基于SVM模型的基坑监测数据预测

Data Prediction of Foundation Pit Monitoring Based on SVM Model

随着不敏感损失函数ε在支持向量机(Support Vector Machine,SVM)中的引进,这一机器学习模型在非线性领域中得到有效运用.SVM在工程领域主要用于进行回归分析、模式识别以及分类,这些应用逐渐成为工程领域的热点.笔者基于SVM构建立模型,对郴州某紧邻边坡条件下的基坑支护桩位移与锚索拉力监测数据进行分析后,建立基于两者施工过程中的监测数据的预测模型.通过与后期所得监测数据对比,检验预测的准确率.之后利用模型对将来某段时间基坑支护桩的位移进行预测,用于评价基坑在预测时间内的稳定性.经模型计算后发现,在对紧邻边坡的基坑支护桩不同深度变形的预测上,在短期内具有较好的一致性,最大误差不超过9%.但随着时间的增长,预测值越来越偏离实际情况,预测周期超过3d误差陡增,预测值不再具有参考意义.预测模型可用于类似工程中对基坑的稳定性作出判断.

With the introduction of insensitive loss function ε in support vector machine(SVM), this machine learning model is effectively used in the non-linear field. SVM is mainly used for regression analysis, pattern recognition and classification in engineering field. These applications have gradually become a hot spot in engineering field. Based on the SVM model, the prediction model of the monitoring data in the construction process is established after the monitoring data analysis of foundation pit support pile displacement and anchor cable tension under the the condition of close to the slope in Chenzhou. The accuracy of the prediction is tested by comparing with the monitoring data obtained in the later stage. Then, the model is used to predict the displacement of foundation pit pile for a certain period of time in the future, which is used to evaluate the stability of the foundation pit in the predicted time. After the model calculation, it is found that the prediction of deformation in different depth of foundation pit supporting pile adjacent to slope has good consistency in the short term, and the maximum error is not more than 9%. With the increase of time, the forecast value deviates from the actual situation more and more. The error increases sharply when the forecast period exceeds 3 days, and the forecast value is no longer of reference significance. The prediction model can be used to judge the stability of foundation pit in similar projects.