Dam anomaly assessment based on sequential variational autoencoder and evidence theory

Considering the multi-sources, heterogeneity and complexity of dam anomaly assessment, a novel anomaly assessment model based on sequential variational autoencoder and evidence theory is proposed to perform dam anomaly assessment. The anomaly assessment model consists of two stages: anomaly detectio...

Full description

Saved in:
Bibliographic Details
Published in:Applied Mathematical Modelling Vol. 98; p. 576
Main Authors: Shu, Xiaosong, Bao, Tengfei, Xu, Ruichen, Li, Yangtao, Zhang, Kang
Format: Journal Article
Language:English
Published: New York Elsevier BV 01.10.2021
Subjects:
Anomalies
Arch dams
Fuzzy sets
Heterogeneity
Homology
Smoothness
ISSN:1088-8691, 0307-904X
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Considering the multi-sources, heterogeneity and complexity of dam anomaly assessment, a novel anomaly assessment model based on sequential variational autoencoder and evidence theory is proposed to perform dam anomaly assessment. The anomaly assessment model consists of two stages: anomaly detection and anomaly fusion. At the stage of anomaly detection, a novel sequential variational autoencoder is proposed. The backbone of sequential variational autoencoder is fulfilled by the long short-term memory model to capture the latent structures of both generative and inference models. To obtain a robust density estimation, the smoothness-inducing loss is added into the loss function. At the stage of anomaly fusion, the fusion method based on Dezert-Smarandance theory and the interval-valued intuitionistic fuzzy set is employed. The dynamic reliability based on supporting degree is applied to integrating the homologous information among monitoring points. Then the Dezert-Smarandance theory is employed to integrate the heterogeneous information. For verification, an arch dam is taken as an example. Through comparison among representative time series, the proposed model detects anomalies more accurately than other benchmark models. From the analysis of fusion results, the dam is severely abnormal in May 2019.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1088-8691
0307-904X
DOI:10.1016/j.apm.2021.05.021