基于灰色相关分析的GRFM倾斜量预测模型  

GRFM forecasting mode of inclination based on the grey relation analysis

作　　者：张小红[1,2] 王慧琴[1,3] 于洪磊[3] 高大峰[4] 王展 ZHANG Xiaohong WANG Huiqin YU Honglei GAO Dafeng WANG Zhan(School of Management, Xi＇an Univ. o f Arch. ＆ Tech., Xi＇an 710055, China School of Communication＆Information Engineering, Xi＇an Univ. of Scie.＆ Tech.,Xi＇an 710054, China School of Information Control Engineering, Xi＇an Univ. of Arch. ＆ Tech. ,Xi＇an 710055,China School of Civil Engineering, Xi＇an Univ. of Arch. ＆ Tech. ,Xi＇an 710055,China Shanxi Provincial Institute of Cultural Relics Proteciton)

机构地区：[1]西安建筑科技大学管理学院,陕西西安710055 [2]西安科技大学通信与信息工程学院,陕西西安710054 [3]西安建筑科技大学信息与控制工程学院,陕西西安710055 [4]西安建筑科技大学土木工程学院,陕西西安710055 [5]陕西省文物保护研究院,陕西西安710075

出　　处：《西安建筑科技大学学报（自然科学版）》2016年第6期919-924,共6页Journal of Xi'an University of Architecture & Technology(Natural Science Edition)

基　　金：国家教育部归国留学人员科技支撑基金资助项目(K05055);西安市碑林区科技计划基金资助项目(GX1614)

摘　　要：针对时间序列的动态性、相关性、小样本性、非线性等特征,利用灰色模型的小样本适用性和神经网络的预测高精度等性能,提出了一种基于相关分析的灰色神经网络组合预测模型.首先,基于灰色相关理论定量分析了倾斜量与沉降观测指标时间序列之间的相关度;然后,采用GM(1,1)模型对原始序列累加求和,降低各因素原始数据的噪声干扰;利用优化径向基神经网络(Radial Basis Function,RBF)多步拟合训练,其中心点和扩展系数初值采用蚁群算法进行优化.最后,将该模型应用到了砖石古塔的倾斜量预测中,设计了沉降综合指数,通过计算,该指数与倾斜量的灰色相关度为0.789 1,采用该模型对某古塔倾斜量进行了预测,平均相对误差为9.056%.实验结果表明,该模型对小样本、非线性的时间序列预测具有高精度和有效性,为古建筑保护中变形预测提供了理论和实践经验.Considering time series with dynamic, small sample size, relativity, nonlinearity characteristicses, a new grey radial basis function （RBF）neural network forecasting model based on correlation analysis is proposed,which is based on the capability of grey model fitting to the small sample data and the properties of neural network with high prediction precision. At first, the new model analyses the correlation of the inclination and settlements time series quantitatively. And then, generated the accumulation sequences of the original time series using GM（1,1） model, which reduced the noise of the original data; the result is input into the radial basis function neural network, the center pointer and expansion coefficient of which is optimized by ant colony algorithm to forecast the inclination quantity with multi-steps.At last, the new model is applied in small goose pagoda, after the grey correlation of inclination and the inhomogeneous settlement index designed is calculated, the inclination quantity is forecasted with the average relative error of 9.056%.The experimental results show that the proposed model has the capability of the good precision and effectiveness, which supplies the theoretical and practical experience for the ancient building protection.

关 键 词：相关性分析 灰色模型 径向基神经网络 预测模型 砖石古塔 

分 类 号：TU196.4[建筑科学—建筑理论]