Research on heat consumption detection, restoration and prediction methods for discontinuous heating substation

Under the goal of low-carbon heating, the control modes of heating substation will develop from traditional continuous heating to discontinuous heating, indicating that the existing methods for detection, restoration and prediction are not applicable. Based on data-knowledge analysis, this paper conducts a systematic study. Firstly, the associated parameters of the control mode were determined according to professional knowledge, including date, daily heat consumption and standard deviation, and the control mode of discontinuous heating substation was successfully identified by K-means. Secondary, by analyzing the internal heating parameters affecting the changing of heat consumption, the detection methods for different types of abnormal data are proposed. Thirdly, the restoration effect of statistics methods and Data-driven methods on abnormal data are compared, and data-driven methods with input parameters of secondary supply temperature, return temperature and temperature difference can successfully restore different abnormal heat consumption with an average error of less than 5%. Finally, the input parameters for heat consumption prediction of discontinuous heating stations, including outdoor temperature, time point and secondary return temperature were determined, and extreme gradient boosting had a higher prediction accuracy than support vector machine and multiple linear regression, with R2 >0.85. •The characteristics of different heating control models are analyzed.•The identification methods of different heating control models are proposed.•The detection methods for 0 data, explicit and implicit abnormal data are proposed.•Machine learning methods can restore abnormal data successfully.