Study On Dynamic Short-term Load Forecast Of Office Buildings Based On Meteorological Factors Analysis

In recent years,heating,ventilation and air conditioning(HVAC)systems have become the important elements in office buildings.At present,the operation management level of HVAC systems is generally low and the refrigeration(heat)capacity of the equipment doesn’t match with the actual demand,resulting in the large energy consumption.HVAC systems are responsible for around 40% of the energy use in office buildings,which means the great energy-saving potential.Precise load forecasting is the basis of the optimization of HVAC systems operation,which is conducive to formulate the operation strategy according to the load change and can lay the theoretical foundation for enhancing the thermal comfort and reducing the energy consumption of office buildings.Therefore,this study mainly focused on the research of the dynamic load prediction model based on the analysis of the sensitivity property of the meteorological factors to cooling load and heating load.Firstly,this paper analyzed the influence factors of cooling load and heating load.By DesignBuilder simulation software,the significance of the meteorological factors,such as the dry bulb temperature,relative humidity,radiation,was analyzed via the sensitivity analysis method.And the meteorological factors which had great influence on the load were screened out based on the sensitivity coefficient index.Then,the correlation analysis method is used to further optimize the input of the load prediction model.Secondly,the forecasting models were based on wavelet transform,support vector machines,partial least squares regression.Wavelet transform was used to extract the load features effectively and divide the load into periodic and random components,solving the shortcoming of the general forecasting methods which could hardly reflect the influence of the superposition of different terms.Then the appropriate prediction models are selected on the basis of the distinctive features.For the low frequency component with strong periodicity and linearity,partial least squares regression is used for modeling,which can deal with the multicollinearity problems.And for the high frequency components with strong randomness and nonlinearity,the method of support vector machines is adopted for modeling.The prediction method took full use of the feature extraction ability of wavelet transform,the multiple linear processing ability of partial least squares regression and the nonlinear processing ability of support vector machine,which improved the prediction accuracy and generalization ability.Finally,the validity of the model was verified from two aspects: one is to make the comparison between the load forecasting model established in this paper and other forecasting models;the other is to analyze the influence of weather forecast precision on the proposed model accuracy.