关键词: 位移反分析, 大型地下洞室, 围岩扰动, 增量位移, 遗传－神经网络

Abstract: Displacement back analysis based on the in-situ measured information is one of the important parts that compose the dynamics measurement-control and information-oriented construction of the underground project. On the basis of measured displacements of surrounding rockmass , a system analysis method for back analysis of rock mechanics mechanical is established, which is used to gain the unloading surrounding rock mechanical parameters of the large-scale underground powerhouse (excavating span 30 meters and height 78.5 meters) of Pengshui Project on Wujiang River. The analysis method includes uniform design (UD) ，BP artificial neural network（BP-ANN）and genetic algorithm (GA) .Uniform design and numerical simulation are used to gain training samples of BP-ANN, and BP-ANN is used to simulate the relationship between rock mass displacements and rock mechanical parameters. In addition, the initial input weights of BP-ANN model are optimized by genetic algorithm (GA) because the initial input weights of BP artificial neural network model strongly affect the accuracy of neural network. And the rock mechanical parameters are gained through the trained BP-ANN and measured increment displacement of surrounding rock mass. Finally, the surrounding rock mass displacements induced by the excavation of underground powerhouse are forecasted. According the result, the forecasting displacements are approximate to the measured ones; so as to validate, the rightness and reliability of the above method.

Key words: displacement back analysis , large-scale underground powerhouse , unloading rockmass, increment of displacement , GA-ANN