Modeling and predicting reservoir landslide displacement with deep belief network and EWMA control charts: a case study in Three Gorges Reservoir

Abstract

The accurate modeling and predicting of landslide deformation is crucial to the prevention of landslide hazard. This paper presents a pioneering study of modeling and predicting the reservoir landslide displacement with deep learning algorithm. A data-driven framework using deep belief network and control chart has been introduced to explore the temporal patterns of displacement and potential of identifying seasonal faster displacement. First, the continuous wavelet analysis has been applied to decompose the time-series precipitation, reservoir water level, and displacement into seasonal and residual components. Second, the deep belief network has been constructed to predict the future displacement. Third, it utilizes the exponentially weighted moving average (EWMA) control chart to derive the boundaries as alarm conditions of seasonal faster displacement. A group of tests are conducted to compare the performance of the deep belief network with other state-of-the-art machine learning algorithms. Computational results demonstrated the effectiveness of the deep belief network in extracting highly non-linear data features. In addition, the advantage of utilizing control charts has been further validated by the accuracy of examining the seasonal faster displacement based on the case study in Baishuihe landslide in Three Gorges Reservoir, China.