Reservoir Parameter Prediction Modeling Research Based On Neural Network

Non- linear times series is an attention research field in recent years. It is revealed that the traditional methods are becoming unfit for the modem need, because of the non-linear system is going to be even more complicated nowadays. For the complexity of the non-linear system and the good performance of the ANN, the ANN-based non-linear system forecasting has been found encouraging in this thesis. So this paper presents an artificial neutral network based reservoir parameter prediction model of oil. It has an important meaning to improve extraction ratio of oil field and prolong field developed service life.(1) First, in this thesis an ANN algorithm is worked out, and the algorithms based on GRNN and ElmanNN is presented in details.(2) Secondly, the prediction model is built based on GRNN and ElmanNN, the neural network problems such as stable and popularization capability and so on are research. Besides these, each relative merit of GRNN, RBFNN and BPNN is compared.(3) In application of an artificial neutral network based reservoir parameter prediction model, the difficulty modeling and other problems which may be met in process is analyzed and discussed. The main problems are over many training samples, thus resulting long training time and slow convergence speed.(4) To overcome these defects, an ElmanNN model based on ant colony optimization algorithm is proposed. The simulation results for actual reservoir parameter system show that the proposed prediction model can effectively improve the accuracy and this model is stable and adaptable. Its forecasting performance is far better than that of simply ElmanNN.The research result of this paper express that the method of using neural network of reservoir parameter prediction is valid and viable. So, the reservoir parameter prediction study based on ANN has good realistic meanings to launch the oil wells.