关键词: 变形监测, 灰色模型, 最小二乘支持向量机, 几何平均生成变换, Metropolis准则, 自适应降温函数

Abstract: Due to the poor monitoring environment，deformation monitoring sequence often accompanied with large fluctuations.The grey model （GM） is only suitable for solving exponential deformation series，and the least squares support vector machine （LSSVM） is difficult to effectively select the parameters in the deformation prediction.In order to sovle the existing problems of GM and LSSVM，an improved grey least squares support vector machines deformation prediction model （IGM-LSSVM） is proposed.Firstly，the geometric mean generating transformation is introduced into GM （1，1） model to enhance the exponential regularity of its input samples，and the deformation values are initially predicted，the residuals are also calculated；secondly，according to the disadvantages that artificial bee colony algorithm （ABC） is easily fall into local extremum when optimizing the parameters of LSSVM，the metropolis criterion is introduced and adaptive cooling function is designed to get an adaptive metropolis colony algorithm （AMABC）；finally，a set of residual values that predicted by LSSVM based on AMABC algorithm is used to compensate the GM （1，1） model，and the final prediction value is obtained.The deformation prediction results of a mining area show that the shortcomings that the ABC algorithm is easy to fall into local optimal solution is solved effectively，the average relative error of IGM-LSSVM，G（1，1） and ABC-GM-LSSVM are 1.223%，9.565％ and 3.200％，the prediction precise of IGM-LSSVM is higher than other two models，which further show that IGM-LSSVM is suitable for the the large fluctuation deformation monitoring sequence，and it has certain reference for realizing high precision deformation prediction of mine slope.

Key words: Deformation monitoring, Gray model, Least squares support vector machine, Geometric mean generating transformation, Metropolis criterion, Adaptive colling function