Deep Learning-Based Diagnosing Structural Behavior in Dam Safety Monitoring System

Abstract

Collecting a myriad of prototype data through various types of monitoring sensors plays a virtual important role in many aspects of dam safety such as real-time grasp of safety state, exposure of hidden dangers, and inspection design and construction. However, the current methods of prediction are weak in the long-term sequence of nodes with missing and abnormal error value. Moreover, the limitation caused by the apparatus, environmental factors, and network transmission can lead to the deviation and inconsistency of diagnosis and evaluation of local region. In this paper, we consider the correlation of data on nodes in the entire monitoring network. To avoid the deviation caused by noise and missing value in the single-node data sequence, we calculate the correlation between the multiple sequences. A single-node assessment model based on multiple relevant sequence (SAM) is proposed to improve the accuracy of single node assessment. Given the different nodes of a local region have varying impacts on the evaluation results, a local region evaluation algorithm based on node credibility (LREA) is presented to model the credibility of nodes in order to alleviate inconsistent evaluation results in the local region of dam. LREA can assess the dam’s operation state by considering the variations in credibility and multiple nodes coordination. The experimental results illustrate the LREA can reveal the trends of the monitoring values change in a timely and accurate way, which can elevate the accuracy of evaluation results of dam safety.