基于概率神经网络的无粘性土渗透破坏预测研究  

Prediction of Seepage Failure of Cohesionless Soil Based on Probabilistic Neural Network

作　　者：欧阳磊 刘家印 何利军[1] 韩燚 宋金博[2] OU Yang-lei;LIU Jia-yin;HE Li-jun;HAN Yi;SONG Jin-bo(Nanchang Hangkong University,Nanchang 330063,China;Jiangxi V&T College of Communications,Nanchang 330013,China)

机构地区：[1]南昌航空大学,江西南昌330063 [2]江西交通职业技术学院,江西南昌330013

出　　处：《云南水力发电》2022年第6期37-40,共4页Yunnan Water Power

基　　金：广西防灾减灾与工程安全重点实验室开放课题(2016ZDK013);广西大学工程防灾与结构安全教育部重点实验室开放课题(2016ZDK013);基于人工智能的隧道多源异构监测数据融合及塌方安全预警方法研究(2022H0026)。

摘　　要：为了避免人为因素及判别方法对渗透破坏判别方法的限制,引入非线性映射能力强的概率神经网络,利用其容错性能好、分类正确率高等优势,输入无粘性土的物性参数,构建“物性参数—渗透破坏类别”预测模型进行渗透破坏类别预测,预测结果与实际相符。并选取了不同平滑因子输入网络,结果表明平滑因子的正确选择对网络输出至关重要。与其他算法的预测模型的预测值做对比分析,可以直观的看出基于PNN的渗透破坏类型预测模型的分类结果可靠。In order to avoid the limitation of artificial factors on the method of judging penetration failure,a probabilistic neural network with strong non-linear mapping ability is introduced in this paper.With the advantages of good fault tolerance and high classification accuracy,the physical parameters of non-cohesive soil are input,and the prediction model of"physical parameters-category of seepage failure"is constructed to predict the category of seepage failure.The prediction results are consistent with the actual situation.The input networks with different smoothing factors are selected,and the results show that the correct selection of smoothing factors is very important to the network output.By comparing and analyzing the predicted value of the prediction model with that of other algorithms,it can be concluded that the classification result of the prediction model of penetration failure type based on PNN is reliable.

关 键 词：无粘性土 渗透破坏 概率神经网络 预测 平滑因子 对比 

分 类 号：TU441.33[建筑科学—岩土工程]