Abstract: The deep reservoir of Dongying sag is a set of red beds under dry sedimentary environment. The reservoir of Ek₁^z has a great exploration potential with large thickness and less drilling. However, the reservoir prediction is very difficult,as the layer structure in the target area is complex, the burial depth becomes deeper from south to north gradually, and the consistency of well-to-seismic relationship is not good. Through a comprehensive analysis of well-to-seismic characteristics, the seismic attributes were optimized and the relationship of well-to-seismic thickness was established. Using the single best attribute method, multiple linear regression method, and learning-type nonlinear reservoir prediction method, the understandings and conclusions are drawn as follows:1) The red bed in the study area has the typical characteristics of thin inter-beds, with higher frequency than the overlying and underlying strata, medium amplitude seismic reflection, low continuity, and low signal-to-noise ratio; 2) Based on the dominant frequency, the newly developed mobility attribute has the best thickness correlation and the maximum amplitude, the root mean square amplitude, zero crossing count,and the energy half-time attribute is also better; 3) Whether it is linear regression or nonlinear support vector machine (SVM) method, the optimized five attributes reservoir prediction effect is better than that of seven attributes; 4) The cross-validation based SVM method is the most suitable one for the thickness prediction of small sample wells in deep reservoirs, and its effect is better than the single best attribute method and the multiple regression method.

Key words: Dongying sag, red beds, well-to-seismic relationship, mobility, multiple linear regression, cross validation SVM, reservoir prediction