小通道气液两相流流型辨识与LSTM短期预测研究

Pattern Identification and Prediction of Air-Water Flow in Small Channel with LSTM Recurrent Neural Network

以空气与水为工质,运用压差传感器与光学位置传感器和高速摄像机对水力直径3.0 mm的小通道水平圆管内气液两相流进行实验研究。根据压差波动信号图、光电传感器模拟信号图及高速摄像机拍摄所得流型图对小通道内气液两相流进行流型辨识,辨识结果表明:存在环状流、层状流、间歇流、以及段塞流四种流型;对四种流型所对应压差波动信号进行LSTM循环神经网络的分析预测,结果表明:LSTM循环神经网络预测模型可实现小通道气液两相流压差信号的在线预测,且预测结果精确,四种流型的均方误差分别为0.004、0.0099、0.0075、0.0156。

The air-water flow in a horizontal ring of small organic-glass tube,3 mm in hydraulic diameter,was investigated with the lab-built test platform,comprising pressure-difference sensor,photoelectric position sensor,high-speed video camera and host computer.Four distinctive flow patterns,including the annular,layered,intermittent and slug flow-patterns.In addition,the fluctuation signals of pressure difference,involving the four flow patterns,were analyzed and predicted in a short term with the model of Long and Short Term Memory(LSTM)recurrent neural network.The results show that when it comes to four flow-patterns,on line prediction with LSTM recurrent neural network model was relatively accurate.To be specific,for the annular/layered/intermittent/slug flow-patterns,the mean-square errors were estimated to be 0.004,0.0099,0.0075 and 0.0156,respectively.