Stratigraphic Division And Sedimentary Facies Analysis Based On Logging Curves

With the maturity of logging comprehensive interpretation, it has become a tendency to make full use of different logging curves. This paper takes logging curve as the research object, discuss and realize some key methods in the stratigraphic division and log facies analysis. Primary research achievements and innovation points of this paper are described as follows:1. Take advantage of the analytic method based on object-oriented, through analyzing WIS documents which contain logging curves, parse and store logging curves data accurately. In view of the influence factors of logging data, adopt fixed window length correlation contrast method for depth correction of curve, then use Savitzky-Golay smoothing filter to remove noises, which is aimed at eliminating noises caused by non-stratigraphic factors, which provide a scientific data support for stratigraphic division and log facies analysis.2. According to the small sample properties of core oil fields, and the similarity of same regional geological characteristics, research log layering method based on support vector machine(SVM). On account of inconvenience in calculation and stratigraphic in continuity of traditional multi-classification algorithms, improve a one to one multi-classification method, so that enhance learning and prediction speed of the sample. As every stratum has some continuity points, we apply filter function and step function to recognize stratigraphic interface, further promoting the accuracy of stratigraphic division.3. Combine with the similarity of log facies located at the same region and same layer, study support vector machine(SVM) algorithm based on logging curves, then use it for log facies analysis. Since logging curves reflecting log facies is more and there is correlation between curve data, we employ principal component to analyze and extract principal component of log facies, which not only reduce the sample dimension, but also improve computational efficiency. By comparing results of experiment, show the availability of coding multi-classification method. Meanwhile utilize support vector machine(SVM) coding multi-classification method based on principal component analysis to analyze prediction logging, which has obtained good analysis results.