Anomaly identification of foundation uplift pressures of gravity dams based on DTW and LOF

Abstract

Anomaly can provide valuable information for dam safety monitoring. In this paper, a methodology integrating dynamic time warping and local outlier factor to identify anomalies of time series on various time scales is proposed. The main steps of the methodology are introduced in detail. First, measured time series are preprocessed using moving average and normalization, respectively, to eliminate influence of random and amplitude variations. Following this, time series of independent variables (predictors) are selected as templates to calculate dynamic time warping distances using the global constraint to measure similarities with the dependent variable (a predicant) on a large time scale. Then, traditional regression models are built to find out contributions of independent variables to structural behavior, as well as to highlight unwanted behaviors at relatively on a small time scale. Furthermore, local outlier factor values of multivariate are computed, which helps finding anomalies on a small time scale. Finally, causes of anomalies are analyzed comprehensively. A case study is presented focus on the measured foundation uplift beneath the local riverbed blocks of Xixi reservoir dam. Results show that point anomalies occur with sudden changes of independent variables, but contextual anomalies boil down to coupled effect of high reservoir level and low ambient temperature. The natural condition of the rock foundation underlying blocks, which is characterized as intensively open tension joints, is responsible for uplift anomalies.

Statistical modelling for high arch dam deformation during the initial impoundment period

“…The outlier judgment is performed to determine whether the outlier is a misdetection or a sudden change reflecting the structural behavior evolution. The analysis of outliers can be found in Hu et al 31 Then time series with inconsistent monitoring frequency are interpolated to obtain multivariate time series with a constant measurement step.…”

Zoned safety monitoring model for uplift pressures of concrete dams

“…Similarity measure is of fundamental importance for a variety of time series analyses and data mining tasks (Hu et al, 2018a; Zhen et al, 2013). To measure the similarity between two time series, the most popular approach is to calculate the Euclidean distance on the transformed representation.…”

“…The results show that this technique can efficiently identify dam performance changes with higher flexibility and reliability than simple regression models. Hu et al (2018a) employed local outlier factor values of multivariate to identify anomalies of foundation uplift pressures of gravity dams. For the investigated dam, contextual anomalies boil down to the coupled effect of high reservoir level and low ambient temperature.…”

Dynamic structural health monitoring for concrete gravity dams based on the Bayesian inference

“…Most studies are defined in a deterministic setting and regressive methods are used to calibrate dam predictive models. Different approaches have been proposed to improve the performance of predictive models: the hybrid simplex artificial bee colony algorithm (HSABCA) [22], boosted regression trees [23], multilevel-recursive method [17], dynamic time warping (DTW) method, local outlier factor (LOF) [24], chaotic residual errors [19], Random Forest Regression (RFR) [25] and Bayesian inference [3] have been successfully applied in the scientific literature.…”