基于因子分析的BP神经网络在岩体变形模量预测中的应用

岩体变形模量是研究岩体变形特性的重要参数,它对工程岩体稳定性评价与优化设计具有重要意义。本文提出了基于因子分析的BP神经网络预测岩体变形模量的方法。以西藏某水电站为例,在现场调查、室内外试验的基础上,建立了48组包括密度、吸水率、纵波波速、单轴抗压强度、岩块变形模量以及泊松比等因素的数据库,采用因子分析法对6个影响因素进行分析,可得3个公共因子,该3个公共因子作为神经网络的输入参数,采用BP神经网络进行预测。结果表明:利用因子分析法可降维输入数据,消除BP神经网络中由于输入数据太多而影响数据处理速度的缺陷; 把因子分析法和BP神经网络结合进行岩体变形模量的预测,可使预测精度提高; 该研究思路不仅对岩体变形参数的预测是一个有益的尝试,而且对类似岩土工程问题的预测也有借鉴意义。

APPLICATION OF BP NEURAL NETWORK BASED ON FACTOR ANALYSIS TO PREDICTION OF ROCK MASS DEFORMATION MODULUS

Rock mass deformation modulus is the important parameter in the study of rock mass deformation characteristics. It is also of great importance to the stability analysis and optimal design of engineering rock mass. A method for predicting the rock mass deformation modulus is presented in this paper. It uses the BP neural network based on factor analysis. It is applied to the case of a hydropower station in Tibet. On the basis of laboratory tests and in-situ tests, a database of 48 data sets including density, water absorption, vertical-pace, uniaxial compressive strength, rock mass deformation modulus and poisson's ratio factors is established. Three public factors are obtained using the factor analysis method to analyze the six factors. The three public factors act as the input parameters and are used to make BP neural network predictions. Some important conclusions are drawn: The factor analysis can eliminate the defect that the excessive inputting data slows down the processing speed in BP neural network. The prediction accuracy can be improved using this method. This research idea is not only an useful attempt to predict rock mass deformation modulus, but also a great reference value to solve similar geotechnical engineering problems.