Research On Ensemble Support Vector Regression And Its Prediciton Of Flooding Velocity In Packed Tower

The flooding velocity is the limit of the operating conditions in packed towers. In order to guarantee the stable operation and keep high efficiency production in packed columns, it is necessary to manipulate the gas velocity suitably. Consequently, it is the very important to construct a model which can predict the related flooding velocity accurately.After the overview of the traditional methods of flooding velocity prediction in packed towers, the least squares support vector regression (LSSVR) is presented for predicting the flooding velocity in this thesis. In order to improve the prediction performance, several improved LSSVR models based on the ensemble learning theory are proposed to predict the flooding velocity in the thesis.The main contributions in this thesis are as follows:(1) Traditional empirical models for flooding velocity prediction often show a lack of generalization because the packing constants needed in these traditional models are difficult to obtain. To overcome this problem, the LSSVR model is proposed to predict the flooding velocity. Compared to the traditional methods and the neural network models, the LSSVR model shows better prediction performance and generalization ability.(2) Single models are often difficult to achieve a satisfied prediction performance because of the diversity and imbalance of the packing data in industrial processes. To overcome the problem, an ensemble LSSVR modeling method combined with fuzzy c-means clustering (FCM) algorithm is proposed for predicting the flooding velocity. This integrated model can extract the feature information of flooding data effectively. Thus, improves the prediction performance can be further improved. The obtained results of the flooding velocity modeling experiment for several kinds of packing indicate that the proposed ensemble LSSVR method obtains better and more reliable prediction performance, compared with other methods.(3) To overcome the disadvantage of the Euclidean distance in FCM clustering methods, a new clustering method integrated with the fuzzy set theory is used to construct the LSSVR ensemble model. The proposed clustering method combines both the advantage of the FCM clustering algorithm and the Gaussian mixture model. Consequently, the new LSSVR ensemble model is presented to predict the flooding velocity. And the predicted results show that this novel LSSVR ensemble model can achieve better performance in terms of higher prediction accuracy and reliability, compared with other two ensemble models.