And Seismic Joint Lateral Prediction Research, Logging Class Attribute

Log data and their derived reservoir parameters such as porosity, permeability, saturation, named log's attributes, are important foundations of describing reservoir attributes in the process of oil field exploitation. As the drillings in a specific survey zone usually are much fewer, it is very difficult to control the variation of reservoir spatially. Based on the statistical relations, the paper has broken away from the limits of conventional inversion method, and set up the relationships between the seismic attributes and the log's attributes to realize lateral prediction of log's attributes, thus the problem mentioned above has been solved reasonablly.By analysing the overseas and domestic research actualities of lateral prediction of log's attributes, the research train of thought is put forward correspondingly. On the bases of probing into extractive methods of log's attributes and their applied fields actively, extractive methods of seismic attributes and their geologic application foundations are mainly studied, simultaneously, the methods to select or search for optimal ones from slews of attributes are also elaborated on of necessity. All the results studied provide a theoretical basis for setting up the relationships between the seismic attributes and the log's attributes. The emphasis of the paper lay on the study of mathematical methods, such as the studies ofmulti-attribute linear regression method and convolutional regression method. Furthermore, the fact that applying convolution operator is closed to inducting new attributes, just the time shifts of the primary attributes, is proved. In addition, a new fast neural network algorithm which is better to reveal the complex nonlinear relationships between the log's attributes and the seismic attributes than regression methods do, demonstrated by testing calculation for model and real data, is also put forward.On the bases of research works on attribute extraction, selection, regression methods and neural network algorithms etc., a corresponding applied software system is developed using VB programming language. Relative ideal effects have been achieved by using the system to do trial application to logs and seismic data from a specific working area. The actual effects of application show that the neural network algorithm is superior to the linear regression method.Along with the development of the research for direct oil discovery, it should be expected that the lateral prediction of log's attributes for real seismic data, by jiont using log data and seismic data to realize, would have wide application foreground.