关键词: 滑坡, 先验知识, 非平稳变形, SVM预测, 外延模式

Abstract: Defining extensive patterns of nonstationary deformation by analyzing the physical significance and types of monitor time series, aiming at the lower reliability of landslide early-warning due to the key information data, reflected in stages, repeatability and mutability of landslide process, are limited or uncompleted observed. A method, added or imposed extensive pattern is identified according to current status of landslide, such as deformation stage analyzed use factors superposition technique and the types of time series, and the potential causative factors of prediction time into model of support vector machine (SVM) algorithm is presented. The efficiencies of new idea are tested by predicting two groups of displacement time series. All experimental results show that the project prior knowledge can be availably used to increase data message of nonstationary deformation and guide forecast modeling. In contrast to common SVM, the model based on extensive pattern is more reliable and accurate. Particularly, the average relative error is markedly cut down about 2 to 3 times.

Key words: landslide, prior knowledge, instability deformation, support vector machine prediction, extensive pattern