Research On Lithology Identification Method Of Logging Curve For The Reservoir Characterization

Reservoir characterization is an inevitable requirement for reducing the current exploration cost of oil and gas resources,for improving the reliability and accuracy of oil and gas resources.Lithology identification is one of the most important aspects in reservoir characterization,because it is the key to the translation of rock-properties to relevant reservoir parameters,which can effectively reduce the development cost and improve the exploration efficiency.Therefore,in order to solve the problem of lithology identification according to the actual production demand of oilfield,this paper studies three aspects,namely,missing logging curve recovery,multi-frequency feature extraction of logging curve,and the boosting neural network classification of logging data after feature extraction:(1)In order to solve the missing logging data due to the imperfect logging equipment,a logging curve recovery method is studied by using the discrete cosine transform(DCT)frequency division sparse representation.First,the DCT was used to divide the missing training logging curve,and then the dictionary learning was carried out for the logging curve in different frequency bands.Finally,a high precision logging curve recovery method is realized by sparse representation.(2)Aiming at the problem of complicated statistical distribution and mixed random noise,a feature extraction method of multi-frequency feature combination is studied.First,the local statistical features and texture features of the original logging data are extracted.On this basis,DCT technology is used to frequency division processing on feature curve of different attributes and extract frequency signals of different characteristic data.Besides,the feature data of different frequency band components are combined to realize the feature extraction of original logging curve.(3)In order to improve identification accuracy of lithology,a Boosting Neural Network method is studied.The proposed method is based on the traditional artificial neural network(ANN).Through the identification of single lithology,all lithology identification label is obtained.Then different types of lithology labels are weighted to vote,and finally the unknown wellhead of the whole lithology of high accuracy identification.