Air Conditioning Load Prediction Model On Support Vector Machine

The process of urbanization is still in the high speed development in 21 st century China.In the public building air-conditioning energy consumption is increased year by year and reached about 60% of building energy consumption. But sharp contradiction has been increased between supply and demand, electricity supply and supply during the day and night hasn't in balance which causes serious waste of electric power industry in China. In order to solve this problem fundamentally, load prediction of air conditioning is needed. It is necessary to accompany the optimization of air conditioning control by changing the supply and demand fundamentally. The ice storage air conditioning is one of the most energy conservation and emissions reduction in the engineering application. Ice storage air conditioning usually adopt control strategy of transferring the peak from valley. It can meet the demand of air conditioning load economically by arranging and distributing ordinary period and peak period of electricity price in proportion to the time reasonably. So the building of load forecasting model is of great significance for building energy saving.The financial office building air conditioning load prediction model of the Beijing chang'an avenue is studied In this paper. First of all, load prediction of model method about the central air conditioning are illuminated. Scientist puts forward the basic algorithm of establish forecasting model--support vector machine(SVM). The construction of the air conditioning load prediction model is established based on statistical theory and the algorithm of support vector machine(SVM) principle in detail. The main studies on the paper is as follows:First we introduce three kinds of improved SVM algorithm(v-SVR,?-SVR and LS-SVR). Prediction model of RBF, linear and Sigmoid kernel function for the common kernel functions of three kinds is analyzed based on the improved algorithm. Prediction model based on the RBF kernel function has better accuracy and generalization ability.Second, confirming the input and output data of the project by combining with the engineering situation,according to the modeling steps. Then we deal with data of the input data and select RBF as the kernel function for establishment of the forecasting model.Parameters of punishing and nuclear g cross validation by K- CV method. Air conditioning load prediction model based on improved algorithm of three kinds is analyzed. We receive conclusion that air conditioning load prediction model based on the LS-SVR has better generalization and learning ability.Air conditioning load forecasting model of BP neural network are compared with load forecasting model based on the LS-SVR of RBF through MATLAB simulation. It is concluded that the LS-SVR based on RBF has high learning efficiency and strong generalization ability by the regression coefficient(R) and error(MSE) which analyzed and summarized. We also conclude that the BP artificial neural network model is affected by the training sample and personal experience. It is easy to fall into local minimum value.