Abstract:

A piecewise linear interpolation and Boostrap resample methods are introduced to enrich information of deformation monitoring data in detail and enlarged data capacity. Moreover, the coordination transforms or space reconstruction is also used to change data representation and rewrite indistinct patterns or rules of monitor data in order to enhance the finiteness and incompleteness of sample data in process of prediction modeling for deformation behavior on rock and soil system. The efficiencies of new ideas are tested by the comparative studies of prediction with support vector machine (SVM), RBF and Elman neural network for deformation behavior of 3 typical projects of which the monitor data are deal back and forth with reproduction and space reconstruction. The results show that the approach method can effectively extend data information and highlight essential laws of evolution. Specifically, the SVM prediction model based on data representation is superior to SVM, RBF and Elman model which directly used monitor data. The data reproduction and space reconstruction can thus be utilized to more server deformation prediction of rock and soil.

Key words: data reproduction, space reconstruction, deformation behavior, prediction