Research On Reservoir Lithology Recognition Based On Integrated Clustering Fine Subdivision

With most oil fields entering the mid-to-late development stage,the exploration and development of oil and gas wells at this stage are mainly for unconventional storage resources.However,due to the impact of today's tense international relations and the global spread of the new coronavirus on oil supply and demand,crude oil prices are about to remain relatively high for a long time Low level.Fine reservoir description plays a vital role in improving exploration quality and reducing production and development costs.Among them,using logging data to identify reservoir lithology is an important part of fine reservoir description.To this end,based on modern signal processing and analysis,machine learning and pattern recognition technology,this article is committed to effectively improving the ability to express the formation information of cross-well logging curves,and develops a series of studies on reservoir lithology recognition method based on integrated clustering and fine subdivision.(1)Aiming at the problem that the basic identification unit constructed by the existing lithology identification methods cannot make full use of the context information provided by the continuity of the logging signal curve,a lithology identification study based on the characterization of logging meta-objects is proposed.This method draws on the main object-oriented ideas in the image field,and aims at mining potential related information in logging information.Specifically,the original logging data is first segmented based on logging characteristics to obtain multiple interval meta-objects,and then feature extraction is performed from multiple angles such as statistical features,morphological features,and geological attributes.Lithology recognition is performed by using a classifier in the feature space.Through the verification experiments of a variety of different classifiers,the accuracy of lithology recognition using this method and the accuracy and recall rate of each sub-category lithology have been significantly improved.(2)In view of the problems of current logging data segmentation methods that mostly focus on singular values and are easily affected by measurement errors,as well as to further improve the accuracy of meta-object division,a logging data segmentation method based on integrated clustering is proposed.This method integrates multiple K-means clusterers.In order to effectively cluster the logging data,it proposes the construction of local trust space,differentiated generation base clusterers and base cluster relations,and adopts the spectral clustering method Perform the division of the base cluster relationship graph.Through the stratification experiment of actual logging data,this method can effectively avoid the interference of singular values,enhance the effectiveness of the interval division,and obtain a great improvement in the lithology identification experiment of the meta-object characteristics.(3)Aiming at the problem that existing machine learning algorithms have high requirements for practitioners' experience in tuning parameters,which is not conducive to the application and promotion of some machine learning algorithms,this paper proposes a lithology identification method based on differential evolution and XGBoost.This method uses XGBoost with excellent performance as the basic classifier,and uses the differential evolution method to iteratively optimize the parameter population of the classifier,and finally select the optimal model parameters.Through comparative experiments on lithology recognition of original data and meta-object feature data,this method has better recognition accuracy,can give full play to the excellent performance of XGBoost,and has better recognition stability.