摘要
对已有的BP神经网络预测方法做了进一步的改进,通过对比选取了最优的网络训练模式和传递函数;利用反复训练和统计学原理推导了适用于确定单个隐含层神经元个数的解析式,并提出了与其相适应的归一化方法、最优归一区间和最优隐层神经元个数的取值范围;指出当输入神经元大于3个时,采用具有双隐含层结构的BP神经网络进行预测的效果远好于单个隐含层结构.对隧道结构整体沉降进行了预测,效果满意,为合理选择后续施工工艺提供了依据.
In this paper, the existing BP neural network forecasting method is first improved. The made improvements include that: The best training mode of network and transfer function are got by contrastive analysis ; For the BP neural network with only one output neuron, the explicit expression of the number of neuron in hidden layer is derived by using repeated training and statistical theory; A normalization approach, best normalized interval and range of neuron in hidden layer are proposed ; When the number of input neurons exceeds 3, it is showed that the prediction effect of BP neural network with double hidden layers is superior to that of the network with one output neuron. Based on the improved BP neural network, the settlement of the whole tunnel structure is forecasted and the results is satisfying, which is very useful for choosing the follow-up construction technology.
出处
《北京工业大学学报》
EI
CAS
CSCD
北大核心
2011年第1期92-97,共6页
Journal of Beijing University of Technology
基金
国家'九七三'计划资助项目(2010CD732102)
关键词
隧道沉降
预测方法
神经网络
deformation of the tunnel structure
forecasting method
improved BP neural network