Robust Anomaly Recognition in Hydraulic Structural Safety Monitoring: A Methodology Based on Deconfounding Boosted Regression Trees

Analyzing monitoring data to recognize structural anomalies is a typical intelligent application of safety monitoring, which great significance hydraulic engineering operational management. Many regression modeling methods have been developed describe the complex statistical relationships between points, in turn can be used abnormal data. However, existing studies are devoted introducing correlation adjacent response points improve prediction accuracy, ignoring detrimental effects on anomaly recognition, especially pseudo-regression problem. In this paper, an recognition method proposed from perspective causal inference realize best exploitation various types information model construction, including four steps constructing graph, modeling, interpretation, and recognition. stage, two deconfounding machine learning models, two-stage boosted trees copula debiased trees, for recovering correlated points. The validation was carried out with Shanmen River culvert data, experiment results showed that paper has better compared methods, as shown by lower false alarm rates averaged missing under different scenarios.