The Method Of The Soft Measurement Based On Mixed Particle Swarm For Oil-Gas-Water Three Phase Flow

The application of the soft measurement technology has increasingly widespread inthe field of multiphase flow research with the development of the modern industry. Atpresent it is a hot issue that the soft measurement of the three-phase flow is achieved by theright estimated model. The flow is a very important parameter in the multiphase flowsystems, and is difficult to measure. Currently, the flow measurement of the three-phaseflow doesn't have a better solution, especially the flow measurement of the oil-gas-waterthree-phase, which is necessary to do further research.Firstly, to solve the problem of the feature redundancy, the feature selection of theoil-gas-water three-phase flow is proposed by introducing the discrete particle swarmalgorithm and the principal component analysis. The input of the algorithm use thestatistical, chaos, fractal analysis and so on, which obtain the29characteristic parametersbeing sensitive to changes in the flow pattern.Secondly, to solve the problem of the parameter selection for the least squares supportvector machine and the local optimal solution that the elementary particle swarmoptimization can easily fall into, the least squares support vector machine based on mixedparticle swarm algorithm is proposed by introducing the genetic algorithm and naturalselection mechanism, and the identification model of the flow pattern is estalished, whichidentifys the two training and prediction samples composed of the different featureselection methods and comparison the two methods.Again, the model is proposed by accounting for the flow pattern for the measurementof the three-phase feature parameter impacted a wide range of flow pattern. The oil flowand the gas flow are predicted in the bubble flow, slug flow and transition flow by usingleast squares support vector machine based on the mixed particle swarm algorithm, usingthe average of the turbine speed, impedance instrument response, gas flow rate and liquidflow rate as the feature vector.Finally, the modeling software is developed by using the Visual C++andMatlabRa7.0mixed programming. The extraction of data feature vector is achieved, flow pattern is classified, flow is predicted.