Interval early warning method for state of engineering structures based on structural health monitoring data

Abstract

In order to carry out early warning for the abnormal state of engineering structures in time and optimize early warning interval (WI), a new construction of WI method based on relevance vector machine (RVM) and particle swarm optimization (PSO) under the framework of lower upper bound estimation (LUBE) is proposed. First, extract time-domain features of structural monitoring data and then two RVM models are trained using time-domain features to construct lower and upper bound of WI. Kernel parameters of RVM are optimized by PSO algorithm to minimize the loss function of LUBE and obtain the best optimal evaluation indices of WI. The LUBE based on PSO-RVM interval warning method is verified through a numerical simulation and a case study on structural monitoring data of quayside container crane (QCC), and the performance of WI is compared with other prediction methods. The results indicate that the proposed construction of WI method can effectively improve the quality of WI, realize early warning damage, and provide better performance for early WI.

Interval Prediction Model of Deformation Behavior for Dam Safety during Long-Term Operation Using Bootstrap-GBDT

“…Te uncertainties caused by the environmental factors and subjective assumptions when establishing prediction models can be classifed as cognitive as the former. Te latter is usually caused by random noises in the measured time series [22,25]. Hence, the present work is a contribution to the quantifcation of these uncertainties for dams during longterm operation.…”

“…where y l (X i ) is the prediction result of the l th model and y(X i ) is the predicted result of the B models, i.e., the point prediction result. Te noise variance can be estimated by equation (25). In the residual dataset D r 2 � (X i , r 2 (X i )), i � 1, 2, .…”

“…Te variance of the model errors σ 2 ∆ (X i ) can be obtained by the B GBDT model in equation (24). Te variance of the random noise σ 2 ε (X i ) can be estimated by training the B + 1 st GBDTmodel in equation (25). Te 1 − α/2 percentile p and q can be calculated by the KDA in equations ( 20) and (21).…”

Machine Learning‐Based Distraction‐Free Method for Measuring the Optical Displacement of Long‐Span Bridge Structures

“…Among the existing machine learning algorithms, relevance vector machine (RVM) can minimize the regression error and exhibits a high generalization and antinoise disturbance abilities. Moreover, the RVM can adapt to the characteristics of nonlinear time sequences of bridge health monitoring system data and exploit the correlation between the data and selected training samples to predict the missing data or correct the abnormal data [37]. Optical methods can be used to synchronously monitor multiple measurement points and provide data support for the model training of RVM.…”