基于AdaBoost集成的WPSO-RBF大坝变形监控模型

Dam Deformation Monitoring by Radial Basis Function Model Optimized by Particle Swarm Optimization with Inertia Weight and Ada Boost

变形监测是大坝安全监测的必设项目,由于影响因子众多,常利用神经网络(如BP,RBF等)进行参数选取和模型建立。传统的径向基函数(RBF)神经网络因网络结构简单、收敛速度快而被广泛运用,但其在预测中易陷入局部最优且参数选取不当会对其收敛性产生影响。因此,首先利用动态权重粒子群算法(WPSO)对RBF神经网络的3个参数(隐含层基函数的中心c、宽度d及隐含层到输出层的权值w)进行优化,建立基于WPSO-RBF的大坝变形监控模型,然后将WPSO-RBF模型作为弱分类器,采用Ada Boost算法进行集成,建立基于WPSO-RBF-Ada Boost的大坝变形监控模型。将该模型运用到工程实例中,实例结果显示该模型具有收敛速度快、分类精度高、泛化能力好,可建立较优的大坝变形监控模型。

Deformation monitoring is a requisite for dam safety monitoring. Due to a large number of factors,neural networks such as back propagation（ BP） and radial basis function（ RBF） are often used for parameters selection and model establishment,of which RBF has been widely employed on account of its simple network structure and rapid convergence. Nonetheless,local optimality and inappropriate selection of parameters will exert great impact on the convergence rate. In view of this,the Particle Swarm Optimization with Inertia Weight（ referred to as WPSO）is adopted to optimize three parameters of RBF（ central value c of hidden layer base function parameter,width d and connection weight w between hidden layer and output layer parameter）. In subsequence,the WPSO-RBF model is integrated as a weaker classifier by Ada Boost algorithm,hence establishing a WPSO-RBF-Ada Boost model for dam deformation monitoring. The model is applied to practical engineering,and results suggest that the present model is of fast convergence,high classification precision and good generalization ability.