The Sensitivity Prediction Of The Ultra-deep Reservior Based On Artificial Neural Network

Most reservoirs of ultra-deep wells are low permeability formation. Because of the extreme low permeability, it is difficult to carry out core flooding experiments through avaible methods, and very long time will be spend to evaluate formation sensitivity, though no results can be get some times. One hand, the data are also not always precise and can t be repeated easily. On the other hand, the representative reservoir cores are difficult to be obtained, so the evaluation experiments even can t be established. Through existing reservoir lithology, physical character and experiment data that accumulated before, the relation between five-sensitive damage and reservoir lithology, physical character can be found, so the mathematics and physical model can be established and formation damage can be normalized and quantified. Then the formation sensitivity can be predicted and the result had been verified by experiment.Owing to the strong nonlinear mapping self-organizing, self-adapting and self-learning ability of artificial neural network technology, it is especially applicable to solve the problems such as undetermined reasoning, determining, predicting, classifying and so on. After selecting the proper model and parameters, the predicting model can be established. Based on BP net theory, combining artificial neural network and formation sensitivity, this paper has established and educated the simulation predicting model.This paper introduced the situation of formation sensitivity predicting model in home and abroad briefly, and at the same time presented the fundamental of artificial neural network, basic configuration, learning arithmetic and network model of BP network and neural network kit of MATLAB. Several available predicting models have been analyzed and compared, and discussed formation sensitivity predicting model based on BP network.The formation sensitivity predicting model based on BP network has been established, educated and simulated by use of Shengli Oilfield formation sensitivity data. Comparing with conventional regress method, the results neural network predicts are more precise. All above shows that the newly predicting mode exhibits perfect predicting ability. Secondly, by fitting with the datas of Well Shengke1, the first ultradeep well in East China, the validity of the predicting model has been verified.