摘要
采用粒子群优化算法代替传统共轭梯度法对高斯回归学习机进行最优超参数搜索,克服共轭梯度法在优化过程中对初值依赖性太强、迭代次数难以确定、易陷入局部最优的缺点,进而形成粒子群—高斯过程回归算法。基于该算法提出一种根据现场岩石回弹强度σH预测岩石单轴饱和抗压强度σC的新途径,并对绩黄高速佛岭隧道4个掌子面20组σH及σC数据进行学习预测以检验该算法性能。工程应用表明:与传统采用点荷载实验确定σC方法相比,该算法能够更快建立和反映二者之间的映射关系,预测精度较高,加速了施工现场围岩分级变更的信息反馈及σC指标提取,可以此取代传统的数值回归模型进而指导施工。
As the traditional conjugate gradient method shows the disadvantages of being too dependent on the initial value,difficult to determine the iteration steps and easy to fall into the local optimum during the optimization process,the particle swarm optimization（PSO）is used,instead of it,to optimize the hyper parameters of the Gaussian process regression（GPR）,upon the basis of which the PSO-GPR algorithm is formed.Furthermore,it is adopted as a new approach to establish the mapping between the rebound strength and saturated uniaxial compressive strength of the rock for evaluating the compressive strengthσC.Besides,20 groups ofσHandσCfrom 4 work faces of the Foling Tunnel of the Ji-Huang Expressway are learnt-predicted to test the performance of the algorithm.The application results show that,compared with the traditional method of using the point load tests to determineσC,the algorithm can describe the relationship between the two very well and perform more accurate prediction,which boosts the information feedback in the process of the classification of the surrounding rock,and thus can replace the traditional numerical model in guiding construction.
出处
《国防交通工程与技术》
2015年第2期16-19,31,共5页
Traffic Engineering and Technology for National Defence
关键词
围岩分级
岩石单轴饱和抗压强度
粒子群优化算法
高斯过程回归
智能预测
classification of the surrounding rock
saturated uniaxial compressive strength of the rock
particle swarm optimization algorithm
Gaussian process regression
intelligent prediction
