A hybrid prediction model of landslide displacement with risk-averse adaptation. (August 2020)

• Record Type: Journal Article
• Authors: Xing, Yin; Yue, Jianping; Chen, Chuang; Qin, Yuluo; Hu, Jia
• Is Part Of: Computers & geosciences. Volume 141(2020)
• Publication Date: 2020-08
• Subjects: Landslide -- Displacement prediction -- Risk-averse adaptation -- Hybrid model -- Penalty mechanism; Environmental policy -- Periodicals; 550.5
• DOI: 10.1016/j.cageo.2020.104527
• Languages: English
• ISSNs: 0098-3004

Abstract

Abstract: Reliable and accurate landslide displacement prediction is the key to early warning system. Most studies focus only on accuracy of landslide displacement estimation and ignore catastrophic consequences caused by underestimated landslide displacement. This paper investigates a landslide displacement prediction method with risk-averse adaptation. In this methodology, double exponential smoothing method is utilized to predict trend term of landslide displacement, while hybrid model of support vector regression and long short-term memory network is developed to predict periodic term of landslide displacement. Considering the adverse effect of underestimated displacement, a cost function with a penalty mechanism is proposed to force the underestimated displacement to shift to an overestimation. Finally, the predicted cumulative displacement is obtained by superposing the predicted trend displacement with the periodic displacement. Verification results on Baishuihe landslide in China indicate that the proposed approach for landslide displacement prediction can maintain a high prediction accuracy and reduce the underestimation rate, thereby achieving adaptive avoidance of risks. Highlights: Double exponential smoothing is suitable for landslide trend displacement extraction. Hybrid model improves prediction accuracy of landslide periodic displacement. Swarm intelligence algorithm helps to determine hybrid model parameters. Cost function with a penalty mechanism allows fostering risk-averse adaptation.