Research And Implementation Of Early Kick Detection Algorithm In Drilling Process

Kick is a common underground complex accident in oil drilling engineering,without timely warning of kick,which may lead to kick developing into blow out,causing casualties and property losses.Therefore,timely warning of kick has always been a difficult problem to be solved in the oil industry,which is very significant for well control safety.At present,the warning of kick mainly depends on the artificial judgement on-site,the engineers judge whether there is kick in combination with the change law of various parameters,this method depends on personal experience of field engineers and has high labor intensity,which is easy to lead to false alarm and missing alarm.Aiming at the problems of strong subjectivity and low stability of artificial early warning,combined with the relevant knowledge in the field of deep learning,an adaptive kick detection algorithm based on Long Short-Term Memory is proposed in this thesis,which mainly includes the adaptive feature extraction of kick process,the design and implementation of kick detection model based on LSTM,and the construction of on-site real-time early warning scheme.This thesis includes the following work:1)Select the parameters that can represent the kick characteristics by analyzing the kick process,and construct the kick database by extracting,interpolating,filtering and labeling the original data.An adaptive feature extraction method based on sliding window is used to construct training samples,this method shields the numerical difference of different drilling data,smooths the jitter of data,and ensures the stability of kick warning algorithm;2)After completing the construction of kick samples,LSTM is selected as the basic model according to the time sequence property of drilling data,and a kick detection model framework based on LSTM is built.The optimal model structure and sliding window length are obtained by taking false alarm and missing alarm as indicators,the model can automatically calculate the current kick probability according to the input drilling data without manual participation,which avoids the subjectivity caused by manual on-site early warning;3)Combined with the needs of on-site early warning,a set of real-time early warning scheme is designed,including drilling data analysis,drilling state judgment,model deployment,etc.The results of on-site experiments show that the model can complete the early kick warning with low false alarm,and basically reach the expected effect.In order to promote the application of the kick detection model,a design idea of early warning software system is proposed,and the core function of the software is completed.The research results show that the adaptive feature extraction algorithm can effectively weaken the impact of data differences,the kick detection model can calculate the current kick probability according to the input drilling data in low frequency of false alarm and missing alarm,and can assist on-site engineers to make effective judgment and reduce the pressure of early warning.The application of this algorithm is expected to improve the intelligence and automation of drilling kick detection.