Research On Forecasting Method Of Production Index Based On Data Mining

The reservoir production indices are important parameters to characterize the development status of the reservoir.The predecessors have carried out a large number of researches in this area,forming various methods represented by analytical methods,empirical methods and numerical simulation methods,which have played an important role in accurately grasping the dynamics of oil fields.In recent years,technologies such as deep learning and data mining have quickly combined with various industrial fields to form a variety of new technologies,providing power for industrial production.In the long-term development process of the oilfield,a large amount of valuable information has been accumulated.How to dig deep into the internal relationship of these data is a common concern of reservoir engineers.In this paper,data mining technology was used to deeply study various data in oilfield development process,and the main production dynamic index prediction method of reservoir was formed,which realized the organic combination of data mining technology and reservoir engineering research.This paper took the historical production data of DXY2S3 reservoir in Shengli Oilfield as the research object,and explored the possibility and method of using long-short-term memory network(LSTM)for data mining to predict the main production indices of blocks and oil wells.Firstly,the process of establishing the sample bank and the process of sample data processing were studied: By analyzing the main production indices of the block and the oil wells,the monthly oil production and water cut were determined as the basic predictive indices;Using the grey correlation analysis and other methods to analyze the production indices,the main factors affecting the monthly oil production and water cut were clarified,and the sample database was established.Then,the sample data was processed by data cleaning,data dimensionality reduction and data normalization.Secondly,the LSTM index prediction model was built: the model input and output form and specific framework were designed,the model training method and the method of avoiding over-fitting of model training were discussed,and the model performance evaluation index was determined.Thirdly,taking DXY2S3 block and its oil well as examples,the established model was trained and tested.In the process,the influence of each model parameter on the running result of the model was studied by using the control variable method,and the automatic parameter adjustment method was designed to determine the value of these parameters.Finally,the difference between the prediction results of the model and the traditional reservoir engineering method under normal production conditions and the change of a certain production index during the forecast period was compared,and a platform for predicting the main production indices of the reservoir was established.The trained LSTM indices prediction model was used to predict oil wells and blocks.The results showed that the relative error of monthly oil production prediction was within 4%,and the relative error of water cut prediction was within 0.3%,which proves that the production index prediction model has high prediction accuracy and strong generalization ability.The comparison between the production prediction model and the classical reservoir production decline model shows that the prediction results of the two models are consistent,and the method can realize the production prediction under given control conditions,and the prediction results are more in line with the actual production situation.