Abstract: With powerful nonlinear mapping and fitting ability of neural network, a production predicting neural network model is constructed. In view of high error, easy to default and other characteristics of oil field production data or data fitting and prediction is not easy to converge, a method of extended training data set and improved mean square error loss function are presented to get remarkable results in oil production fitting.

Key words: fractured-vuggy reservoir, production predict, recurrent neural networks, back propagation algorithm