Classification of Rocks Surrounding Tunnel Based on Improved BP Network Algorithm

The classification of rocks surrounding a tunnel has an important significance for guiding design and construction in underground engineering. This paper introduces an artificial neural network method into the classification of these rocks. Based on traditional back propagation (BP) arithmetic, an enhanced neural network method is obtained by improving the training algorithm, transfer function and network structure. By combining the additive momentum method with the self-adjusting learning speed method, the algorithm has been improved: when the error is bigger than the upper critical limits the learning speed automatically decreases; when the error is smaller than the lower critical limits the learning speed automatically increases. Thus, the training speed can be fast yet at the same time the stability of the network can be ensured. By introducing the parameter of adjusting learning speed, the transfer process becomes more sensitive and the convergent speed becomes faster, thus, increasing the calculating precision of the training function. By giving a data range for a certain implicit layer joint model, the structure of the network is optimized; correspondingly, the functional precision is improved. The improved BP network model is tested in example classifications of some typical rocks surrounding tunnels in the Dong Shen Water Supply Reconstruction Project. The results fit well with the classification according to the code of hydraulic tunnel design in China, which indicates that this improved method has a high practical application.