Optimal Identification Method And Software Development Of Cracks In Tight Sandstone Reservoir

Reservoir crack both the oil and gas storage space, but also oil and gastransportation channel, so in the development of oil and gas exploration, with themain position. Tight sandstone layer to form an effective porosity and permeabilityreservoir is due to the presence of a large number of cracks. Tight sandstone reservoirevaluation of the premise that accurate identification of cracks, cracks and cracks thetype of identification, prediction and the method described is particularly important inthe reservoir exploration and development.Single well fracture identification based on conventional logging data has beenare a hot research, but so far still no a way to carry out single well crack a goodrecognition. This study based on based on previous historical experience, summed upthe response characteristics of the conventional logging crack, come to reflect themore obvious well logging coefficient of cracks, single well fracture identificationreference and basis for, the main acoustic logging deep in the dual-induction logging,micro spherical focused logging, neutron porosity logging. This study, the articleintroduces the BP neural network, probabilistic neural network (PNN), support vectormachine (SVM), the wavelet neural network and gray theory analysis. Introduces thebasic principles of these methods, and to the identify of well logging cracks. Projectstudy area, for example, the core observation, based on the response characteristics ofthe reference to crack the observation segment is divided into two kinds of crack andnon-crack, a total of68samples selected, of which44cracks,24non-crack. Take thesample standard deviation normalized for data processing, the selection of parametersto take the property values normally distributed directly normalization, was go on thenumber owned by one of the non-normalized and then normalized so that removesome of the parameters, parameter samples (AC, CNL RLLD) and four parametersamples (AC, CNL, DEN, RLLD). These two parameters sample as modeling sampleswere selected recognition algorithm to model back to the sentence. Four parameters toestablish the BP neural network model, back to the sentence accuracy rate of89.7058824%83.8235294%three parameters; two samples of the probabilistic neuralnetwork back-contracting rate of82.3529%; nuclear function were selected for theSVM algorithm,"polynomial kernel function, the linear kernel, RBF kernel function "," second kernel function”, model to identify accuracy rate of97.0588%; waveletneural network used three parameters of the sample back to sentenceaccuracy85.2941%; gray theory analysis of selected samples of three-parametermodel, the accuracy rate of80.8824%. The project study area has68wells of thecoring data to identify each model algorithm obtained the effect of poor BP neuralnetwork and support vector machine; the effect is good for the wavelet neural network;analysis algorithms gray theory. Can be drawn,5recognition algorithm is feasible andeffective, can be used in tight sandstone fracture identification, provide a basis forfuture research.Existing crack recognition software is the only one identification method, thisstudy is mainly based on conventional well log data, a variety of reservoir fractureidentification method is integrated in one software. Software framework designedusing stand-alone Windows application, a single-document view architecture, whichcontains the BP neural network, probabilistic neural networks, support vector machine,wavelet neural network method, gray theory and drawing module. Developed usingVisual Studio2008integrated development platform, the project type of MFCapplication development language is Visual C++. The software is simple, friendlyinterface, and can call external mapping software for mapping functions. The softwareis simple, friendly interface, and can call external mapping software for mappingfunction. Software through debugging, testing and use, fast response, there was noexception. Can be drawn, complete the study of software functional structure ofexcellent performance, can be used as a research tool for future fracture identificationto facilitate research and improve efficiency.