基于PCA-GA-RF的地震液化场地侧移预测模型研究

Prediction Model for Liquefaction-Induced Lateral Spread Displacement Based on PCA-GA-RF

为了准确预测地震所引起的场地液化侧向位移,基于主成分分析(Principal Component Analysis,PCA)、遗传算法(Genetic Algorithm,GA)、随机森林算法(Random Forests,RF)提出了一种场地液化侧移预测模型。首先通过PCA降维处理搜集到的地震液化侧移数据库,其次利用GA算法优化RF算法的参数决策树个数和分裂属性个数等,再次选择最佳参数完成模型的训练,最后将其结果与其他模型进行对比分析。结果表明,PCA-GA-RF预测模型的决定系数R2、均方根误差(Root Mean Square Error,RMSE)和平均绝对误差(Mean Absolute Error,MAE)分别为95%、0.41、0.27,相较于传统的多元线性回归法(Multiple Linear Regression,MLR)和RF模型在拟合优度和预测精度等方面有了显著提升,因此能够满足工程实际运用需求。在所有输入参数中,震级、震中距、可液化层厚度、细粒土含量以及临空比的敏感性程度较高,对液化侧向位移的影响较大,因此在勘察或试验时应该更注重这些参数的测量精度以确保预测的准确率。

In order to accurately and effectively predict the lateral displacement of site liquefaction caused by earthquake,a prediction model of site liquefaction lateral displacement is proposed based on Principal Component Analysis(PCA),Genetic Algorithm(GA)and Random Forest Algorithm(RF).PCA is used to reduce the dimension of the collected seismic liquefaction lateral displacement database in the world,and GA algorithm is used to optimize the number of parameter decision trees and split attributes of RF algorithm,and the best parameters are selected to complete the training of the model.Finally,the results are compared with other models.The results show that the determination coefficient R2,root mean square error RMSE and average absolute error MAE of PCA-GA-RF model are 95%,0.41 and 0.27,respectively.Compared with the traditional Multiple Linear Regression(MLR)and RF model,it has significantly improved the goodness of fit and prediction accuracy,and can be used in engineering practice.Among all the input parameters,magnitude,epicentral distance,liquefiable layer thickness,fine soil content and open space ratio are highly sensitive and have greater influence on liquefaction lateral displacement.Therefore,more attention should be paid to the measurement accuracy of these parameters in investigation or test to ensure the prediction accuracy.