Study On Influencing Factors And Forecasting Methods Of Heating Load In Central Heating System

With the development of chinese central heating,heat metering technology and digital operation of heating systems are gradually applied."Heat supply on demand" is becoming the focus of research in the field of central heating.The heat load is an important indicator for the operation and control of the heating system,and its accurate prediction can guide the production and distribution of heat,and achieve the purpose of energy saving while ensuring the heating effect of users.Therefore,how to achieve accurate prediction of heat load is of great significance for improving heating quality and achieving energy saving.This paper takes a central heating system in Harbin as the research object,and based on the historical operating data of each thermal station,carries out a statistical analysis of the factors affecting the heat load;compares the advantages and disadvantages of using various traditional load forecasting methods and machine learning algorithms to establish load forecasting models;In addition,the heat load forecast period is divided to explore the influence of staged modeling on the accuracy of model forecasting.First of all,using correlation analysis,variance analysis to explore the influence of weather factors,building factors and heating system factors on the heat load of the heating station.The results show that weather factors such as outdoor temperature,solar radiation intensity,rainfall and snow all determine the heat load of the building.In addition,the construction period and function of the heating building,and the type of equipment or system used in the heating system can also directly or indirectly affect the heat consumption.Secondly,use multivariable linear regression,BP neural network,genetic algorithm to optimize BP neural network,and grid search to optimize support vector regression to establish heat load forecasting models.Analyze the forecast results,compare the forecast accuracy of different methods,and evaluate its applicability.The results show that the error of each prediction method is as follows: BP neural network> multivariable linear regression> BP neural network optimized by genetic algorithm> grid search optimized support vector regression.Finally,in order to improve the prediction accuracy,the model prediction period is divided.The outdoor temperature series are processed with moving averages of different time scales.In order to divide the heating season into two periods,the outdoor calculated temperatures are clustered.On this basis,the heat load forecasting models for different periods are established based on different forecasting methods.The results show that the model accuracy of staged forecasting has improved compared with the previous ones.