Intelligent Reservoir Management Based On Ensemble Kalman Filter And Machine Learning

In recent years,with the development of intelligent technology,domestic and foreign oil companies have successively regarded "the application of big data and artificial intelligence technology in oil and gas exploration,development and production" as an innovation driving point,and its main goal is to realize intelligent reservoir management.History matching is of great significance to the integrated closed-loop management of oilfields,a key step in intelligent reservoir management.Traditional history-matching methods have been difficult to meet the requirements of integrated closed-loop management of oilfields.History fitting method.On the basis of summarizing and analyzing the research of automatic history matching at home and abroad,this thesis proposes a new method of history matching based on the combination of Ensemble Kalman Filter(En KF)and Long Short-Term Neural Network(LSTM).The short-term neural network is used for data mining,which solves the problem that the prediction is not easy to converge due to insufficient observation data in the process of ensemble Kalman filter history fitting.In this thesis,the automatic history matching of oil reservoirs based on the ensemble Kalman filter method is firstly studied,and it is applied to the one-production-one-injection reservoir model to analyze the history matching results;Improve its accuracy for parameter optimization,and verify its accuracy by predicting the random daily production data of a single well in a certain block;thirdly,in order to make up for the shortcomings of ensemble Kalman filtering,long-term and short-term neural networks are used for optimization,that is,LSTM-En KF history Fit the new method,apply the new method to the previous one-production-one-injection reservoir model again,and compare the historical matching results of ensemble Kalman filter and LSTM-En KF.The results show that the history matching results of LSTM-En KF are closer to the real values and can describe the geological parameters more accurately,the permeability results are close to the real value,and the overall trend of the dynamic data is the same as the real dynamic data,which verifies the feasibility and accuracy of the method for the actual reservoir model.