Characterization And Evaluation Of Flow Field Based On Flow Field Diagnosis And Machine Learning Method

Water flooding development,as the main means to improve oil recovery,has a very wide range of applications.However,the remaining oil distribution of most reservoirs after long-term water drive development is disordered and dispersed,and it is difficult to effectively understand the production law of water flooding reservoirs,resulting in difficult adjustment of flow field and affecting the efficiency of water flooding development.Domestic scholars characterize and evaluate the flow field by determining the influencing parameters of flow field,which provides support for the decision-making of flow field adjustment.However,the method is subjective.Foreign scholars mostly optimize the water injection system of water flooding reservoirs by streamline simulation,but the convergence of the method is poor under complex geological conditions.Therefore,taking GX1Q1 complex fault block reservoir as an example,this paper characterizes and evaluates the flow field by robust mathematical method,which provides reasonable support for the adjustment decision of complex flow field and improves the water flooding efficiency of high water cut reservoir.This paper presents a set of percolation field characterization and evaluation methods for complex water flooding reservoirs.The contents of the dissertation are as follows:(1)For flow field with multi-well and long production history,according to the fluid flow exchange between grid nodes in numerical simulation results,the time distribution of flow field propagation is calculated by the flow field diagnosis algorithm in the open source toolbox of MATLAB,and the injection and production wells belonging to the grid are divided according to the time of transmission of any well,and the control area of injection and production wells is divided,so as to facilitate the complex flow field.Visual representation was performed.According to the propagation time,the grid nodes in the control area are sorted,and the F-? diagnostic map is made.The flow heterogeneity of the local flow field is evaluated by Lorentz index and flow heterogeneous index.Flow heterogeneity and movable oil reserves are considered to evaluate flooding potential of injection wells,which provides a basis for flow field adjustment.The evaluation of oil displacement potential of injection wells can be based on a single perforation section,which provides theoretical support for the fine adjustment of flow field.(2)With spatial location,flow rate and oil-water volume ratio as characteristics,reasonable clustering number is selected by density peak clustering algorithm,and flow field clustering is carried out on grid nodes.Clustering algorithm regards clustering results as flow field with different displacement capacity,evaluates development potential of local flow field,identifies ineffective water injection cycle or area with potential,and provides macro-adjustment for flow field.Basis.(3)Fit the historical production data of production wells,establish machine learning model,capture the relationship between production volume and historical data of any production wells that may exist.In our method,Daily production of produced wells and daily water injection of injected wells were used as input data to predict future daily production of produced wells.First,the relationship between historical flow rates of injected wells and produced wells was modeled by vector auto regression algorithm,and the generalization performance of the model was improved by regularization coefficient.Through the data before May 2017,the oil production of the produced wells from May 2017 to March 2018 can be predicted with an accuracy of 86.92%,and uncertainty analysis can be carried out to ensure the safety and accuracy of the prediction results.The method can be used to model complex data relations,with high accuracy and short calculation time.The model can be used to simulate the injection efficiency of water injection wells to evaluate the oil production contribution of water injection wells,and to ensure the robustness of the evaluation method when the numerical simulation is difficult to fit.(4)Flow field diagnosis and machine learning evaluation results are comprehensively evaluated by Technology for Order Preference by Similarity to an Ideal Solution(TOPSIS).Because they are based on different assumptions and the evaluation results are similar,they are all effective evaluation methods.Comprehensive evaluation can improve the accuracy of evaluation.Through the flow field adjustment algorithm,the flow field diagnosis,machine learning and flow field clustering module evaluation results are combined.According to the comprehensive evaluation results,the water injection adjustment is carried out.The numerical simulation shows that the oil recovery is increased by 0.3112%in 2 years,and the oil production is increased by 34770 m3.It shows that the water flooding ability can be effectively improved by adjusting the flow field through the evaluation results.Finally,the flow field clustering and flow field diagnosis function are realized by MATLAB programming language,and the machine learning module function is realized by Python programming language to ensure the practicability and convenience of the algorithm.The evaluation results of the software can provide theoretical support for the comprehensive adjustment of flow field,and further improve the water flooding efficiency and utilization of complex high water cut reservoirs.