【摘要】 泥石流危险性的评价是地质灾害预测的重要研究课题之一,根据泥石流危险性评价因子,建立了遗传BP神经网络评价模型。模型利用BP神经网络的函数逼近能力,模拟泥石流某些主要评价指标与危险程度之间的非线性函数关系,实现对泥石流危险性的评价。为克服BP神经网络收敛慢、易陷入局部极小、泛化能力差等缺陷,引入遗传算法和对比分析方法优化BP评价网络的权值、阈值和网络结构。实验证明,采用所述方法优化后的BP神经网络的拟合精度、准确度、效率大幅提高,泛化能力也得到增强,该方法可作为解决泥石流危险性评价问题的一种新思路。更多还原

【Abstract】 The risk assessment of debris flow is an important research field of geological disaster prediction.According to risk as-sessment indexes of debris flow,this paper constructs genetic BP Neural Network assessment model.The model utilizes function approximation capability of BP Neural Network to simulate nonlinear relation between some primary assessment indexes and risk of debris flow,and finally realizes risk assessment of debris flow.In order to overcome the shortcoming of BP Neural Network,such as slow convergence,easily getting into local dinky value and low generalization ability,the paper introduces genetic algo-rithm and comparative analysis method to optimize the weight,threshold and network structure of BP Assessment Network.Experiment shows that the precision,accuracy and efficiency of simulation have been greatly improved,and generalization ability has been enforced after optimizing by adopting the method.Therefore,this method should be a new way to solve risk assessment of debris flow.更多还原