Study On The Method Of Efficient Back Analysis Of Rock Mechanics Parameters And Displacement Losses In Underground Engineering Structure

Recently, the development of underground engineering has brought the challenge to the safety problem in construction constantly. The safety problem of underground engineering meanly caused by the consequence that it hard to find a efficient method difficult to provide an effective way to predict the nonlinear behavior of rock mass in the process of construction. As a key factor of analysis and identification of rock nonlinear behavior, the mechanics parameters and displacement losses of tunnel rock is difficult to obtain by the laboratory test and field test. As a method to calculating the mechanics parameters and displacement losses in underground engineering, back analysis has become a hot problem in geotechnical field. Facing with the problem above,Thanks to the National Natural Science and Technology Fund Project (" high geothermal gradient of hydraulic pressure tunnel THM under coupling action of load bearing characteristics of "51369007), in this paper, the following several aspects of research work has been done:(1)The traditional optimization can only get the local optimal solution once it facing with the complicated problem. What’s more, in order to get global optimal solution, thousands and millions of fitness function evaluation is needed when adopted random global optimal solution. To sum up, in this paper, lobal surrogate model based on Gaussian process-particle swarm optimization cooperative optimization algorithm is proposed. The algorithm used GPR surrogate model replace real fitness evaluation to calculated fitness in the process of local optimization. And it also adopts the dynamic update of the learning sample strategy to improve the precision of Gaussian process regression agent model continuously, realized the ability to solve complex function problem efficiently. It has tested the low computational cost and efficient ability of method in this paper by Benchmark functions.(2)The complexity of rock mass of underground engineering structure determines complexity of optimization problem for the mechanical parameters, and the calculation of the value of the objective function often need the help of numerical computation, which determines the calculation function is implicit. The global surrogate model based on Gaussian process-particle swarm optimization cooperative optimization algorithm is proposed which combined mathine learning, optimization algorithm and mumerical computation. And the algorithm is better then PSO algorithm facing with the problem of back analysis.(3) Facing with the complexity objective function, the algorithm of global surrogate model based on Gaussian process-particle swarm optimization depends on the initial set of sample points, and it has its drawback facing with the high dimension problems. Therefore, a cooperative optimization of local surrogate model based on Gaussian process-particle swarm optimization is proposed, and it shows the feasibility and ad advancing by comparing with the local surrogate model based on Gaussian process-particle swarm optimization and particle swarm optimization.(4) As a factor that it is difficult to observing but have a big influence in the process of excavation, displacement losses is the key element to evaluate the secure grade of tunnel reasonably and guide the construction safety. In this paper, lobal surrogate model based on Gaussian process-particle swarm optimization cooperative optimization algorithm is adopted to calculating the displacement losses. And it has confirmed the feasibility of the method by the example of the engineering of Kam screen secondary power station.