一种基于PCA-RF的卡钻预测方法

Research on PCA-RF-based sticking prediction method

为实现井下卡钻工况预测,提出一种基于主成分分析法(PCA)与随机森林(RF)相结合的卡钻预测方法(PCA-RF)。该方法通过PCA方法对井下测量工程参数进行降维处理;用降维后的数据对RF模型进行训练和测试,判断是否发生卡钻事故。通过PCA-RF卡钻预测模型与RF、SVM、PCA-SVM模型对比发现:PCA-RF卡钻预测模型准确率可达98.53%,训练与测试时间分别为4.64 s与0.71 s,准确率和运算效率均优于RF、SVM、PCA-SVM模型。

In order to realize the prediction of downhole sticking condition,a method based on principal component analysis(PCA)and random forest(RF)is proposed.For this method,PCA method is used to reduce the dimension of underground measurement engineering parameters,and the reduced dimension data is used to train and test the RF model to judge whether there is a sticking accident.By comparing PCA-RF model with RF,SVM and PCA-SVM model,it is found that the accuracy of PCA-RF model can reach 98.53%,and the training and testing time are 4.64 s and 0.71 s respectively.The accuracy and the operation efficiency of PCA-RF model are better than those of RF,SVM and PCA-SVM model.