Research On The Improvement Of PBA Method That Based On SVM Optimized By Genetic Algorithm

Depending on the project of the Liberation Road subway station in Changchun, the PBA method and the research status of inversion about rock and soil mechanics parameters at home and abroad are first expound. To make sure the rock and soil mechanics parameters reasonable, accurate and stable during the numerical simulation, the Support Vector Machine optimized by Genetic Algorithm is used for the inversion of rock and soil mechanics parameters. In combination with the global optimization of Genetic Algorithm and the efficient regression of Support Vector Machine, the parameters of Support Vector Machine are gotten by screening in the way of Genetic Algorithm, which are used to establish a reasonable SVM model. Then according to the measured displacement, the rock and soil mechanics parameters are inversed. And then, considering the effect of the initial supporting arch on the whole construction process, in order to control the settlement better and improve the safety of this stage, the construction plan of the initial supporting arch in view of the PBA method is improved the adjusted. Meanwhile, the plans, both the former and the improved ones, are analyzed from the construction sequence, excavation methods and supporting measures,quantities and the stress situation, and are compared comprehensively with each other from six aspects, including safety, quality, progress, efficiency, stress distribution and operability and so on. On the basis of rock and soil mechanics parameters inversed, the numerical models of PBA method of the former and the optimized ones are established by finite difference program, FLAC 3D. The changes of settlement during the construction process by the PBA method before and after optimization are analyzed and compared, including the surface subsidence analysis and arch sedimentation analysis. In addition, the safety of the structure,including vault and the key node, during the construction process is validated. Through the above research, the following conclusions are gotten.a. The relative error of the results of inversion about rock and soil mechanics parameters are all controlled within 5%, which illustrates the method is accurate and feasible. The learning samples is constructed by orthogonal design method, which can get satisfactory results with less scientific samples. The inversion result shows that the RBF kernel function has better recognition. And the inversion precision has the positive correlation with the number of learning samples.b. In terms of the construction technology of the former and the optimized ones, the optimized one is better than the former in quantity, safety, quality, progress and so on. While the stress is more complicated than the former, the stress safety of the optimized one need to be validated in the subsequent numerical simulation.c. As for the control on settlement, during the initial supporting arch stage, both the surface settlement and the vault settlement of optimized scheme is much less than the former one. When the whole construction is completed, the maximum surface settlement and vault settlement is 2.8% and 3.2% less than the former respectively.d. In terms of the structure stress, during the initial supporting arch stage, both two schemes exist obvious stress transformation process. However, the maximum arch stress appears at different stage and different location. Both of them are less than the concrete compressive strength design values. The maximum key node stress, the maximum tensile stress and the optimized maximum shear stress are less than the concrete compressive strength design values. The arch stress and key node stress are all within the structures withstand range,which is accordance with the requirements of the structure safety.