摘要
针对GM(1,1)模型在建筑物变形预测中精度和泛化能力较低的缺陷,提出一种基于LS-SVM的灰色补偿RBF神经网络的建筑物变形组合预测方法。利用最小二乘支持向量机训练由灰色GM(1,1)模型预测得到的一组结果的残差值,直接获得RBF网络的中心函数训练RBF网络,得到RBF误差补偿器,去补偿GM(1,1)模型。实验证明,最小二乘支持向量机、灰色系统以及神经网络3者相结合的方法,能有效提高建筑物变形沉降预测的精度。
Based on the ability of the GM(1,1) model in building deformation prediction accuracy and generalization is low, this paper proposes a gray compensating RBF neural network based on the LS-SVMbuilding deformation of the combination forecast method. This method uses the least squares support vector machine training by the grey GM (1, 1) model to predict the residual value of a set of results, direct access to the center of the RBF network function training RBF network, and the RBF error compensator, to compensate the GM (1, 1) model. The experiments show that the least squares sup- port vector machine, along with the grey system and the neural network methods, effectively improve the accuracy of deformation of building settlement prediction.
出处
《大地测量与地球动力学》
CSCD
北大核心
2016年第1期66-68,74,共4页
Journal of Geodesy and Geodynamics
基金
国家自然科学基金(41461089)
广西“八桂学者”岗位专项
广西空间信息与测绘重点实验室基金(桂科能140452402,130511402)
广西自然科学基金(2014GXNSFAA118288)
广西矿冶与环境科学实验中心课题(KH2012ZD004)
广西研究生教育创新计划(YCSZ2014151,YCSZ2012083)~~