| Rock media is predominantly fracture-karst in karst, which is strongly heterogeneity and anisotropy. It is frequently to occur water-inrush,mud pumping and collapse disasters in karst tunnels. How to increase the prediction accuracy of water-inrush is always a tough question in karst. Research the water-inrush disaster, on the one hand we need research the prediction methods; on the other hand it's necessary to research the disaster environmental together. To concluded the advantages and disadvantages of each method and the suitability of various storage structures.To research the prediction methods, currently, it proposed many methods. Review many literature, the prediction methods can be divided as follows five categories, Approximate method,Mathematical method,Numerical simulation,Random mathematical and Nonlinear theory.To research the disaster environmental, From the point of prediction water-inrush, the karst water-storing structure can be divided into three categories, Monoclinic-type,Syncline-type and Anticline-type by structural control, and be divided into pure carbonate and non-soluble rock clip rock type by lithology respectively. This paper as examples in Maoba syncline of Yuanliang mountain tunnel and Mingyue ravine anticline of Mingyue mountain tunnel, Used groundwater dynamics method,rainfall infiltration method,VM numerical simulation,AHP-fuzzy and Neural network to predicte the tunnel water-inrush respectively.The calculation results of Maoba syncline. The total tunnel maximum water-inrush result show taht the rainfall infiltration (2057859m~3/d) is far larger than the groundwater dynamics method (285333m~3/d). The total tunnel stability water-inrush, result of groundwater dynamics method (168819m~3/d) is largest, followed by the rainfall infiltration (118504m~3/d),and VM is the smallest calculated(106620m~3/d). Analyze of, the larger section of water is in the aquifer P2w+P2c and P1q+P1m, and the small amount of water is in the Non-soluble rock. Compare with the actual disaster, the result of groundwater dynamics method is relatively realistic, and the general trend of the rainfall infiltration and VM resurts can reflect the relative risk, but water of rainfall infiltration is larger than the actual, and VM is smaller. For the evaluation level, the result of AHP-fuzzy and neural network is consistent. Paragraph 8,9,11,12 and 14 has occurred disasters, and the evaluation are theâ…£orâ…¤level. For the paragraphs which has no water-inrush hazards, the evaluation are theâ… -â…¢level. The predicted with satisfactory results.The calculation results of Maoba syncline. The total tunnel maximum water-inrush result show taht rainfall infiltration (6332363m~3/d) is far greater than groundwater dynamics method (289873m~3/d). The total tunnel stability water-inrush, result of rainfall infiltration (283632m~3/d) is the largest, followed by the groundwater dynamics (137441m~3/d),and VM is the smallest calculated(12626m~3/d). Analyze of paragraphs, the water-inrush is concentrated in the anticlinal core where is soluble stratum T2l and T1j, and is much lower in the wings. Compare with the actual, the result of groundwater dynamics is relatively realistic. The general trend of rainfall infiltration and VM resurts can reflect the relative risk, but water of rainfall infiltration is larger, and VM is smaller. For the evaluation of level, AHP-fuzzy evaluat paragraph 5-12 areâ…£level, but these paragraphs had no water-inrush disasters. The neural network predicted value isâ…£orâ…¤level except paragraph 5 and 12, so the neural network results is relative slightly better.Analysis the method suitability, various methods show little difference in the two examples. In the Mathematical methods, groundwater dynamics is more suitable for the two structures than rainfall infiltration method. Simultaneously, rainfall infiltration is more suitable for Mingyue ravine anticline, so that rainfall infiltration is more suitable for the tunnel which through diving aquifer or the shallow tunnel, and the groundwater dynamics is more suitable for the tunnels which is buried deep or have high head. Numerical simulation,AHP-fuzzy and neural network is suitable for both examples. Forecast accuracy is more importantly determined by methods and parameter selection. |
