气液两相流流型特征无监督提取的卷积自编码器：机理及应用

doi:10.11949/0438-1157.20231038

摘要/Abstract

摘要：

在气液两相流的测量中，流型的准确识别是压降、换热等热工参数测量的基础。传统的两相流方法由于试验条件和数据的局限性使得在不同工况下的适用性有限。人工智能算法可以同时兼顾效率和精度，但特征提取方法仍是识别的难点。流型的准确识别对于解释数据，改进模型，以及提高应用效果方面具有重要意义。因此提出了一种基于卷积自编码器的无监督学习的特征提取方法，将特征提取后分别输入到随机森林，支持向量机以及前馈神经网络分类器中进行分类。实验结果表明，对四种流型的识别精度都达到了99%以上，说明卷积自编码器特征提取方法能显著地提高分类算法的准确率，对不同的分类器具有很好的兼容性，也为今后流型识别的特征提取方法提供了帮助。

关键词: 气液两相流, 气泡, 神经网络, 算法

Abstract:

In two-phase flow measurements, the accurate identification of the flow pattern is the basis for the measurement of pressure drop, heat transfer and other thermal parameters. Traditional two-phase flow methods have limited applicability under different operating conditions due to the limitations of test conditions and data. Artificial intelligence algorithms can take into account both efficiency and accuracy, but the feature extraction method is still the difficulty in its identification. The accurate identification of flow patterns is of great significance for interpreting data, improving models, and improving application effects. Therefore, this paper proposes a feature extraction method based on unsupervised learning of convolutional autoencoder, which inputs the extracted features into random forest, support vector machine, and back propagation neural network classifiers for classification, respectively. The experimental results show that the recognition accuracy for all four stream types reaches more than 99%, which indicates that the convolutional autoencoder feature extraction method can significantly improve the accuracy of the classification algorithm, has good compatibility with different classifiers, and also provides help for the feature extraction method for subsequent popular recognition.

Key words: gas-liquid flow, bubble, neural networks, algorithm

中图分类号:

TQ 028.8

陈思睿, 毕景良, 王雷, 李元媛, 陆规. 气液两相流流型特征无监督提取的卷积自编码器：机理及应用[J]. 化工学报, 2024, 75(3): 847-857.

Sirui CHEN, Jingliang BI, Lei WANG, Yuanyuan LI, Gui LU. Unsupervised-feature extraction of gas-liquid two-phase flow pattern based on convolutional autoencoder: principle and application[J]. CIESC Journal, 2024, 75(3): 847-857.

图/表 23

图1 两层网络结构的自编码器

Fig.1 Autoencoder of two-layer network structure

图2 深层自编码器网络结构

Fig.2 Deep autoencoder network structure

图3 深层自编码器训练过程

Fig.3 Training process of deep autoencoder

图4 卷积自编码器网络结构

Fig.4 Convolutional autoencoder network structure

图5 随机森林网络结构

Fig.5 Random forest network structure

图6 BP分类器网络结构

Fig.6 Network structure of BP classifier

图7 数据预处理过程

Fig.7 Data preprocessing process

图8 流型识别结构

Fig.8 Flow pattern recognition structure

表1 流型的数值表示形式

Table 1 Numerical representation of flow pattern

流型	数值表示
泡状流	[1, 0, 0, 0]
塞状流	[0, 1, 0, 0]
弥散流	[0, 0, 1, 0]
环状流	[0, 0, 0, 1]

图9 卷积自编码器训练损失函数

Fig.9 Autoencoder training loss function

图10 特征提取可视化

Fig.10 Feature extraction visualization

表2 分类器指标

Table 2 Classifier index

指标	概念
精确度（precision）	查准率，表示所有被预测少数类的样本中，真正的少数类所占的比例。精确度越低代表判错了越多的多数类。是将多数类判错后所需付出成本的衡量
召回率（recall）	查全率，表示所有真实为少数类的样本中，被预测正确的样本所占的比例。越低代表捕捉到了越少的少数类
F1分数（F1 score）	精确度和召回率的调和平均

图11 随机森林分类超参数学习曲线

Fig.11 Hyperparameter learning curve of random forest classification

图12 随机森林分类结果

Fig.12 Results of random forest classification

图13 卷积自编码器提取特征后分类结果

Fig.13 Classification results after feature extraction by convolutional autoencoder

表3 卷积自编码器效果评估

Table 3 Evaluation of the effect of convolutional autoencoder

流型类别	Random forest	RF+Con_AE
泡状流	90.16%	98.39%
塞状流	89.09%	98.15%
弥散流	91.34%	100.00%
环状流	95.17%	100.00%
样本个数	252	252

图14 卷积自编码器特征提取效果对比

Fig.14 Comparison of feature extraction effects of convolutional autoencoder

图15 卷积自编码器对流型整体识别效果

Fig.15 Overall recognition effect of convective type on convolutional autoencoder

图16 特征分类器的混淆矩阵

Fig.16 Confusion matrix of feature classifier

表4 不同分类器效果评估

Table 4 Effect evaluation of different classifiers

分类器	查准率/%	查全率/%	F1分数/%	样本个数
random forest	99.09	99.21	99.13	252
SVM	99.21	99.21	99.13	252
BP	99.18	99.21	99.18	252

图17 BP分类器训练准确率

Fig.17 Training loss function of BP classifier

图18 BP分类器训练损失值

Fig.18 Training accuracy of BP classifier

图19 不同分类器对四种流型图片分类结果

Fig. 19 Classification results of four kinds of stream images by different classifiers

参考文献 30

1	Wang T, Liu Z H, Gui M, et al. Flow regime identification of steam-water two-phase flow using optical probes, based on local parameters in vertical tube bundles[J]. Flow Measurement and Instrumentation, 2021, 79: 101928.
2	O’Donovan A, Grimes R. Two-phase flow regime identification through local temperature mapping[J]. Experimental Thermal and Fluid Science, 2020, 115: 110077.
3	Zhang Y C, Azman A N, Xu K W, et al. Two-phase flow regime identification based on the liquid-phase velocity information and machine learning[J]. Experiments in Fluids, 2020, 61(10): 212.
4	Nnabuife S G, Kuang B Y, Whidborne J F, et al. Development of gas-liquid flow regimes identification using a noninvasive ultrasonic sensor, belt-shape features, and convolutional neural network in an S-shaped riser[J]. IEEE Transactions on Cybernetics, 2023, 53(1): 3-17.
5	Li X N, Liu M Y, Ma Y L, et al. Experiments and meso-scale modeling of phase holdups and bubble behavior in gas-liquid-solid mini-fluidized beds[J]. Chemical Engineering Science, 2018, 192: 725-738.
6	周志华. 机器学习[M]. 北京: 清华大学出版社, 2016: 101-108.
	Zhou Z H. Machine Learning[M]. Beijing: Tsinghua University Press, 2016: 101-108.
7	仝卫国, 庞雪纯, 朱赓宏. 基于卷积神经网络的气液两相流流型识别方法[J]. 系统仿真学报, 2021, 33(4): 883-891.
	Tong W G, Pang X C, Zhu G H. Flow pattern identification method of gas-liquid two-phase flow based on convolutional neural network[J]. Journal of System Simulation, 2021, 33(4): 883-891.
8	梁法春, 陈婧, 冉云麒, 等. 基于迁移学习的水平管气液两相流型智能识别[J]. 实验室研究与探索, 2021, 40(7): 1-5.
	Liang F C, Chen J, Ran Y Q, et al. Investigation of transfer learning for horizontal pipe gas-liquid two-phase flow pattern identification[J]. Research and Exploration in Laboratory, 2021, 40(7): 1-5.
9	Chiu P H, Lin Y S, Manie Y C, et al. Intensity and wavelength-division multiplexing fiber sensor interrogation using a combination of autoencoder pre-trained convolution neural network and differential evolution algorithm[J]. IEEE Photonics Journal, 2021, 13(1): 6600709.
10	Xia Z M, Chen Y, Xu C. Multiview PCA: a methodology of feature extraction and dimension reduction for high-order data[J]. IEEE Transactions on Cybernetics, 2022, 52(10): 11068-11080.
11	石延新, 何进荣, 李照奎, 等. 3D卷积自编码器高光谱图像分类模型[J]. 中国图象图形学报, 2021, 26(8): 2021-2036.
	Shi Y X, He J R, Li Z K, et al. Hyperspectral image classification model based on 3D convolutional auto-encoder[J]. Journal of Image and Graphics, 2021, 26(8): 2021-2036.
12	马龙博, 张宏建, 周洪亮, 等. 基于Hilbert-Huang变换和支持向量机的油水两相流流型识别[J]. 化工学报, 2007, 58(3): 617-622.
	Ma L B, Zhang H J, Zhou H L, et al. Flow regime identification of oil-water two-phase flow based on HHT and SVM[J]. Journal of Chemical Industry and Engineering (China), 2007, 58(3): 617-622.
13	周云龙, 孙斌, 陆军. 改进BP神经网络在气液两相流流型识别中的应用[J]. 化工学报, 2005, 56(1): 110-115.
	Zhou Y L, Sun B, Lu J. Application of improved BP neural network in identification of air-water two-phase flow patterns[J]. Journal of Chemical Industry and Engineering (China), 2005, 56(1): 110-115.
14	翁润滢, 孙斌, 赵玉晓, 等. 基于自适应最优核和卷积神经网络的气液两相流流型识别方法[J]. 化工学报, 2018, 69(12): 5065-5072.
	Weng R Y, Sun B, Zhao Y X, et al. Flow pattern recognition method of gas-liquid two-phase flow based on adaptive optimal kernel and convolution neural network[J]. CIESC Journal, 2018, 69(12): 5065-5072.
15	Tan C, Dong F, Wu M M. Identification of gas/liquid two-phase flow regime through ERT-based measurement and feature extraction[J]. Flow Measurement and Instrumentation, 2007, 18(5/6): 255-261.
16	Sarker I H. Deep learning: a comprehensive overview on techniques, taxonomy, applications and research directions[J]. SN Computer Science, 2021, 2(6): 420.
17	Liu W B, Wang Z D, Liu X H, et al. A survey of deep neural network architectures and their applications[J]. Neurocomputing, 2017, 234: 11-26.
18	Breiman L. Random forests[J]. Machine Learning, 2001, 45: 5-32.
19	Chang C C, Lin C J. Training v-support vector regression: theory and algorithms[J]. Neural Computation, 2002, 14(8): 1959-1977.
20	邱锡鹏. 神经网络与深度学习[M]. 北京: 机械工业出版社, 2020: 224-228.
	Qiu X P. Neural Networks and Deep Learning[M]. Beijing: China Machine Press, 2020: 224-228.
21	薛英杰, 陈颀, 周松斌, 等. 基于自监督特征提取的机械异常声音检测[J]. 激光与光电子学进展, 2022, 59(12): 1215013.
	Xue Y J, Chen Q, Zhou S B, et al. Mechanical abnormal sound detection based on self-supervised feature extraction[J]. Laser & Optoelectronics Progress, 2022, 59(12): 1215013..
22	Fan X J, Wang X Y, Jiang M Y, et al. An improved stacked autoencoder for metabolomic data classification[J]. Computational Intelligence and Neuroscience, 2021, 2021: 1051172.
23	Hou L, Luo X Y, Wang Z Y, et al. Representation learning via a semi-supervised stacked distance autoencoder for image classification[J]. Frontiers of Information Technology & Electronic Engineering, 2020, 21(7): 1005-1018.
24	Vincent P, Larochelle H, Lajoie I, et al. Stacked denoising autoencoders: learning useful representations in a deep network with a local denoising criterion[J]. J. Mach. Learn. Res., 2010, 11: 3371-3408.
25	Chauhan R, Ghanshala K K, Joshi R C. Convolutional neural network (CNN) for image detection and recognition[C]//2018 First International Conference on Secure Cyber Computing and Communication (ICSCCC). Piscataway, NJ: IEEE, 2018: 278-282.
26	Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[EB/OL]. .
27	Lalitha R, Tamilselvi D J J. Improved deep convolution neural networks classification amygdala of image mining technique using ASD accuracy[J]. International Journal of Engineering and Advanced Technology, 2019, 8(6s): 796-803.
28	Hassanzadeh T, Essam D, Sarker R. EvoDCNN: an evolutionary deep convolutional neural network for image classification[J]. Neurocomputing, 2022, 488: 271-283.
29	Yu H G, Tao J F, Qin C J, et al. A novel constrained dense convolutional autoencoder and DNN-based semi-supervised method for shield machine tunnel geological formation recognition[J]. Mechanical Systems and Signal Processing, 2022, 165: 108353.
30	Wang J, Liang J Y, Yao K X, et al. Graph convolutional autoencoders with co-learning of graph structure and node attributes[J]. Pattern Recognition, 2022, 121: 108215.

[1]	王娟, 李秀明, 邵炜涛, 丁续, 霍莹, 付连超, 白云宇, 李迪. 多孔板鼓泡塔流动与传质特性数值模拟[J]. 化工学报, 2024, 75(3): 801-814.
[2]	张领先, 刘斌, 邓琳, 任宇航. 基于改进TSO优化Xception的PEMFC故障诊断[J]. 化工学报, 2024, 75(3): 945-955.
[3]	董益斌, 熊敬超, 王敬宇, 汪守康, 王亚飞, 黄群星. 融合激光雷达料位测算的锅炉燃烧优化模型预测控制[J]. 化工学报, 2024, 75(3): 924-935.
[4]	蒙西, 王岩, 孙子健, 乔俊飞. 基于注意力模块化神经网络的城市固废焚烧过程氮氧化物排放预测[J]. 化工学报, 2024, 75(2): 593-603.
[5]	刘志鹏, 赵长颖, 吴睿, 张智昊. 基于水电解制氢的梯度多孔传输层中气液流动可视化实验研究[J]. 化工学报, 2024, 75(2): 520-530.
[6]	詹小斌, 王会彬, 蒋亚龙, 史铁林. 声共振混合器高黏度流体混合的功耗特性研究[J]. 化工学报, 2024, 75(2): 531-542.
[7]	肖拥君, 时兆翀, 万仁, 宋璠, 彭昌军, 刘洪来. 反向传播神经网络用于预测离子液体的自扩散系数[J]. 化工学报, 2024, 75(2): 429-438.
[8]	刘起超, 张世博, 周云龙, 李昱庆, 陈聪, 冉议文. 起伏振动水平管气液两相流型及转变机理[J]. 化工学报, 2024, 75(2): 493-504.
[9]	韩东, 高宁宁, 唐新德, 龚升高, 夏良树. 适用欧拉-拉格朗日方法模拟气液泡状流的气泡破碎模型[J]. 化工学报, 2024, 75(2): 553-565.
[10]	李乃良, 刘常松, 杜雪平, 张一帆, 韩东太. 基于Hurst指数的严重段塞流多尺度分形特性[J]. 化工学报, 2024, 75(2): 484-492.
[11]	崔怡洲, 李成祥, 翟霖晓, 刘束玉, 石孝刚, 高金森, 蓝兴英. 亚毫米气泡和常规尺寸气泡气液两相流流动与传质特性对比[J]. 化工学报, 2024, 75(1): 197-210.
[12]	赵碧丹, 代伊杨, 王军武, 张永民. CFD-DEM-IBM方法探究流化床倾斜挡板内构件受力特性[J]. 化工学报, 2024, 75(1): 255-267.
[13]	尹刚, 钱中友, 曹文琦, 全鹏程, 许亨权, 颜非亚, 王民, 向禹, 向冬梅, 卢剑, 左玉海, 何文, 卢润廷. 基于Adaboost-PSO-SVM的铝电解槽健康状态诊断方法研究[J]. 化工学报, 2024, 75(1): 354-365.
[14]	邵苛苛, 宋孟杰, 江正勇, 张旋, 张龙, 高润淼, 甄泽康. 水平方向上冰中受陷气泡形成和分布实验研究[J]. 化工学报, 2023, 74(S1): 161-164.
[15]	江河, 袁俊飞, 王林, 邢谷雨. 均流腔结构对微细通道内相变流动特性影响的实验研究[J]. 化工学报, 2023, 74(S1): 235-244.