Improvement of the Improved BP Neural Network Forecast Method and its Application in Long-term Settlement of the Tunnel

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.