基于PSO-SVM的边坡稳定性组合预测方法研究

为利用不同边坡稳定预测方法的特征信息，改进预测质量，提出了一种基于微粒群优化--支持向量机(PSO-SVM)的边坡稳定性非线性组合预测模型。该模型能够利用边坡的特征参数快速预测出边坡的稳定性，且在建模过程中可对不同建模方法的特征信息进行整合，避免了单一方法的偶然性。为提高SVM的学习、泛化能力，采用混合核函数，并用具有并行性和分布式特点的PSO算法优化选择SVM模型参数。利用该非线性组合预测模型对73个边坡实例进行学习，对另外10个边坡实例进行推广预测，研究结果表明，该模型较好地整合了不同建模方法的特征信息，较单一模型、加权组合模型和BP网络组合模型具有更高的预测精度和更小的峰值误差，为边坡稳定性评价提供了一种新的途径。

Combined forecasting model for slope stability based on support vector machines with particle swarm optimization

In order to use the characteristic information of different modeling methods sufficiently and improve the quality of prediction result, a nonlinear combining forecasting model for slope stability based on support vector machines with particle swarm optimization was presented. The new model could predict the safety factors using only characteristic parameters of slopes. Characteristic information of different modeling methods were integrated and the chanciness of single prediction method was avoided. In the interest of improving the forecasting accuracy, support vector machines which using a mixed kernel function was used; and the parameters of SVM were selected by PSO algorithm which had the characteristics of parallel and distributing. Applying the nonlinear combined forecasting model to predict safety factors of 10 slope cases after learning with other 73 samples; result showed that the model combining the characteristic information of different modeling methods very well. Compared with single predicting model, regressive combining model and BP network combining model, the PSO-SVM combining model had a higher predicting precision and smallest peak error; thus a new approach to estimate the slope stability more accurately was provided