An intelligent displacement back-analysis method for the right-bank slope of Dagangshan Hydropower Station

Abstract

A novel intelligent method for cut slope displacement back-analysis is proposed. The method employs the back-propagation (BP) neural network to establish a nonlinear relation between mechanical parameters and deformation behaviors of rock masses affected by excavation and reinforcement. Then genetic algorithm (GA) is incorporated to evolve the BP network topology and their connection weights in order to create the best matched network, instead of exploiting traditional time-consuming Finite Difference Method (FDM) calculations. Moreover, once the BP network model is established, GA is adopted once again to search for the most appropriate mechanical parameters so as to achieve a global minimum in the accumulated error between the calculated displacements (By BP network) and their corresponding observed values. The proposed method is verified by applying it to the displacement back-analysis of right-bank slope of Dagangshan Hydropower Station. The results of the forward analysis carried out by FLAC3D with the back-analyzed parameters demonstrate that the calculated displacements of the monitoring points involved in back analysis are reasonable and very close to the observed ones. Furthermore, the results also demonstrate that the calculated displacements for different depths of two multi-point extensometers match well with the monitored values, which indicate that the back-analyzed parameters are representative and acceptable. Therefore the proposed method has important application value with enough accuracy in geotechnical engineering projects.