Study On Soft Measurement Method Of Gas/Liquid Two-Phase Flow Parameters

Support Vector Machine (SVM) is a novel machine learning method recently. Gauss kernel function is widely used in Support Vector Machine because of its good properties. Model selection in this class of SVMs involves two parameters: the penalty parameter C and the Gauss kernel parameterσ. Appropriate parameters are very crucial to SVM learning results and generalization ability. At present, parameter optimization is also a problem of SVMs application. Chaos optimization algorithm is a global search method of selecting SVMs parameters, simulations show that the proposed method is an effective approach for parameter selection.Based on nonlinear processing technology, such as the attractor morphological dynamic theory, the recurrence quantity analysis theory, the symbolic time series analysis, the complexity measure theory and the linear prediction method from the speech signal processing method, the feature quantities from nonlinear time series analysis and frequency domain are extracted by using the signals of the conductance sensor and differential pressure sensor, also the feature quantities extracted from the time domain ,combining experiment observed flow pattern information and phase volume fraction to compose the data set. A method based on rough sets and support vector machines is applied to two-phase flow parameter soft measurement. Using the rough set reduction algorithm as the pretreatment of data set, it can get rid of redundant attributes of decision tables, and never lose valid information. Then SVMs are used to build flow pattern classification model and phase volume fraction forecast model after rough set reduction. The method can reduce the dimensions of the data set and the complexity of the model of SVMs, and doesn't affect its classification and prediction performance. Applied it to the two-phase flow decision table above, it can obtain good results and provide a new approach to the measurement of two-phase flow parameters. Comparing with every feature quantities data mining method, this paper provides a valid approach for gas/liquid two-phase flow data mining.