关键词: BP神经网络, GM（1, 1）, 支持向量机, 组合预测, 滑坡

Abstract: In order to realize the high precision forecast of landslide deformation and achieve the purpose of landslide stability evaluation, we applies the BP neural network, support vector machine and GM (1,1) model to the traditional single forecast of landslide deformation. To enhance the precision of the single forecast, the genetic algorithm, particle swarm algorithm and semi-parametric method were used to optimize the single forecast models. Based on multiple combination indexes, the comprehensive combined weights were determined through build-up method and tired multiplication to optimize combination forecast of landslide deformation. The results show that the precision and stability of the combination forecast results are higher than the single forecast. In the process of determining the comprehensive combined weights, tired multiplication is superior to build-up method. The average value and standard derivation of relative error of the optimal combination forecast results are 0.81% and 0.62 respectively, which proves that the forecast precision is high and stable and that the forecast method proposed in the paper is applicable and effective to landslide deformation forecast.

Key words: BP neural network, GM (1,1), support vector machine, combination forecast, landslide