基于无监督学习的井漏风险层位预测方法OA

Lost circulation is a common engineering problem in drilling engineering,and particularly important to accurately predict the risk horizon of lost circulation before drilling for safe drilling.The prediction of lost circulation risk is based on the study of the matching relationship of seismic attributes.To reduce the dependence on the calibration results of lost circulation horizon attributes,the key parameters closely related to lost circulation events are extracted by unsupervised clustering algorithm.Aiming at the problem of anomaly detection in the case of unbalanced data processing,based on the support vector machine(one-class SVM for anomaly detection)algorithm,the lost circulation risk assessment model and lost circulation risk index are constructed to assess the lost circulation risk horizon of new drilling wells.The constructed lost circulation risk index corresponds to the probability of lost circulation in the predicted risk horizon,and the predicted lost circulation risk can be divided into four grades.The coincidence rate of the predicted results and the actual results can reach 85.8%.The results show that this method can solve the problem of abnormal detection under the condition of high dependence of the calibration results of lost circulation horizon attributes and unbalanced processing data,and provide a new technical idea for the prevention and treatment of lost circulation.