District Heating System Load Prediction Using Machine Learning Method

Abstract

Accurate prediction of heating load can help improve operational efficiency of district heating systems (DHSs). The selection of feature variables is of great significance to prediction performance. Most existing methods only use the meteorological data and historical thermal demand data. In this study, correlation analysis method is employed to analyze predominant variables affecting prediction accuracy. The correlation of supply/return temperature, outdoor temperature, and historical load data were examined. The obtained results were used to select minimal input variables subset so as to avoid multiple input variables. The extreme learning machine (ELM) was used to predict the energy consumption of the next 6, 12, and 24 h. The approach was adopted to predict heating load of a DHS in Changchun, China. Historical heating load data were proved to be the most essential prediction inputs. The results show that the root-mean-square error predicted by the ELM model can reach 4.1%.