Combining forecast of landslide displacement based on chaos theory

Abstract

This study aimed to propose a new forecast combination method for landslide displacement based on the monitoring dataset. Considering the existence of outlier data, the 3σ method was used to eliminate them firstly. In order to diminish the effect of different length units, Z-score normalization was employed in this paper. Space reconstruction was performed to calculate the Lyapunov exponent, which was to verify whether the time series were chaotic. Three models were further studied here, namely, adding-weight one-rank local-region prediction method, adding-weight zero-rank local-region prediction method, and the largest Lyapunov exponent method. Furthermore, they are compared with respect to accuracy with each other and measured data so as to establish a forecast combination method. In order to check its validity, this method was applied to the analysis of the dataset of a 434-day displacement monitoring project performed on 12 monitoring points distributed on 4 profile maps on Qinghuasi Landslide Treatment Project in Yan’an, China. This model, which was not only implemented on a single point but also extended some results to a profile map, effectively, predicted the slope deformation in Qinghuasi landslide site. Prediction results indicate that the displacement of A-A profile map is relatively minimal, while slope deformation has been significant on the whole profile map of C-C. Situations at B-B and D-D are better than C-C profile map, though local failures have occurred at these two profile maps. This new model achieved a higher accuracy. Engineering measures based on these predicting results work effectively, which shows that it would be helpful in practical applications. This paper provides new idea and calculating method that can be used to develop deep data-mining approaches for landslide displacement prediction.