摘要
泥石流是我国山区的主要地质灾害之一。影响泥石流活动性的因素十分复杂,并且具有随机性和模糊性。遗传神经网络结合了神经网络和遗传算法的优点,可以模拟学习和进化之间的交互作用,很适合用于分析泥石流活动性。文章简要讨论了遗传神经网络的原理,建立了泥石流活动性分析的遗传神经网络模型,并将该模型用于川藏公路沿线30条泥石流沟的活动性分析。网络的拓扑结构为(9,6,4,3),即输入节点(评价指标)、第1隐含层、第2隐含层和输出接点(分析结果)分别为9、6、4、3。首先以其中25条泥石流沟作为样本对网络进行训练,训练时网络的连接权采用遗传算法进行自适应演化,待模型稳定后将其余5条泥石流沟的数据输入模型,计算它们的活动性,计算结果与实际观测基本相符,证明模型是可行的,各个参数的选取也是合适的。
Debris flow is one of the geological hazards in mountainous regions in China and the factors which control the activity of debris flows are random and fuzzy.Genetic neural network,which is the combination of neural network and genetic algorithm,is able to simulate the interaction of learning and evolution and,therefore,is suitable for analyzing the activity of debris flows.Firstly,the principals of genetic neural network are briefly discussed in this paper.Then a genetic neural network model for analyzing the activity of debris flows is set up and this model is applied to the analysis of actity of 30 debris flow gullies distributing along Sichuan\|Tibet road.The topological structure of the model is (9,6,4,3),which means the input nodes(evaluation indexes),the first hidden layer,the second hidden layer and the output nodes(analysis results) are 9,6,4,3 respectively.The data of 25 gullies are used as samples to train the network and,during the training,the linking weights of the network evolve self\|adaptively using genetic algorithm.After the model is stable,the data of the other 5 gullies are input into it to calculate their activity and the results accord with that of observation.Therefore,this model is feasible and the parameters used in the model are appropriate.
出处
《中国地质灾害与防治学报》
CSCD
2003年第3期16-20,共5页
The Chinese Journal of Geological Hazard and Control
基金
中国科学院知识创新资助项目(编号:KZDX2 306)
