论文部分内容阅读
目前颗粒流计算方法中所采用的细观力学参数,均需根据岩土体宏观力学参数试验结果反复调试获取,调试效率很低且有较大的盲目性,因此,急需引入一种新方法,建立岩土体宏观力学参数与细观力学参数的关系。基于PFC3D程序,采用BP(back propagation)神经网络方法,建立宏观力学参数与细观力学参数的非线性模型,通过输入宏观力学参数,即可快速、准确地反演岩土体细观力学参数。研究结果表明:(1)根据BP神经网络模型反演确定的细观力学参数,输入数值模型计算其宏观力学参数,结果与试验值相比,精度一般高于90%;(2)当模型最小尺度上的颗粒数RES=10、隐含层含6个神经元时,BP神经网络模型的反演性能最佳。实例计算表明,BP神经网络模型对于岩土体细观力学参数的确定具有良好的反演性能,该方法为颗粒流理论的推广应用提供了新的技术手段。
At present, the meso-mechanics parameters used in the particle flow calculation method need to be debugged according to the test results of the macroscopic mechanical parameters of rock and soil. The commissioning efficiency is very low and has a large blindness. Therefore, it is urgent to introduce a new method, RELATIONSHIP BETWEEN MACRO AND MECHANICAL PARAMETERS OF ROCK AND SOIL. Based on the PFC3D program, BP (back propagation) neural network method is used to establish a nonlinear model of macroscopic mechanical parameters and mesomechanical parameters. By inputting macroscopic mechanical parameters, the mesomechanical parameters of rock and soil can be quickly and accurately retrieved. The results show that: (1) According to the mesomechanical parameters determined by BP neural network model inversion, the macroscopic mechanical parameters are input into the numerical model, and the accuracy is generally higher than 90% compared with the experimental values; (2) When the model is the minimum When the number of particles on the scale RES = 10 and the hidden layer contains 6 neurons, BP neural network model has the best inversion performance. The case study shows that the BP neural network model has good inversion performance for the determination of mesomechanical parameters of rock and soil mass. This method provides a new technical means for the popularization and application of particle flow theory.