| Since the seismic survey technique been build,many new methods have been proposed, and many new parameters are created at the same time. Researchers are not satisfied to use a single parameter information to analyze the reservoir, multiple information have been combined to describe the relationship between the data and target, and increase the accuracy of prediction. Nowadays, researchers usually using two methods to analyze the relationship between the data and target: The first method, researchers use two or three parameters to form a cross-plot, then use the difference of distribution characteristics between the two target to recognized targets. In another method, seismic attribute slices or sections can be used to reflect the certain characteristics of known targets synthetically, the region which matches the attribute property of the target is potential target. In the practical application, these methods can achieve good results, but they still have some deficiencies: At first, the research targets have lots of attributes, the difference of attributes between target varies. So, different combination of attributes brings different result, some contains high accurate, and some are not. The urgent need for improvement is how to combine more attributes to make the prediction more accurately. Secondly, the creation of discriminant function is easily disrupted by human factors. The expressions are too simple, so that the flexibility and accuracy is low.Since the 1990 s,the theory of SVM developed rapidly and has received much applications in the field of seismic exploration. Support vector machine(SVM) is a supervised machine learning method, which was put forward based on structural risk minimization principle. It helps to get the global optimal solution without under-fitting or over-fitting. Compared with the neural network and traditional pattern recognition methods, SVM do has higher reliability. In recent years, the theory of PSVM is gradually developing, which could keep the discriminant accuracy up with the traditional support vector machine(SVM) and furthermore, greatly improved operational performance. PSVM, therefore, being applied in seismic exploration, can bring more seismic attributes and training samples, and improve discriminant accuracy by reducing the proportion of error data and increasing the amount of data. It also takes some SVM methods which need large amount of data across the bottleneck of the computational efficiency.Based on the above reasons, PSVM can be used to discriminate by combination of some attributes. In this way, the problem of lacking of attributes in discrimination and the problem of low accuracy can be solved.The research contents and results of this paper listed as follows:First, the development history, research status and advantages of SVM and PSVM has been introduced. Then, the research goal and significance have been proposed.Secondly, the theories of SVM and PSVM have been introduced. The advantage and characteristics of the application of PSVM have been summarized by case study.Combine PSVM method and well data, PSVM method can be applied to lithology discrimination and fluid identification to get satisfied results, so that feasibility of using PSVM method can be proved.Thirdly, the application of PSVM method can be branched form the “point” which stand for the well data to the “surface” which stand for seismic data. Form the training set by attributes optimization, and using PSVM to analyze the attribute volume of the targets. Then the favorable hydrocarbon-bearing area can be predicted, and an accurate prediction results can be available. Additionally, extreme points, slopes and other parameters which could describe the shape features of curve can be obtained from four kinds of gas sand models. Consequently, by the application on discrimination of seismic attribute and AVO, the practicability of PSVM method have been proved. This kind of method can perform well in the reservoir analysis. |
PDF Full Text Request