Application of Grey-Cascade Neural Network Model to Reservoir Prediction

Well-log data and the associated extracted attributes have allowed better description of reservoir heterogeneities and more realistic assessment of oil and gas in place. However, the establishment of a complicated nonlinear relationship between logging attributes and reservoir properties has been a major challenge for working geoscientists. Although back propagation neural network is widely and successfully adopted in reservoir prediction, there have been several problems encountered, such as being slow to converge and easy to reach extreme minimum value. To overcome the shortcomings of traditional BP algorithm, a novel reservoir prediction model is presented which uses grey relational analysis technique to optimize the training samples of BP neural network, and a cascade neural network to achieve a higher speed and a lower error rate in identifying reservoir. The effectiveness of these neural network techniques in well-log interpretation is demonstrated in this paper through a real data example from Tarim Basin in China.