Hydrate decomposition

During gas hydrate extraction, insufficient heating or depressurization can lead to incomplete hydrate decomposition, resultingin lower-than-expected actual gas production. In this paper, a method for assessing the degree of hydrate decomposition based on an electronic nosesystem is proposed. First, the electronic nose system was developed tocollect the information of gas produced by gas hydrate decompositionunder different humidity conditions. Then, a Convolutional NeuralNetwork combined with Domain Adaptive Compensation featureextractor (CNN-DAC) was used to extract moisture-insensitive features, which can improve the humidity generalization ability of the assessment model. Finally, a CNN-DAC-RF model by combining CNN-DAC withrandom forest (RF) was proposed, which can accurately assess the hydrate decomposition degree level. The experimental results show that the accuracy of the model reached 98.67%. In the comparison experiments with other feature extraction methods, the classification accuracy ofCNN-DAC-RF was improved by 1.32% in the source domain (high humidity data), and by 42.49% in the target domain (lowhumidity data). In summary, the combination of CNN-DAC-RF and electronic nose provides a reliable technical means for the assessment of the degree of hydrate decomposition during hydrate mining.

在天然气水合物提取过程中，若加热不足或降压不当，将导致水合物分解不完全，进而造成实际天然气产量低于预期。本文提出了一种基于电子鼻系统的水合物分解程度评估方法。首先，开发了一套电子鼻系统，用以收集不同湿度条件下水合物分解产生的气体信息。随后，采用卷积神经网络（CNN）结合领域自适应补偿特征提取器（CNN-DAC）提取对湿度不敏感的特征，从而提升评估模型的湿度泛化能力。最终，提出了一种结合CNN-DAC与随机森林（RF）的CNN-DAC-RF模型，该模型能够准确评估水合物分解程度等级。实验结果表明，该模型的准确率达到了98.67%。在与其他特征提取方法的比较实验中，CNN-DAC-RF在源域（高湿度数据）的分类准确率提升了1.32%，在目标域（低湿度数据）的分类准确率提升了42.49%。总之，CNN-DAC-RF与电子鼻的结合为水合物开采过程中水合物分解程度的评估提供了一种可靠的技术手段。