A Comparative Study Of Multiple Attribute Sensitivity Methods

The resources exploration has become increasingly complex, and seismic attribute technology has widely extended its application in the fields of lithological interpretation, reservoir evaluation or reservoir description in the state of exploration and dynamic monitor of reservoir in development state. With the introducing and extensively using of new knowledge in mathematics and information science and the rapidly developing of computer technology, more and more seismic attributes based on a method or aimed to a certain study goal have been built. And now the quantity is near200.Although the more attributes mean the more available information we can acquire, their unlimited increase also brings unfavorable effect to the reservoir prediction. The relationship between seismic attributes and the object of prediction is complicated. Different area, depth or reservoir has different geologic conditions and reservoir conditions, and so the sensitivity rate of attributes to the reservoir is different. Even in a same area and a same set of reservoir, different object of study will lead to obvious difference in the sensitive attribute set. Therefore, we have to do an optimum selection from the large quantity of extracted attributes.There are also many kinds of attribute sensitivity methods, and each method is with its own theory of information science or statistics. For this reason, we should find out a reasonable attribute sensitivity method to select attributes. At first part of this paper, we have preliminary come to the available range by the analysis and comparison among the SDC and the methods based on Pearson’s coefficient, Spearman’s coefficient or Significance. Then, the degree that affected by the quantity of samples of each attribute sensitivity method has been concluded by the tests with practical data. By the way of comparison in theory and data test, we have made some advance in SDC, and built a new attribute sensitivity method. It has been proved by the comparison of effect that this new method’s selection of attributes achieves a better prediction result than other four methods.