A Study On Displacement Prediction By Evolution Neural Network And Elastic-plastic Back Analysis From Measured Displacement

Since NATM came out and made a figure, it has arisen great recognition in geotechnical engineering world and many researchers are finding numerical solutions about the design of NATM's supporting structure. But the geotechnical engineering , faced by us, is engineering ground mass which is composed of natural material and structure, ground mass behaves great ramdomness, fuzziness, uncertainty, nonintegrality of information, so geotechnical engineering is impossibly engineering of "precise solution". The time series of displacement around tunnels contain the information of systematic evolution, and we can find out the embedded rule from these data. The ANN starts with simulating cerebric visual thought, and it has wide application foreworld for geological information of underground engineering,taking on strong buzzing, fuzziness, nonlinear. The physical properties of geotechnical material such as mechanical proerties, are so complex that we must establish rather accurate constitutive relation equation if we want to predict the dynamic mechanical properties of all kinds of underground constructions such as underground cavern, tunnels, dams, bridge foundation and so on, by analytic methods. The more constitutive relation is complex,the more parameters need be inputed, and no matter how difficulty these parameters are gotten in the lab or locale , personal influence is certainly rather much according to a good many mechanical parameters. The initial displacement after the cavern is excavated can't be monitored except that we embed monitoring device. But the difficulty and cost is so large. Based on local measured information, the problem can be solved, utilizing the strong non-linear-mapping capacity of evolutional neural network. The elastic-plasticity theory should be acted as computing gist of analysing mechanical condition of strata. The elastic-plastic optimizational back-analysis used to establish each sort of parameter, can be utilized to solve complex geotechnical engineering problem and have considerable economic value.This paper focuses on 4 facets as below:1. Searching the most optimistic structure parameters of Elman neural network by GA, which puts up better temporal dynamic character, we can predict thecertain point's displacement within some fore scope via sample learning, so making up the lost because of constructive influence and achieving the final displacement more accurately.2. Erecting analytic model by FLAC3D,we will compute the stress and strain of surrounding rock and achieve displacement of underground carvern.3. Using several controlling points' actual displacement in underground cavern, the article adopt elastic-plastic theory. We can invert Young's modulus E of rock mass and component of horizontal ground stress P, acting on the side of computing model, using golden section optimization and coordinate conversion method.4. Utilizing inversion results, we can verify the feasibility of the method of the article combining geological condition, constructive situation and monitoring data.