Risk prediction of water inrush of karst tunnels based on BP neural network

Abstract

To evaluate precisely the risk level of karst tunnel helps reduce the risk of sudden flood water accidents in the process of tunnel construction. On the basis of relevant literature, statistical study and comprehensive analysis of hydrogeological condition in karst tunnel, and select unfavorable geology, formation lithology, underground water level, topography and geomorphology, strata dip Angle, fracture of surrounding rock as risk evaluation index of karst tunnel water gushing. In different hydrogeological conditions, varies a lot. Using BP neural network method to analysis water gushing risk of karst tunnel and avoid the weight of factors. In engineering applications, assess water risk of tunnel by method of BP neural network, avoid the occurrence of sudden flood water, which provides reference for risk prediction of water gushing in karst tunnel. KEYWORD: karst tunnel; water inrush; BP neural network; risk prediction; advanced geological prediction 4th International Conference on Mechanical Materials and Manufacturing Engineering (MMME 2016) © 2016. The authors Published by Atlantis Press 362 gushing of factor, its hierarchical structure shown in Figure 1. Fig.1 hierarchical chart of water and mud in rush risk factors in Karst tunnel BP network learning and training process owe two parts, forward propagation by the network input signal and the error signal reverse propagation, the signal is passed forward, error propagation direction. When the forward propagation, the input information transmitted to the output layer, each neuron output corresponding to the output layer of the network in response to the input mode from the input layer, hidden layer by layer by layer calculation; If system cannot get the desired output the output layer, the error into the back-propagation, by reducing the expected output and the actual output of the error in principle, from the output layer to the intermediate layers, and finally back to the input, correction layers each connection weights. With back propagation training ongoing, correct input mode network response rate is also rising, so the cycle until the error signal to within the permissible range so far (SU Bo et al, 2006). 3 BP EVALUATION SYSTEM NEURAL NETWORK ANALYSIS 3.1 Evaluation System. Karst water burst is phenomenon which underground water storage conditions by outside interference power instability essentially. In this paper, hydrological and geological conditions influencing factors were evaluated. A. Unfavorable geology. Water inrush tunnel construction are encountered with karst pipes, water bearing fault zone, river and other adverse geological related. B. Formation lithology. Tunnel water inrush disaster occurred in limestone, dolomite and other karst formations, lithology higher purity, the greater the layer thickness, the karst development, the more easily to form large karst pipeline (LI Shucai et al, 2014). C. Underground water level. Groundwater plays a material carrier and the source of power in the dual role of water inrush process, is an important factor affecting water burst (LI Liping et al, 2011). Groundwater mainly in pore water, karst fissure water and water form, so that there is a strong disaster-causing ability of groundwater, and the energy in the form of groundwater depends mainly on groundwater level. Therefore, the water table and the floor of the tunnel elevation difference largely characterize the degree of danger of water inrush based on the groundwater table elevation difference h divided four levels, h 60m. D. Topography and geomorphology. Water inrush in underground engineering is closely related to topography, surface karst depressions, sinkholes cause precipitation into the ground, called groundwater recharge source, direct influence of underground karst development. E. Strata dip angle. Strata dip angle have a major impact on underground water flow angle is too large or too small is not conducive to karst. Strata dip angle is too big, small catchment area, the water cycle is weak; Strata dip angle is too small, is not conducive to the infiltration of surface water, karst development is also affected. F. Fracture of surrounding rock. Tunnel Rock fractured situation to a large extent influence the size of water inflow, the fractured media penetration has unevenness, maximum permeability, and fracture and fracture permeability vertical direction parallel to the minimum (LIU Wei, 2014). 3.2 Sudden gushing risk classification. Based on the above analysis, the tunnel sudden gushing risk classification index is determined by experience or expert score, after BP network training sample derived from the specific rates are shown in Table 1.