Abstract: To improve the standard orifice plate design a better outflow characteristics of tapered orifice. Using tapered orifice differential pressure fluctuation characteristics of the empirical formula and Least Square Support Vector Machine （mixed kernel） which is improved by particle swarm optimization establish the soft sensing model of two-phase flow dryness. The mixed kernel of this model is constituted of radial basis function and polynomial kernel，it can be both the local fitting ability and global fitting ability of Least Square Support Vector Machine. What’s more，by optimizing parameters of the Least Square Support Vector Machine，this model can reduce the dependence on the initial point and the number of samples. The experimental results show that such a prediction method is feasible than not optimized least squares support vector machine model and the optimized mononuclear least squares vector machine model generalization ability. So，it can be widely used in engineering practice.

Key words: gas-liquid two phase flow, dryness, particle swarm option, least squares support vector machine

摘要： 对标准孔板进行改进，设计了一种流出特性较好的锥形孔板。将锥形孔板压差脉动特性的经验公式与经过粒子群算法优化的混核最小二乘向量机模型结合，建立了气液两相流干度的软测量模型。该模型的优点是采取径向基核函数和多项式核函数的混合核函数兼顾了局部和全局的拟合能力，且通过对传统最小二乘向量机模型的参数优化，可减少对初始点和样本数量的依赖。实验结果表明，该预测方法可行，且比未经过优化的最小二乘支持向量机模型和经过优化的单核最小二乘向量机模型的泛化能力强，具有实际工程意义。

关键词: 气液两相流, 干度, 粒子群算法, 最小二乘支持向量机