| 摘要: |
| 【目的】建立原岩应力准确预测方法，为岩石力学研究及地下岩土开挖工程设计与施工提供参考。【方法】充分利用区域实测原岩应力数据资料，选取岩石埋藏深度、岩石类别等参数作为原岩应力的评判指标，在分析基于群体智能（GI）的粒子群优化算法（PSO）和BP神经网络算法特点的基础上，提出一种新的组合训练方法，建立了PSO-BP组合人工神经网络模型，并对原岩应力进行实际算例预测。【结果】PSO-BP组合人工神经网络模型整体工作性能优良，研究区域原岩应力场最大主应力、最小主应力、垂直应力的网络输出与目标输出相关程度较高，相关系数分别为0.994 0，0.997 0，0.992 0，该组合模型基本可以预测研究区域原岩应力场的分布规律。【结论】应用建立的PSO-BP组合人工神经网络模型可以进行原岩应力的准确预测，对岩体初始应力研究和地下工程设计具有一定的指导意义。 |
| 关键词: 原岩应力 预测模型 组合人工神经网络 粒子群算法 |
| DOI： |
| 分类号: |
| 基金项目:国家自然科学基金项目（50874103）；山西省科学技术发展计划项目(20100322013) |
|
| Research on in-situ rock stress prediction based on particle swarm combined artificial neural network |
|
|
| Abstract: |
| 【Objective】The method to predict the in-situ rock stress is built to provide reference for rock mechanics research and underground geotechnical excavation engineering design and construction.【Method】The factors of rock burial depth and rock category etc.are selected as the judging indexes,and the measured rock stress data are used for the researched area.On the basis of the Particle Swarm Optimization (PSO) algorithm based on Group Intelligence (GI) and the standard Back Propagation (BP) artificial neural network method,a new type of combined training method is put forward,and PSO-BP combined artificial neural network model is successfully built in the end.【Result】The combined artificial neural network has good performance from the global aspect.The network output data and target output data of maximum principal stress,minimum principal stress and vertical stress of the studying in-situ rock stress field show closer relationship.The correlation coefficients are respectively equal to 0.994 0,0.997 0,0.992 0,in the multivariate regression calculation results. The combined artificial neural network can predict stress field distribution law in the studyed area rock.【Conclusion】The PSO-BP combined model can accurately predict in-situ rock stress,and it has certain guiding significance to theoretical study of rock initial stress and underground engineering design. |
| Key words: in-situ rock stress prediction model combined artificial neural network PSO algorithm |
