The Recognition Of The Sedimentary Facies Based On Well Logging Curves

In order to further discuss the sedimentary relative control relationship ofoil and gas when oil and gas exploration, to analyze the sedimentary facies ofreservoir types, sedimentary environment and its space-time evolution, revealand sedimentary environment of the medium sand body size, distribution,geometry shape and its longitudinal and lateral heterogeneity characteristicsof connectivity, the depositional model is established. So a correctidentification, logging facies is an important part of the oil field explorationand development research.For the automatic recognition of log facies, mainly by means of thecriterion of the Bayes and neural network analysis method, the disadvantagesof these methods are too dependent on mathematical model but ignoring theexperience of the geological experts, this often lead to low accuracyrecognition results. This article embarks from the depositional environmentand lithology characteristics in the study area, first of all, extract the loggingcurve characteristic parameters, and then analysis the relationship betweensedimentary micro facies and the characteristic parameters of logging curve,find the certain corresponding relationship between sedimentary microfaciesand specific characteristics, input the corresponding relation to the expertdatabase as an expert knowledge, which can identify the curve characteristicand the experience data in the expert system matching. the recognition resultsis established on the basis of statistics and expert experience. Through a fieldWza3well and Wza4well test analysis the discriminant accuracy can reach90%.Supplement identification methods, this article also USES the supportvector machine (SVM) method for well logging facies identification. TheSVM method can be very successfully deal with classification and regression problems. Compared with the traditional statistical learning theory, themethods are different, because it is not based on empirical risk minimization(ERM), but based on structural risk minimization (SRM). Its advantages aresimple structure, and technical performance especially in generalizationability is improved obviously, so for small sample problems in reality hasbetter treatment effect.