基于主成分分析法优化神经网络的滆湖组黏性土抗剪强度预测

为了研究苏锡常地区滆湖组黏性土抗剪强度特性，建立抗剪强度参数预测模型；以研究区711组滆湖组黏性土物理力学试验数据为载体，运用主成分分析(principal component analysis, PCA)方法，从样本11个指标中提取影响目标变量的主成分；将其作为反向传播神经网络(back propagation neural network, BPNN)模型的输入层，建立基于PCA-BPNN算法的滆湖组黏性土抗剪强度预测模型。结果表明：当主成分数量为3时，主成分累计贡献率达93.4%;第一、二主成分贡献率分别为52.1%和36.6%;PCA算法即保留了样本大部分信息，又实现了对多维变量的降维。第一主成分可归纳为土体孔隙特性，与黏聚力和内摩擦角均呈负相关关系；第二主成分可归纳为土体水稳性，与黏聚力和内摩擦角均呈正相关关系；土体孔隙特性越显著，水稳性越弱，抗剪强度越低。建立了滆湖组黏性土抗剪强度参数PCA-BPNN预测模型，模型抗剪强度拟合优度为0.85,内摩擦角拟合优度为0.72;模型可靠性总体较高。可见PCA-BPNN预测模型即可降低解释变量间的多重共线性，简化了模型，又能够提升模型...

Prediction of Shear Strength of Cohesive Soil in Gehu Formation based on Back Propagation Neural Network Optimized by Principal Component Analysis

In order to study the shear strength characteristics of cohesive soil of Ge-Hu Formation in Suzhou-Wuxi-Changzhou area and establish a prediction model of shear strength parameters, the principal component analysis (PCA) method was used to extract the principal components affecting the target variables from 11 indicators of the sample based on 711 sets of physical and mechanical test data of cohesive soil in Gehu Formation in the area. After the selected principal components being used as the input layer of the Back Propagation Neural Network (BPNN) model, the prediction model based on PCA-BPNN algorithm is established. Results show that :( 1) when the number of principal components is 3, the cumulative contribution rate of principal components reaches 93.4%. The contribution rates of the first and second principal components are 52.1% and 36.6% respectively. PCA algorithm not only retains most of the information of the sample, but also reduces the dimension of multidimensional variables. (2) The first principal component can be summarized as soil pore characteristics, which are negatively correlated with cohesion and internal friction angle. The second principal component can be summarized as soil water stability, which is positively correlated with cohesion and internal friction angle. Therefore, the more significant the pore characteristics of the soil, the weaker the water stability and the lower the shear strength. (3) PCA-BPNN model was established for predicting the shear strength parameters of cohesive soil in Gehu Formation .The goodness of fit of the model for cohesion is R=0.85 and the goodness of fit for internal friction angle is R = 0.72. The prediction accuracy and reliability of the model are generally high. In summary, the PCA-BPNN prediction model can not only reduce the multicollinearity between explanatory variables, simplify the model, but also improve the generalization ability of the model; it provides a reference for using mathematical methods to study soil engineering geological parameters.