Research On Interwell Connectivity Evaluation Based On Machine Learning

During the oilfield development process,interwell connectivity is an important indicator for reservoir evaluation.After the development adjustments of rough water flooding,layer-by-layer water injection,well network densification,and polymer flooding,X oilfield has now entered a high water cut period,making it difficult to explore and exploit,with increasingly severe development adjustments in the middle and later stages.Evaluating interwell connectivity has become the key to improving injection and production efficiency.Based on actual geological data of the reservoir,this study clarifies the main controlling factors of interwell connectivity,and establishes and optimizes machine learning models for evaluating interwell connectivity and predicting oil production,to guide the mid-to late-term development adjustments.Analyzing the regional and well areas overview of X oilfield,X1 and X2 independent well areas were selected as study objects.Fully considering geological static and production dynamic parameters,the entropy weight gray correlation method and Spearman coefficient method are used to determine the seven factors that affect the interwell connectivity of X oilfield,including effective thickness,porosity,permeability,casing pressure,oil pressure,wellhead pressure,and dynamic liquid level.A combination of dynamic and static methods was used to carry out inter-well connectivity evaluation research.In terms of static analysis,the production capacity changes,production capacity testing,and water injection response of 29 injection-production wells in X oilfield were analyzed from the perspective of fluid and geometric connectivity.A machine learning method was used to establish a static connectivity evaluation model based on a multi-layer perception,which was used to quantitatively evaluate the static interwell connectivity.The overall static interwell connectivity in the study area is good,with 10 connected layers in X1 well area belonging to the first class,accounting for 59%,and 5 in X2 well area,accounting for 56%.In terms of dynamic analysis,a connectivity coefficient inversion model based on multiple linear regression and elastic network regression was established,and dynamic connectivity was quantitatively evaluated using a selected multiple linear regression model combined with tracer analysis and static connectivity evaluation results.Overall,the interwell dynamic connectivity in X1 well area is good,with a total of 11 connected layers in the first and second classes,accounting for 58%,while the interwell dynamic connectivity in X2 well area is poor,with a total of 3 connected layers in the first and second classes,accounting for 25%.Based on the results of interwell connectivity evaluation,a detailed analysis of injection-production well groups was conducted,and measures such as profile control,water plugging,fracturing and perforation,and injection-production optimization were adopted to formulate corresponding adjustment plans.At the same time,a production forecasting model based on random forest was established to evaluate the effectiveness of the plan.The comprehensive plan increased the recovery rate of the X1 well area by 2.05% and that of the X2 well area by 1.50%.The analysis shows that the injection-production adjustment plan formulated based on the results of interwell connectivity evaluation effectively improves the injection-production efficiency of the reservoir,and proves that the machine learning-based interwell connectivity evaluation method established in this study has practical significance for the formulation and adjustment of water injection development plans in high water cut oilfields.