Algorithm for Dike Abnormal Behavior Detection Based on Transfer Function Model and One-Class Classification
Abstract
Purpose: Dike monitoring with sensors installed in the dike usually assumes that the sensor readings are compared with some predefined threshold values derived from the analysis or modeling of the dike structure. This method can detect only the simplest cases of dike failure like crest overtopping, being useless for more complex cases caused by internal erosion. The purpose of this work is developing and testing an algorithm for detecting an abnormal dike condition caused by internal erosion. Results: The proposed algorithm is based on the simulation of the transfer function between the measured signals of the water level and the pore pressure inside the dike. A one-class classifier “Neural Clouds” estimates a nonlinear fuzzy membership function which checks whether the model error belongs to the area of normal state of the dike. The classifier is taught on historical data of normal dike behavior obtained from the sensors. The fuzzy response of the classifier varies from 0 to 1, giving an estimation of how close the current state of the dike is to an abnormal state. The algorithm has been tested on natural experimental data. Practical relevance: The results and algorithms were used by Siemens in its AI component of a dike condition monitoring system.Published
2015-12-01
How to Cite
Kozionov, A., Pyayt, A., Mokhov, I., & Ivanov, U. (2015). Algorithm for Dike Abnormal Behavior Detection Based on Transfer Function Model and One-Class Classification. Information and Control Systems, (6), 10-18. https://doi.org/10.15217/issn1684-8853.2015.6.10
Issue
Section
Information processing and control