SVM大坝变形监测模型研究

Dam Deformation Monitoring Model Based on Improved MMA-SVM

大坝的变形监测数据是一个复杂的非线性的时间序列,采用传统的建模方法存在拟合和预报精度低等问题。传统算法中,基本遗传算法不能确保全局最优收敛,而普通多变异位自适应遗传算法在进化初期对群体不利,容易导致进化走向局部最优。针对这一问题,提出一种基于改进的多变异位自适应遗传优化支持向量机(SVM)的建模方法。多变异自适应遗传算法采用二进制多点交叉,可根据个体适应值大小,自动选取合适的交叉概率和遗传概率,针对遗传算法易陷入局部最优点,对上述遗传算法进行改进,并利用该算法对支持向量机的模型参数进行寻优。将上述建模方法用于大坝变形监控模型的建立,结果表明该组合算法能有效提高模型的拟合和预报精度。

Dam deformation monitoring data are a complex nonlinear time series.While modeling with traditional modeling methods,problems like low accuracy fitting and forecasting arise.In traditional algorithms,the basic genetic algorithm can＇t ensure global optimal convergence,while the average multiple mutation adaptive genetic algorithm is unfavorable for groups in the early stage of evolution,which generates a high possibility of leading the evolution towards local optimum.In response to this problem,this paper presents a varied ectopic modeling method based on improved adaptive genetic optimization support vector machine（SVM）.Multiple mutation adaptive genetic algorithm uses binary multi-point crossover method,in which it automatically selects the appropriate crossover probability and genetic probability according to the size of individual fitness value.As genetic algorithm falls into local optimum easily,the genetic algorithm above is improved to appropriately seek the optimization of SVM parameters.The modeling method above is used to establish the model of dam deformation monitoring,and the results show that the combination algorithm can effectively improve the accuracy of the model fitting and forecasting.