Prediction Research Of Energy Consumption In Public Buildings And Application

With the deepening of domestic industrialization and urbanization,the proportion of building energy consumption in the total social energy consumption is also increasing year by year.Against this background,the building energy saving technologies has received more and more attention.Short-term building energy consumption forecasting is an important work to realize economic operation of public buildings.Thus,research on the forecasting methods to improve the accuracy of building energy consumption prediction is significant.This paper focuses on the forecasting methods of public building energy consumption and air conditioning load.A model based on K-means clustering and wavelet neural networks(WNN)is proposed to predict air conditioning load by analyzing the characteristics of air conditioning load for office building.According to the similarity statistics,K-means clustering method was employed to divide the historical load data into several clusters which could reduce the interference between samples and eliminate the noise in load sample data.Then,the wavelet neural network of the identified cluster was constructed with the training samples and test data set.Experiment results of two examples shown that the proposed model can provide accurate and effective prediction.In order to make the predicting method more expandable,a model based on fuzzy C-means clustering(FCM)and least squares support vector machines(LSSVM)is established to predict air conditioning load after analyzing the shortcomings of the model based on K-means and WNN.FCM can do the same work as K-means clustering,but more flexible.LSSVM has the ability to handle small samples and nonlinear data,which was established for each cluster instead of WNN.Furthermore,gravitational search algorithm(GSA)was used to optimize the parameters of LSSVM model so as to avoid artificial arbitrariness and enhance the predicting performance.Based on the simulated data of an office building from the DeST platform,the predicting method is used to predict the hourly air-conditioning load.The results shown that the FCM-LSSVM-GSM model has a more accurate and stable prediction efficiency compared with LSSVM model and FCM-LSSVM model.Based on reasonably finding out the similar days,the thesis researches a forecasting model which combines the similar day,wavelet transform and LSSVM.The model decomposes the time sequences of the similar day energy consumption in high-frequency and low-frequency part.To the high-frequency part,forecast with LSSVM,and to the low-frequency part,forecast with average method,the final result is the refactoring of these two parts.During the LSSVM modeling process,gravitational search algorithm(GSA)was also applied to optimize the parameters of LSSVM model.Finally,after researching the hourly energy consumption sequences of a college library,the predicted result of W-LSSVM model is compared with LSSVM model.The result shows that the model based on similar day and Wavelet-LSSVM-GSA is more accurate than other models.The key steps in constructing energy consumption monitoring system of an office building in Nanping city is described in this paper.The system has been put into use and it is running well.