基于数据挖掘技术的黄土湿陷性评价

Assessment of loess collapsibility based on data mining

为了运用数据挖掘技术进行黄土湿陷性评价,根据实际工程资料建立了黄土物理力学数据库,用主成分分析法对原数据进行压缩,用压缩后的新变量依据人工神经网络理论建立了预测模型,用BP算法进行了模型的校正及预测。工程实例分析表明,预测湿陷系数与试验值所得湿陷系数的湿陷量计算值相比,准确率可达96%以上,说明这种智能化评价方法具有可行性和实用性。

This paper presents a method for assessment of loess collapsibility based on the dada mining technology. Loess collapsibility is predicted by using the function of data mining. The database should be created based on practical engineering. Data in the database are compressed with principal component analysis （PCA）. Prediction model is built with BP neural network. Variables processed through PCA are used as input part of prediction model. Loess collapsibility is predicted by the model. The predicted loess collapse settlement is compared with measured loess collapse settlement. The result shows that prediction precision of collapse settlement is up to 96% by a specific project example,indicating that the intelligent method of evaluating loess collapsibility is very useful in engineering.