Air Conditioning Load Prediction Algorithm Based On Support Vector Machine

Chilled water storage air conditioning system storage chilled water at the low price periods in the evening, and release it during peak periods. In order to precisely controlling the storage tank to storage and release chilled water, we need an effective conditioning load predict means to forecast the next day’s cooling load. through this predict means, the storage tank knows how much cooling load the next day will need, and we can set the conditioning system running in the best procedure at the next day. Due to this means, the user can maximize saving electricity bills and electricity. Limit by the accuracy, speed and operability, the existing prediction algorithm often difficult to meet the need of conditioning system’s energy-optimized operation and control. Face with this situation, this paper will use SVM to build a rolling prediction model and apply it in the optimal control of the chilled water storage system, for the powerful nonlinear mapping ability of SVM. The main work in this paper is as follows.(1)In order to improve the prediction accuracy, the paper divides the prediction into two processes. The temperature of the next day will predict at first, then the predicted data will be used to predict cooling load. And for every hour of next day, the paper will build a LS-SVM prediction model, this means can reduce the output parameter to improve the prediction accuracy.(2)In the algorithm of forecast outdoor temperature,the paper improves the algorithm which through exponential smoothing to improve the traditional ASHRAE method. This improved algorithm uses hourly and day correction to update the temperature coefficient and hourly correct the predicted temperature. In which the algorithm bases on weight coefficient of time and temperature forecasted error to improve predicted temperature and temperature coefficient, and through exponential smoothing to update daily temperature coefficient.(3)Also in order to improve the prediction accuracy, this paper will adopt rolling and weighed algorithm to optimize each LS-SVM prediction model. Through rolling algorithm, the prediction model can revise and update itself with every passing hour. Trough weighed algorithm, this model can enhance the effect to it of recent data.