Research And Application Of Hybrid Single Variable Time Series Air-conditioning Load Forecasting Methods

The office building air-conditioning load is usually affected by many factors, such as the weather, building envelope, holidays and etc. It is affected by many features and it has strong non-linear feature. The mall building air-conditioning load is more complicated than the office building due to its larger flow of people and the more complex functional areas. So, the prediction methods and theories for different kinds of buildings with high precision become more and more important and urgent.In this paper, the support vector regression is combined with wavelet decomposition algorithm and chaotic time series analysis respectively. Two hybrid air-conditioning prediction methods(Wavelet Decomposition- Support Vector Regression and Chaos-Support Vector Regression) are put forward. The main research of this paper is including:(1) Air-conditioning load for office building has a certain regularity and nonlinear feature. WD-SVR single variable time series prediction method is applied in office building air-conditioning load forecasting. Moreover, different wavelet basises and wavelet decomposition level which affect the prediction accuracy of air-conditioning load prediction results are discussed. The wavelet basis and wavelet decomposition level selection method are proposed. The simulation result shows that Db2 wavelet is suitable for office building WD-SVR single variable time series prediction. The EEP index of WD-SVR method decreases by 33.6% than the SVR method and 29.9% than the BPNN method.(2) Air conditioning load for the mall building has a certain chaotic feature. So, a hybrid Chaos- Support Vector Rgression single variable time series prediction method is proposed. Mutual information method is used for the delay time selection and CAO algorithm for the embedding dimension selection. Largest Lyapunov exponent is applied to determine the chaotic characteristic of the system. Then, the phase state reconstruction and SVR are combined to predict the next air-conditioning load. Simulation results reveal that the EEP index of C-SVR prediction result decreases by 30.6% than SVR prediction results while MBE decreases by 84.2%. The overall prediction accuracy increases to some extent.(3) The WD-SVR method and the C-SVR method are applied in office building and mall building air-conditioning load prediction. The prediction results show that, in office building cooling load prediction, the WD-SVR has the highest prediction accuracy. The EEP index decreases by 3.28% than the C-SVR method and MBE decreases 45.2% than the C-SVR prediction method. In mall building cooling load prediction, the C-SVR has the highest prediction accuracy. The EEP index decreases by 11.7% than the WD-SVR method and MBE decreases 82.4% than the WD-SVR prediction method. The simulation results reveal that when the chaotic characteristic of air-conditioning load is not so strong, the WD-SVR can find out its regular term and random term to obtain better prediction accuracy. But when the air-conditioning load shows strong chaotic feature, the C-SVR method can obtain better prediction accuracy. WD-SVR and C-SVR proposed in this paper both have better prediction accuracy than SVR and BPNN. These two methods are both suitable for the office building and mall building air conditioning load prediction.