基于径向基函数神经网络泥石流危险性评价.docVIP

基于径向基函数神经网络泥石流危险性评价 　　 (1.成都理工大学 a.信息工程学院;b.地球科学学院, 成都 610059;2.中石化西南油气分公司 勘探开发研究院 信息中心, 成都 610081) 　　摘 要：泥石流危险性的主要评价指标与危险程度之间有着某种复杂的非线性的关系，通常采用统计分析、模糊评价、BP神经网络等评价方法，但这些方法均存在不足之处，难以进行准确评价。为了克服以上方法的不足，结合泥石流危险性评价指标，建立了基于径向基函数神经网络的泥石流危险性评价模型，并将该模型结果与BP神经网络的评价结果进行了对比。实验结果表明，径向基函数神经网络的模拟结果比BP神经网络更接近测量数据，精度更高，训练所需时间更少。因此，径向基函数神经网络经过充分训练后，能够较为准确地对泥石流的危险性进行评价，具有较好的应用价值。 　　关键词：径向；网络；泥石流；评价 　　中图分类号：TP301 文献标志码：A 　　 文章编号：1001?B3695(2009)01?B0241?B03 　　 　　Risk assessment of debris flow based on radial basis function neural network 　　CHEN Gang1a,2,HE Zhengwei1b,YANG Yang1a,YANG Bin1a 　　(1.a.College of Information Engineering, b.College of Earth Sciences, Chengdu University of Technology, Chengdu 610059, China;2.Information Center, Exploration Production Research Institute, SWPB SINOPEC, Chengdu 610081, China) 　　Abstract:There were some nonlinear relation between main assessment indexes of debris flow and risk. Statistics analysis, fuzzy evaluation,BP network and other methods were usually adopted. However, they had some insufficiencies and they were difficult to accurately assess risk. In order to overcome insufficiencies of these methods, this paper combined with assessment indexes of debris flow and construct risk assessment model of debris flow based on radial basis function neural network (RBFNN).It also conducts a contrast of assesment result between the RBFNN model and BPNN.Experiment shows that, compared with BPNN, RBFNN simulation have higher data precision, less training time and closer to measurement data. Therefore, RBFNN is capable of making more precise risk assessment of debris flow after enough training, and it’s more valuable in application. 　　Key words:radial; network; debris flow; assessment 　　泥石流危险性的评价一直是地质灾害预测的重要研究课题之一，传统上通常采用统计分析方法。由于统计分析方法是从积累的经验或大量历史记录中寻找泥石流的发生规律[1，2]，不可避免地存在评价主观性、工作量大、时间长等问题。近年来，随着新理论和技术的发展，在泥石流的危险性评价中又引入了模糊评价[3~8]、BP神经网络[9~16]等方法，取得了较好的应用效果。但上述方法也存在一些不足之处，例如：由于模糊评价中确定隶属度比较困难，一般采用层次分析法[7]，而在应用最大隶属度原则时，常会因忽略有用信息，导致评价结果失真[8，9]