论文部分内容阅读
针对岩土力学参数正算反演方法存在的瓶颈问题,研究了将支持向量回归机(ε-SVR)与粒子群-差分演化杂交智能优化算法(BBDE)相结合的参数反演方法,采用ε-SVR的预测功能代替耗时的数值仿真计算,将其嵌入到改进了控制参数取值方法的BBDE算法中建立了相应的反演程序,应用该程序对一个实际工程进行了弹塑性力学参数反演,并对反演程序的可行性和有效性进行了研究.结果表明:利用ε-SVR具有的BP神经网络所不可比拟的泛化推广能力,在保证反演精度的同时提高了反演效率;引入BBDE算法使得在减少算法控制参数的同时提高了解的全局收敛性和收敛速度;将反演所得参数输入数值仿真模型获得的测点计算增量位移与实测增量位移比较吻合,相对误差不超过10%.
In order to solve the bottleneck problem of the positive inversion method of geotechnical parameters, a parametric inversion method combining support vector regression (ε-SVR) with Particle Swarm Optimization (BBDE) SVR prediction function instead of time-consuming numerical simulation, the embedded BBDE algorithm to improve the control parameters of the value of the establishment of the corresponding inversion program, the application of an actual project on the elastic-plastic mechanical parameters of the anti- And the feasibility and validity of the inversion procedure are studied.The results show that the inversion efficiency can be improved while the accuracy of inversion can be guaranteed by using the generalized generalization ability that is not comparable to the BP neural network that ε-SVR has. The BBDE algorithm is introduced to improve the global convergence and convergence rate of the algorithm while reducing the control parameters of the algorithm. The incremental displacement of the measured point obtained from the inversion parameter input into the numerical simulation model is in good agreement with the measured incremental displacement. The relative error is not significant Over 10%.