Abstract: Fourteen non-linear models are selected to establish a model base, and according to the prediction error analysis of these models during a period of time, the optimum non-linear model is selected. One-order smooth exponential model and neural network model are applied to predicting, correlating and analyzing oil production by polymer flooding process. Optimization arithmetic of objective function is introduced to determine the optimum combination of these prediction methods. The results show that the prediction value of network model and combining optimized model are well closed to the actual curves and better than non-linear optimized model and oneorder smooth exponential model. The reason is that the two models can comprehensively include many factors influencing oil production so that the prediction values are in more accordance with the actual production performance by polymer flooding. And the combining optimized model is established based on maximizing the information values, which comprises all the information of single model and has better prediction effect.

Key words: combination, model, prediction, optimization, oil production