Research on Deep Learning Reconstruction Method for Oil and Gas Drilling Sensor Measurement Data With High Missing Rate

With the extension of oil and gas drilling to deep and complex strata, the drilling sensor network is prone to data quality problems, such as large-scale loss of measurement values and coexistence of heterogeneity in the environment of high temperature, high pressure, and multiphase flow coupling, resulting in a significant reduction in the data reliability of the drilling measurement system. Traditional interpolation methods are difficult to handle effectively due to ignoring the global distribution characteristics and dynamic correlations of measurement data. This article proposes a measurement value reconstruction model of pseudolabel conditional generative Sinkhorn adversarial interpolation network (PC-GSAIN) hierarchical density-based spatial clustering of applications with noise (HDBSCAN) based on semisupervised pseudolabel learning. This model combines density clustering and generative adversarial networks (GANs) to construct a data repair framework: The HDBSCAN algorithm is utilized to extract the multiscale spatial structure of missing data and establish an adversarial training mechanism under pseudolabel constraints. In the design of the generator, the Sinkhorn Distance is introduced as a quantification index for the alignment of the measurement distribution. Combined with the entropy regularization optimal transport (OT) theory, a loss function is constructed to achieve the optimization of the distribution consistency between the repaired data and the original measurement values. Experiments using two real sensor measurement datasets show that the PC-GSAIN (HDBSCAN) model solved by the Sinkhorn-Knopp algorithm outperforms the baseline model within the missing rate range of 10%–60%. While maintaining robustness, it significantly improves the quality of data reconstruction, providing a new method for the construction of highly reliable measurement systems in complex industrial environments.