Method Research And Software Development Of Identification Of Fracture In Oil Or Gas Reservoir Based On Well Logging

Fracture is not only the oil and gas reservoir space, but also the oil and gas migration channel, so it takes a very important part in the oil and gas exploration and development. Fracture identification, fracture accurate description and fracture prediction in fractured reservoir exploration and development are crucial. Due to a large number of fracture existing, sandstone layer has formed an effective reservoir porosity and permeability. Accurate identification of such fracture is the prerequisite for effective evaluation of reservoir. For carbonate reservoir, fracture is a very common type of storage space, so the type of fracture and fracture identification is very important.In this paper, based on the results of previous studies, we concluded the characteristics of fracture responses of conventional logging and imaging logging. The conventional logging series which are more sensitive to fracture include: the depth of dual lateral logging, micro-ball focused logging, deep in the dual induction log, acoustic logging, neutron porosity logging. These log series are references of fracture identification which are based on the conventional single well logging data. Single well fracture identification based on conventional logging data has always been a hot research all the time, but so far there has been no method which can efficiently identify reservoir fracture alone. In view of this, we introduce the stepwise discriminant analysis, BP neural network, probabilistic neural networks, KNN, SVM, R/S fractal methodology, the recognition principle of these methods and their application to fracture identification based on conventional logging data, of which BP neural networks, KNN, SVM are used for the first time to identify reservoir fracture. Take the region of Xinchang as an application example, referring to section of the fracture observed from the core and the characteristics of fracture response of the conventional logging, classified into two categories, namely, fracture and non fracture, we selected 34 samples, including 28 fracture samples and 6 non-fracture samples. Using the extreme value normalization method, according to "normal distribution of attribute values are normalized directly, non-normal distribution of property values (such as deep lateral resistivity) are first taken logarithm and then normalized", we take the normalization processing measures to the property values of the modeling samples and finally we get the property values which are mapped to [0, 1]. To the normalized sample we do stepwise discriminant analysis, and we get: the critical value of the introduction of variable F 5 to 6, excluding the variable F the critical value of 0, back to the sentence of up to 94.117% accuracy rate, the resulting discriminant function variables including the AC (acoustic time difference), CNL (neutron porosity), RT (deep lateral resistivity). On this basis, we remove the parameters besides AC, CNL, DEN (density), RT, and we get respectively the three parameters (AC, CNL, RT) samples and four parameters (AC, CNL, DEN, RT) samples. Then we take these two samples as modeling samples to establish the corresponding sample identification models of BP neural network, probabilistic neural network, KNN, SVM, and these 2 samples are sentenced back by using corresponding sample identification models. BP neural network recognition model created with three parameters samples has a 88.235% accuracy rate of back to the sentence; probabilistic neural network identification models established respectively with three parameters samples and four parameters samples has the best result of back to the sentence with 82.235% accuracy rate; the KNN recognition model established with three-parameter samples (K = 3, K = 6) and the KNN recognition model established with the four parameters of the samples (K = 3, K = 4, K = 6) have the best result of back to the sentence with 91.176% correct rate ;selecting the kernel function: "linear kernel function", "second kernel function", "polynomial kernel function", "Radial basis function", we established respectively SVM recognition models with the correct identification rate up to 97.06%. These models were used to identify fracture sections in the 14 wells with core material of Xinchang zone, relatively speaking, the best application of have stepwise discriminant analysis, BP neural network; applications better have KNN, SVM; less effective is the application of probabilistic neural network method. Came to the conclusion, these methods are applied to fracture identification of a single well is feasible and effective, and a method basis is provided for future research about fracture identification based on well logging. For R/ S fractal method, taking the 8 wells with core data of Xinchang zone as application example, we calculate R / S values of well logging cure, find the logarithmic curve relationship between R / S and N, and then compared with fracture sections observed from the core layer, we obtained that acoustic time difference and deep lateral induction resistivity are the best consistent with fracture sections observed from the core. Calculating R/S values of acoustic time difference and deep lateral resistivity of layers of each section of each well of the 8 wells and drawing the logarithmic relationship curves between R / S and N, we take the figure concave curve segment as identified fracture. As a result, we find that R/S fractal can identify efficiently most fracture sections observed from the core.The existing most fracture identification software is compiled only for one identification method, in this paper, we use conventional log data to identify fractures in the reservoir. Therefore, in this study, we developed a set of fracture identification software integrated with some identification methods. The software integrates stepwise discriminant analysis, BP neural network, probabilistic neural networks, KNN, R / S method, etc. 5 methods. Software framework is designed with the single document view architecture, stand-alone version of the Windows applications, including stepwise discriminant analysis module, BP neural network module, KNN recognition module, R / S identification module and the drawing module of 6 major modules. Software development platform is Visual Studio 2008 integrated development platform, development language is Visual C + + and MFC Application project type is used. In software development process, a visual interface technology, object-oriented technology and automation interface technology are used. The software is friendly on interface, simple operation, an identification method corresponding to a dialog box, easy to read and write data faster, but also by drawing dialogs to easily call external mapping software. Software has got through debugging, testing and trial. Besides the SVM recognition method, the other 5 methods are applied to fracture identification through the software. The software is easy to operate, read and write data faster, faster response, without any exception in process of being used. Thus, we can get the software full functionality, better performance. It provides a powerful tool to facilitate the researchers to identify fracture and improve efficiency for the later study on fracture identification based on conventional well logging.