Possibilities of predicting heat consumption in a dwelling by autoregression models

Abstract

The article describes the possibilities of using the ARIMA and XGBoost models to predict heat consumption in dwellings in a multi-family building. Based on the measurement data from the period 2016-2020 of heat consumption in dwellings in two building complexes, ARIMA and XGBoost models were developed to predict heat consumption in monthly periods, and the R environment was used for the calculations. The results are presented in the article for selected apartments in the form of tables and figures. ARIMA models were found to be good, but not effective for rapid changes in single observations. The applications described in the article also require further research. XGBoost is a much more advanced algorithm, and consequently there are many more model parameters that need to be set and optimized later. Therefore, this aspect will be the subject of further research, because despite the expectation of good results, the use of this algorithm did not give much better prediction for rapid changes than the ARIMA models.