Fractured vuggy reservoir prediction combined seismic and well logging data based on BP network

It is very difficult to detect and predict the fractured vuggy carbonate reservoirs because of their random and multi-scale spatial distribution. To address this problem, we unified the different sampling interval of seismic and well logging responses of the reservoirs by the mathematical method, within a certain dimension. Then discussed the correlation of them by the multiple linear regression(MLR). On that basis, we established the BP neural network model to predict the effective thickness of the reservoirs. The results shows that the thickness and the developed zone of fracture cavity can be predicted in combination of three dynamic attributes of seismic, i.e. amplitude, 3-D coherence data and attenuation factors of frequency. We conclude that the method to predict the thickness of reservoirs is very practical with an effective and believable result and it could be widely used in predicting other parameters of carbonate reservoirs as well.