基于v-SVR和MVPSO算法的边坡位移反分析方法及其应用

摘要

To address the problems of the limit search space and local optimization in traditional particle swarm optimization algorithm,a modified variation particle swarm optimization(MVPSO) algorithm is proposed based on particle migration and variation by introducing migration operator and adapting mutation operator.The results of the benchmark test functions show that the convergence rate of this MVPSO algorithm has significantly improved than the traditional particle swarm optimization algorithm.For the nonlinear and multimodal problems,the proposed MVPSO functions well in searching the global minimum.In order to establish a nonlinear relation between the mechanical parameters of rock mass and the displacements,the MVPSO algorithm is adopted to search for the most suitable parameters of the v-SVR model.The results show that the prediction accuracy and generalization ability of the v-SVR have been significantly increased.Then,the optimal v-SVR model is an alternative for the time-consuming FLAC calculations; and the MVPSO algorithm can be used to search for the best group of the mechanical parameters of rock mass.Consequently,a new displacement back analysis method is developed in combination of the v-SVR with the MVPSO algorithm.Compared with the traditional displacement back analysis methods,including BP-GA and the v-SVR-GA,the proposed method has its merits in inversion efficiency and accuracy.Finally,the new method is applied to the parametric back analysis of rock mass in the right-bank slope ofDagangshan hydropower station.Based on the back-analyzed parameters,the deformation and stability of the slope during subsequent construction period are analyzed.The results demonstrate that the proposed method has high accuracy and good applicability.%针对传统粒子群算法存在搜索空间有限、容易陷入局部最优点的缺陷,通过引入迁徙算子和自适应变异算子,提出基于粒子迁徙和变异的粒子群优化(MWSO)算法.基准测试函数结果表明,改进的MVPSO算法较传统的粒子群优化算法在收敛效率上有大幅度提高,在处理非线性、多峰值的复杂优化问题中能快速地搜索,得到全局最优解.应用改进的MVPSO算法搜索最佳的支持向量机(v-SVR)模型参数,建立岩体力学参数与岩体位移之间的非线性支持向量机模型,提高v-SVR的预测精度和推广泛化性.然后,利用v-SVR模型的外推预测替代耗时的FLAC正向计算,利用改进的MVPSO算法搜索岩体力学参数的最优组合,提出v-SVR和MVPSO相结合的边坡位移反分析方法(v-SVR-MVPSO算法),与传统的BP-GA算法和v-SVR-GA算法相比,该算法在反演精度和反演效率上均有较大幅度提高.最后,将本文发展的v-SVR-MVPSO算法应用到大岗山水电站右岸边坡岩体参数反演分析,并对边坡后续开挖位移和稳定性进行预测,取得较好的效果.