A novel outlier detection method for monitoring data in dam engineering. (1st May 2022)

• Record Type: Journal Article
• Authors: Shao, Chenfei; Zheng, Sen; Gu, Chongshi; Hu, Yating; Qin, Xiangnan
• Is Part Of: Expert systems with applications. Volume 193(2022)
• Publication Date: 2022-05-01
• Subjects: CS algorithm Cuckoo Search algorithm -- DBSCAN Density-Based Spatial Clustering of Applications with Noise -- LOF Local Outlier Factor -- IF Isolation Forest -- RC Robust Covariance -- RMSE Root Mean Squared Error -- R2 Coefficients of determination; Dam monitoring data -- Outlier detection -- Gaussian blur -- Otsu binarization -- Cuckoo Search algorithm; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2021.116476
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: Detecting outliers from the perspective of image processing and artificial intelligence. An universal and robust approach for outlier detection of monitoring data. Performance of monitoring model is improved after eliminating outliers by the proposed method. Abstract: The accuracy of the dam monitoring data influences the evaluation of dam safety. However, the actual monitoring data may contain outliers due to sensor failure or operation mistakes, which consequences to an inaccurate evaluation. To tackle this issue, an universal and robust approach for outlier detection for monitoring data is proposed from the perspective of image processing and artificial intelligence in this work. The monitoring data series consists of isolated-pattern outliers, clustered-pattern outliers and normal points. We transform the monitoring data into a binary image of a scatter plot at first. With Gaussian blur, the gray scales of isolated points (outliers) are reduced and hence are eliminated after Otsu binarization. Then, Cuckoo Search (CS) algorithm is utilized to obtain the most possible connection of the pixel aggregations, distinguishing the clustered-pattern outliers and the process line automatically. To detect the outliers thoroughly, we implement the image processing techniques and CS algorithm cyclically until the detection results converge. The results indicate that the proposed method has the highest detection accuracy compared with other five outlier detection methods. Moreover, the monitoring model established based on the data pre-processed by the proposed method has better fitting and predicting ability.