Analysis and evaluation of the operation data for achieving an on-demand heating consumption prediction model of district heating substation

Abstract Accurate heating consumption prediction is the key to achieve on-demand heating in a heating system. A prediction model needs to go through the steps of data cleaning, model establishment and effect evaluation before being applied. Currently, there is no universal supplement method for missing data, and the evaluation method is based on the error between the predicted and the actual value, if the actual value is not on-demand value, the method cannot accurately evaluate the prediction model which may lead to the poor applicability of the model. In this paper, firstly, the distribution characteristics of heating consumption data are analyzed, and the mean method based on eighth fractile judgment is proposed to supplement the missing data. Secondly, the research method and empirical formula of solar radiation intensity and the percentage for heating consumption reduction is determined by analyzing the change of indoor temperature. Finally, the application effects of SVM, MLR and LR in heating consumption prediction are evaluated from the perspective of on-demand heating, for the daily prediction, LR, SVM and MLR can be well used when the average outdoor temperature fluctuation is small, and the applicability of SVM and MLR are poor when the fluctuation is large, for the hourly prediction, the applicability of SVM and MLR are poor due to influence of improper manual regulation, and the daily heating consumption distribution method is proposed. In summary, the methods of data cleaning, daily prediction, hourly prediction and model modification are proposed, which have a strong applicability and provide theoretical support for the practical application of heating consumption prediction in heating substation.