Abstract: The terrain from Sangri to Jiacha is steep with deep-incised valleys and intense tectonic activities. The strong erosion trenching of Yarlung Zangbo River since early Pleistocene has caused release and redistribution of tectonic stress in this region. The stress field here is extremely sophisticated. Geostress field analysis is of great significance to the route selection and construction of railways. This study combines the terrace characteristics alongside the Yarlung Zangbo River to generalize the evolution laws of the river valley and create a 3D geological-mechanical model according to geological conditions. The existing geostress field from Sangri to Jiacha of Lasa-Linzhi railway is then calculated based on the real-tested geostress data and the utilization of the geostress inverting method combining the RBF neural network and the principle of strata denudation. The result shows that the geostress measurements in each testing point are in good consistency with the real-testing values, and the geostress field obtained through this method is reasonable and reliable. On this basis, the stress field features of the river bank slope of Sang-jia canyon section and the major tunnel projects along the valley river bank are analyzed, and the major problems during the tunnel construction are discussed based on the value and direction characteristics of the main stress field at tunnel axis location.

Key words: geostress field, deep-incised valley, inversion analysis, RBF neural network, Lasa-Linzhi railway