Reservoir Prediction And Description Based On High Resolution Well Log Data

Reservoir is a place where oil and gas revenues are accumulated, and is a very important research object during the exploration and development of petroleum. In order to get a comprehensive understanding of hydrocarbon reservoir distribution, it is necessary to build a visualized 3D geological model which is in accordance with actual strata situation, and an accurate description of hydrocarbon reservoir is the key point. Well log data is the highest-resolution geological data with the best continuity property at present, and is the sum information of pore structure, fluid property, lithology, as well as lithofacies, which reflects the change of a certain geological parameter with the increase of depth. Well log data plays a particularly significant role in high-resolution sequence stratigraphy especially when outcrop research cannot cover the hole basin area, core is insufficient and the resolution of seismic data does not reach the standard required, Therefore, reservoir prediction and description based on well log data has been an extremely important research subject.This thesis investigated the geological model of reservoir, integrated theory with practice, with the log data of oil well area in North Tarim basin as the research object. Meanwhile, this thesis applied many mathematical analysis and research approaches for reservoir prediction and description, such as Multiscale Analysis, Principal Component Analysis, Support Vector Machine as well as Artificial Neural Network. The research contents and technical route are as follows:1. The thesis did a lot of research on Multiscale Analysis Theory, and applied it to well log data compression, filtering, reservoir singular point detection using maximum norm and multiscale fusion of well logs.2. Principal Component Analysis and Support Vector Machine theories were researched, to combine the merits of the both, this thesis proposed a Least Squares Support Vector Machine (LS-SVM) model based on Principal Component Analysis (PCA) for Reservoir Prediction. Meanwhile, the selection rule of wavelet basis function, the multiscale fusion rule, the selection rule of kernel function and Super-parameters for SVM was investigated, the advantages and disadvantages of both LS-SVM classification model and ANN model was also discussed. The ground work finished is as follows:1. Plenty of documents and research results both in Chinese and English related to geological models of reservoir and research approaches together with their application were studied. In the meantime, monographic study and reports on oil well area in North Tarim basin were referred to so as to get a comprehensive understanding of the geology, reservoir character, types of precipitation facies as well as stratigraphic classification of this area.2. Using many mathematical analysis and research approaches such as Multiscale Analysis, Principal Component Analysis, Support Vector Machine as well as Artificial Neural Network to process well log data of oilfield development and build classification models. The innovative points of this thesis are as follows:1. Multiscale analysis method was applied to reservoir prediction and description, such as well log data compression, filtering, reservoir singular point detection using maximum norm and multiscale fusion of well logs.2. This thesis proposed a Least Squares Support Vector Machine (LS-SVM) model based on Principal Component Analysis (PCA) for Reservoir prediction and classification. Its application at oil well in North Tarim basin has proved to be very effective.