the supply and production of thermal energy. The data of a specific heat supply organization on a city scale and the website of the Ministry of Energy of the Russian Federation were used as initial data. The proposed approach is based on a comparison of various mathematical models for predicting the production and supply of heat energy according to the criterion of adequacy (average relative error and coefficient of determination). The results of research carried out using the Matlab software package are presented. A substantiated comparison of the possibility of using different models is carried out. Models have been developed that take into account the seasonal component and higher accuracy. The possibility of using the following time series models is investigated: regression models with a seasonal component, triple exponential smoothing Holt–Winters, multiplicative seasonal autoregression model of the integrated moving average SARIMA (seasonal autoregression integrated moving average), and nonparametric model SSA (singular spectrum analysis). Multiplicative models containing a trend in the form of a linear or autoregression model with the calculation of the seasonality index, as well as additive models with a trend in the form of a linear or autoregression model with a seasonal component in the form of a Fourier series or sum of sines, were compared. In the Holt–Winters model, fminsearch, a function of the Matlab software package, was used to determine the tuning parameters. In the multiplicative SARIMA model, the orders of autoregression and moving average were fitted according to the autocorrelation and partial autocorrelation function. The nonparametric SSA model was used for comparison with the listed parametric models. An opportunity has been obtained to make reasonable forecasts of the production and supply of heat energy It is shown that the best results (average relative error 6.52%) were obtained using the SARIMA multiplicative model.