| This paper makes a study of some common mathematical physical method in stratigraphic division, lithology identification. I dentifying lithology strata, and the actual application of the oil and water layer. Stratigraphical division of the logging interpretation is one of the important links. Artificial layered efficiency is low, and affect by personnel subjective factors. This paper using extremum variance clustering method, activity function method, the formaldehyde method with computer to automatic stratification. The advantages of this method is rapid, quantitative, automatic by different levels; Still can simultaneously according to several different log data provide comprehensive stratification. Of course the layered method effect quality depends largely on the use numerical processing method is appropriate or not.In the process of lithology identification,try to use the intersection method, but can only make lithostrati-graphic classification or points out the change trend, do not achieve good effectl. Therefore, based on the algorithm of the bayesian discriminant method and the support vector machine (SVM), combined with the programming,last achieve computer automatic identify lithology. The effect is obvious well, in this two methods, average coincidence rate reached80%, basically met exploration and development needs, can be used as the main method of logging lithology identification in the production practice.This paper mainly studies the block and the reservoir is called Connaught and kernel oil (Bei301block) in Nantun two section of glutenite reservoir.The reservoir generally belongs to the low porosity, low permeability, low oil saturation, and many faults; the lithology is mainly green gray, gray argillaceous sandstone, siltstone, sandstone, conglomerate, coarse sandstone, conglomerate, with unequal thickness layers.In the understanding of Bei301block of regional geological survey foundation, this article from the logging responses of reservoirs and reservoir lithology feature proceed with, combine the commonly used mathematical and physical methods to the study of block logging response parameters to establish a mathematical model, and then of stratum and lithology identification.Such as the activity function layer, select in the mud layer and sand layer in response to the more obvious characteristics of natural gamma curve, and set the window of length3,7,9and three group activity curve to compare recognition, the window length7and9layered effect is better. In the use of Bayesian method for lithology identification process, feature selection is very important, through analysis and lithological response characteristics are derived sandy conglomerate AC (acoustic logging response mean)、sandy conglomerate ILD (deeply should logging response mean)、sandy conglomerate ILM (medium induction logging response mean)、SFL(spherical focused resistivity)、GR (natural gamma ray logging response mean) curve response values in different lithology on changes in the obviously, this is integrated the five log values as characteristic parameters to establish Bayesian discriminant model, and used for lithology recognition, discriminant results and logging lithologic data comparison was sentenced to rate exceeds80%, show discriminant method has practical value.Through this study, mainly the following results were obtained and understanding:(1) The form and change rule of log not only control by lithology, but also connect to underground instrument types, logging velocity, borehole conditions and so on. But they can be summed up to two basic types with the center of symmetry and asymmetric.(2) This paper use the extreme of the variance of theory is simple, easy to program, fast calculation speed, less interaction, layered value result is reasonable, and the effect is obvious. Besides its applicability is stronger, for any logging curve have applied,such as SP, GR, AC, DEN, CNL, CAL, RA.(3) At present, logging curve activity is mainly used in stratified. But with a maximum value formation interface as activity only applies to some of the logging curve, but like natural potential, natural gamma this kind of half of the logging curve is layered point, error will not be huge. For this kind of lateral and gradient not half range point well logging curves of layer, is not so obvious. To use activity layered, must choose the suitable for well logging curves, the appropriate activity window long, can get satisfactory effect.(4). In this paper, the lithology recognition needs to pay attention to when the data are normalized, eliminate the logging parameters for dimension caused by the difference of potential problems, followed by the selection of logging parameters should be selected to reflect lithology, logging information ability.Establishment of model for testing samples, lithology discriminant function reliability, to determine the lithology discriminant function sample data back to the sentence, the results are as follows:gray argillaceous siltstone, sandstone, sandy conglomerate Correct-judgement rate is respectively85.05%,81.19%,85.7%, the entire layer of the sample is sentenced to a rate of83.33%.Illustrate Bayesian discriminant method in the comprehensive use of geological, logging, core data lithology prediction, can obtain good results.(5). Support vector machine method in solving pattern recognition with small sample, nonlinear and Gao Weizhong demonstrated the unique advantages and good prospects in application.The use of C language procedures, to every kind of lithologic SVM samples for training, and then use the learning after the SVM model to predict reservoir lithology, knowable, using SVM recognition Bei301block formation lithology and practical coring data contrast, the coincidence rate was85%, especially for argillaceous sandstone lithologic classification can achieve this more than90%, the use of the method can identify the complex lithology reservoir geology, improve classification accuracy... |
PDF Full Text Request