Intelligent Prediction Method of Pipe Sticking Based on VC-SVM and Particle Swarm Optimization
In the process of well drilling,pipe sticking frequently occurs due to the factors such as complex downhole conditions and unclear knowledge of the strata,seriously restricting the drilling efficiency.However,the existing prediction methods of pipe sticking are defective in aspects such as accuracy,timeliness and transferability.This paper presents an intelligent prediction method of pipe sticking based on ensemble learning idea and intelligent optimization algorithm.First,based on the actual sticking data at well site and a reasonable label calibration meth-od,the label was accurately positioned at the point before the occurrence of sticking rather than at the freeze-in point,and by means of parametric dependence analysis,characterization significance analysis,and timeliness and creditability analysis,7 input parameters were selected.Second,three algorithms,i.e.random forest(RF),support vector machine(SVM)and BP neural network,were used to build a sticking prediction model,and the performances of each model under seriously uneven proportions of sticking and non-sticking samples(sticking to non-sticking ratio:1∶117)were compared.Third,the SVM model was selected as the basic model for the predic-tion of pipe sticking,and improved using the ensemble learning idea.Meanwhile,the particle swarm optimization(PSO)was used to simultaneously conduct hyperparameter optimization on multiple SVM classifiers,simplifying the parameter tuning process while achieving coupled optimization.Finally,the 10 times of sticking samples of a block were used to conduct training test.The results show that the improved model can effectively search for hyperplanes of different types of pipe sticking,with a transfer prediction false alarm rate controlled at 9%and a missed alarm rate of less than 7%,effectively predicting most of the data points of each pipe sticking.This study is expected to improve the risk warning efficiency of field drilling,and provides support for ensuring safe and efficient drilling of wells.
万方数据
维普
