Air Conditioning Energy Consumption Prediction Based On SARIMA-GARCH Model

With the development of social economy,Chinese energy demand is constantly increasing.In response to huge energy consumption and damage to environment and ecology,China has proposed the goals of"carbon peak"and"carbon neutral"in order to realize the change from quantity to quality in country’s economic development.The construction industry accounts for a large part of the country’s energy consumption.As the main component of construction industry,it is very necessary to accurately predict the energy consumption of central air conditioning.The data of air conditioning energy consumption has the characteristics of nonlinear,dynamic and complexity,which makes the prediction difficult.In the dissertation,an air conditioning energy consumption prediction model is constructed on the actual operation data of an office building,which provide suggestions for air conditioning industry.The dissertation is arranged as follows.Firstly,the research object of the dissertation is air conditioning operation parameter data of an office building from 8:00 to 17:00 on working days from June 1 to August 31,2020.Because building energy monitoring system is not perfect,such as sensor failure,power outage and other faults cause the monitoring data to have missing values and abnormal values.The dissertation proposes a pre-processing method applicable to building energy consumption monitoring data according to the operational characteristics of air conditioning system,i.e.,using the mean values of two data points before and after to make up missing values,using K-Means clustering algorithm to identify and clean outliers,and using down sample method to align equal interval data and instantaneous data time points.The final result makes the pre-processed data relatively perfect and improves the usability of the data.Secondly,the dissertation investigates an office building air conditioning energy consumption prediction problem from the empirical model and the VAR model in HVAC field,respectively.The MAPE of the empirical model and the VAR model are 11.06%and 11.14%,respectively,and the model prediction effects are not very different.However,both models have the same problem that the prediction effect at the sudden drop in energy consumption from 16:00 to 17:00 each day is not very satisfactory.Although the empirical model and the VAR model can predict the air conditioning energy consumption,they both ignore the trend and cyclical changes of the energy consumption series,so the prediction effect at the point of sudden drop in energy consumption is not good.Finally,based on the actual operation data of air conditioning in an office building,this dissertation uses the SARIMA model and the GARCH model to extract the seasonal autocorrelation information and the residual fluctuation information of the sequences,respectively.Based on the trend and periodicity presented by the energy consumption time series graph,and obtains the air conditioning energy consumption prediction model SARIMA（0,1,1）（1,1,1）₉-GARCH（1,2）.And the MAPE of SARIMA-GARCH model is4.46%.Compared with the previous two models,the SARIMA-GARCH model not only significantly reduces the prediction error at the sudden drop of energy consumption,but also has better prediction effect at other points.Because in real-life,air conditioning energy consumption is affected by holiday effects and major emergencies in addition to climate-related,with random fluctuations.Therefore,this dissertation not only gives precise prediction values for 9 time points per day,but also gives energy consumption prediction intervals at 95%confidence level.The SARIMA-GARCH model constructed in this dissertation is applicable to cyclical change data without considering variable factors and only considering time,which is universal.This dissertation also gives the prediction interval of air conditioning energy consumption,which makes it possible for enterprises to grasp the floating changes of air conditioning energy consumption more precisely and make more flexible energy-saving and emission reduction strategies,which is a guiding role for enterprise energy consumption prediction and building energy saving.