| Potential of low resistivity reservoirs in Lower Wuerhe Formation of 8th District,Karamay oil field,Junggar basin is approved by prospecting and exploration.Continuous exploitation for decades has resulted in the decreas of oil and gas production,therefore accurate identification of low resistivity reservoir is the precondition for the exploration of potential in the exploited area.Logging data interpretation is the most important and effective method for identification of low resistivity reservoirs.However,frequently,the results of traditional interpretation methods are unsatisfactory due to the complexity of low resistivity reservoir.In addition,the interpretation standard for low resistivity reservoirs varies with different oilfields.Machine learning algorithm,which has been developed rapidly in recent years,is likely to make a breakthrough in logging interpretation of low resistivity reservoirs owing to its advantages of high computing efficiency and strong generalization ability.As a case of study,the hanging wall of the 256 well fault,Lower Wuerhe Formation,8th District,Karamay is selected as the research object in this thesis.Firstly,the training samples are selected based on well-logging and testing data.Subsequently,4 machine learning algorithms,i.e.Random Forest,Extra Trees,K-Nearest neighbor and Adaboost are used for the interpretation of logging data and identification of low resistivity reservoirs.The results show that the optimized Extra Trees algorithm exhibits the best performance,and the prediction accuracy can reach91.5%.Finally,the performance of trained model is then tested by the samples from main oil-bearing and extension ded area.The experimental results show that the average prediction accuracy of the Extra Trees algorithm is 73.6%in the main oil-bearing area and the algorithm exhibits a good effectiveness;while the average prediction accuracy is only 59.3%in the extension area. |
PDF Full Text Request