【摘要】 液压锤打桩机是一种节能、高效、环保的工程机械,基于其沉桩过程数据采集和传输功能的条件基础,提出了一种基于随机森林的沉桩过程状态模式识别方法。本研究通过对沉桩过程中桩端未入持力层阶段、入持力层阶段和收锤阶段的数据构建随机森林进行预测,并与反向传播神经网络(BPNN)和支持向量机(SVM)进行比较分析。通过嵌入法进行特征选择,并进行重要性排序得出地勘土层信息、单位深度锤击数、单位深度锤击能量为模式识别精度主要影响因素。利用准确率(Accuracy)和混淆矩阵等性能指标对三种算法进行对比验证,分析得到随机森林在预测收锤阶段的可靠性更为优秀,具有较高的预测精度,能够为桩基工程施工提供较好的指导作用。更多还原

【Abstract】 Hydraulic hammer pile driver is a kind of construction machinery with energy saving, high efficiency and environmental protection. Based on the condition basis of its data acquisition and transmission function in pile driving process, a method of state pattern recognition in pile driving process based on random forest was proposed. In this study, the Random Forest algorithm was constructed to predict the pile not into the bearing layer stage, into the bearing layer stage and stopping hammering stage in the process of pile driving, and then compared with algorithms of BPNN and SVM. The embedding method was used to select the features, and the importance ranking was carried out. It was concluded that the information of geological prospecting soil layer, hammer number per unit depth and hammer energy per unit depth were the main influencing factors for the accuracy of pattern recognition. By comparing and verifying the three algorithms with performance indexes such as accuracy and confusion matrix, the analysis shows that the random forest has better reliability and higher prediction Accuracy in predicting stopping hammering stage, which can provide better guidance for the construction of pile foundation engineering.更多还原