Abstract:

The double-wavelength transmission method is a high precision method to detect the phase distribution parameters of multiphase flow. In order to improve the accuracy of phase distribution measurement, the gradient bosting decision tree (GBDT) algorithm was proposed in this paper. Optical simulation was used to simulate the propagation process of bubble light with wavelength of 445nm and 635nm in gas-liquid two-phase flow under different phase distribution. The light intensity distribution of detection plane was collected and the influence of bubble center location and bubble radius on the light intensity distribution curve under different phase distribution was analyzed. The measurement model was developed with GBDT, and the corresponding database between the characteristic parameters and the bubble phase distribution was established with the missing width and missing offset of light intensity as the characteristic parameters. A trained model was used to predict the phase distribution of bubbles. The mean square error of the measurement model was less than 0.0008mm, and the mean square error was reduced by 33.33%, which proved that the measurement model was more suitable for the measurement of the phase distribution. An experimental platform was built to predict the flow phase distribution parameters of vertically rising bubbles, and the movement trajectory of the central position of bubbles was tracked.

Key words: gas-liquid flow, bubble, riser, phase distribution, gradient bosting decision tree (GBDT)

摘要：

双波长透射法是一种高精度的多相流相分布参数检测方法，为了进一步提高相分布测量的精度，本文提出用梯度提升决策树（GBDT）算法建立相分布测量模型。利用光学仿真，模拟气泡在不同相分布下波长为445nm和635nm的光线在气液两相流中的传播过程，采集检测平面的光强分布，分析不同相分布情况下气泡中心位置和气泡半径对光强分布曲线的影响；用GBDT建立测量模型，以光强缺失宽度和缺失偏移为特征参数，建立特征参数与气泡相分布之间的对应数据库，利用数据库对GBDT模型进行训练；用训练好的模型预测气泡的相分布。建立的测量模型对气泡相分布预测的均方误差小于0.0008mm，均方误差减小了33.33%，证明了测量模型更适用于相分布的测量。搭建了实验平台，对竖直上升气泡流相分布参数进行预测，实现了对气泡中心位置运动轨迹的追踪。

关键词: 气液两相流, 气泡, 上升管, 相分布, 梯度提升决策树