Feature Selection Method Of Flow Regime Based On Distance Evaluation And Rough Set

Gas-liquid two-phase flow is widespread among the modern industrial production, accurate flow pattern identification has great significance. Feature selection is the current field of information technology, in particular, is to identify one of the hot areas of research, some of which have been in practical engineering application and show great potential for development. Therefore, gas-liquid two-phase flow pattern identification of the research not only has important practical value and academic significance, but also provides a strong technical support for related industrial production equipment, security, economic design and operation of operation.In this paper, gas - liquid two-phase flow of the comprehensive experimental stage were tested, and the level of tubes collected air - water two-phase flow of the differential pressure fluctuation signals were analyzed. In order to overcome many issues not related to extracted features of gas-liquid two-phase flow characteristics, respectively, from the assessment and proposed based on rough set theory, feature selection method. First, the use of de-noising filter, empirical mode decomposition (EMD), wavelet packet techniques, recurrence quantification analysis (RQA) such as pretreatment technology for mining the characteristics of the signal flow patterns of information, feature extraction, feature fusion to form sets. Assessment method using the distance-sensitive feature selection on the feature set, and then enters the sensitive features of support vector machine classification. Similarly, based on rough set theory to feature-set reduction, in the optimal decision based on the use of the property least squares support vector machine classifier to identify the flow pattern.Horizontal tube gas - liquid two-phase flow of the four kinds of typical flow pattern identification results show that the recognition rate is slightly higher than the rough set method from the assessment method, but the dimension is higher than the latter, the computation is slightly larger, so in the two kinds of methods there are pros and cons. This is two kinds of methods sensitive to features extracted, most of the same features, basically the same, were able to accurately obtain the flow pattern of the sensitive characteristics of information, reduce the size of the operation, improve the classification accuracy obtained satisfactory recognition results, evaluation criteria of characteristics of flow pattern was established, and gives the corresponding algorithm.