Font Size: a A A

Research And Application Of Oil Well Overflow Monitoring Based On Machine Learning

Posted on:2021-07-25Degree:MasterType:Thesis
Country:ChinaCandidate:H L XiaoFull Text:PDF
GTID:2481306308473644Subject:Electronics and Communications Engineering
Abstract/Summary:PDF Full Text Request
Artificial intelligence is the current research hotspot,which has been widely used in engineering applications and scientific research.The combination of traditional industry and artificial intelligence has become one of the development trends,while oil drilling engineering is a typical traditional industrial discipline,and overflow monitoring has been one of the important topics in the industry.Timely and accurate overflow monitoring can effectively improve drilling efficiency,reduce losses and reduce drilling costs,and achieve safe and efficient drilling engineering.In this paper,machine learning algorithm is used to study the problem of drilling overflow monitoring:First of all,the causes of overflow in the process of oil drilling are studied,as well as the characterization rules of drilling parameters when overflow occurs,which provides technical basis for drilling data preprocessing.The overflow data and non overflow data in the logging data obtained in the process of drilling are seriously unbalanced.The processing method of non-equilibrium data is proposed.The overflow monitoring effect of the training model of balanced data and non-equilibrium data is compared,and the better performance of the proposed method is verified by the actual drilling system.Secondly,based on the characterization of logging parameters and the method of non-equilibrium data processing when overflow occurs,according to the specific situation of the data obtained in the drilling process,the pre-processing scheme of drilling data is designed to be divided into four parts:drilling data cleaning,overflow feature selection,data set division and normalization,and non-equilibrium data processing.Overflow feature selection is to select overflow feature of cleaned data according to the symptom law of overflow;data set division and normalization is to divide drilling data into training set,verification set and test set;unbalanced data processing is to process the divided training set.Finally,according to the specific situation of overflow monitoring in the process of drilling,this paper proposes a four layer framework of drilling overflow monitoring scheme,which is data source layer,drilling data preprocessing layer,overflow monitoring model training layer and overflow monitoring model results display and analysis layer.In the training layer of overflow monitoring model,Bayesian classification,logistic regression and neural network are used to train drilling data respectively,and the effects of three models are analyzed.In the result display and analysis layer of overflow monitoring model,the optimal overflow monitoring model is tested and the result display analysis is carried out.
Keywords/Search Tags:Machine learning, non-equilibrium processing, overflow monitoring
PDF Full Text Request
Related items