摘要
Abstract
This paper aims to combine the machine learning framework to warn of the possible failure in the oil well,and ana-lyze the recorded single well data(among which the well ID of all wells is recorded in the single well data table and the correspond-ing well failure cause of each well,the start time of well work,the end time,the reason for closing well,etc.,and each well has cor-responding characteristics,such as daily gas production,moisture content,etc.).The intrinsic feature relationship of the oil well da-ta is excavated,the feature screening is carried out by the random forest algorithm,and the feature extraction is carried out on the existing data in combination with the time domain,so as to find the law of oil well work and well closure due to failure and carry out relevant early warning.By establishing an LSTM model,the processed data is trained,the network search method is used to select parameters to obtain the optimal parameters,the real-time data that will occur in the future is predicted,and the time is speculated when the failure may occur in the future,so as to achieve the effect of fault warning and reduce losses.Experimental results show that the early warning model can effectively improve the accuracy of early warning,can perform real-time early warning analysis on the status of oil wells,and can detect and warn of abnormal situations in single wells in advance.关键词
数据预处理/时域/故障预警/随机森林/LSTMKey words
data pre-processing/time domain/fault warning/random forest/LSTM分类
天文与地球科学
