Productivity Prediction Of Multiple-fractured Horizontal Wells In Tight Reservoirs Based On Machine Learning

Tight oil and gas resources occupy a considerable proportion of China’s petroleum reserves,and the proportion of tight oil production has been increasing year by year.Therefore,in the context of increasingly scarce oil and gas resources,tight oil reservoirs have received more widespread attention as an important unconventional oil and gas resource.The tight oil reservoir has dense lithology and poor physical properties,which make its exploitation very difficult.Large-volume hydraulic fracturing technology is the core technology for successful development of tight oil reservoirs,aiming to improve the reservoir’s conductivity and expand the oil drainage area,which has been widely used in the field.There are many factors that affect the productivity of wells after hydraulic fracturing,making it difficult to predict their productivity.Therefore,establishing a machine learning-based prediction model for the production capacity of tight oil reservoirs based on the main controlling factors can further optimize fracturing parameters,improve the design level of fracturing technology,and improve development efficiency,which has significant theoretical guidance and economic value.Based on the reservoir characteristics of volumetric transformation in tight oil reservoirs and the complex distribution features of fracture network control zones,a mathematical model of unstable seepage during volumetric fracturing horizontal wells is established by using the theory of unstable seepage of volume fracturing horizontal wells.The Laplace transform and Stehfest numerical inversion are used to solve the model and obtain its production solution.Based on the solution of the mathematical model of volumetric fracturing horizontal wells,typical production decline curves of volumetric fracturing horizontal wells are drawn and stage division is carried out.The applicability of the production capacity model is verified by fitting the actual well production curve.By drawing the production decline curve and cumulative production curve of the model,the impact of microcrack permeability,oil layer thickness,production pressure difference,number of fractures,crude oil viscosity,fracture half-length,transformation zone width,matrix permeability,comprehensive compression coefficient,fracture diversion ability,elastic energy storage ratio,and eddy current coefficient on the production capacity of horizontal wells is analyzed,and the production decline law and cumulative production variation law of different parameter combinations of volumetric fracturing horizontal wells are revealed.Based on the established mathematical model of seepage in volumetric fracturing horizontal wells,training test data are generated,and the main control factors affecting the production capacity of the three stages,initial high-yield stage,rapid decline stage,and slow decline stage,are screened by using random forest and XGBoost algorithms respectively.Combining the results of the two algorithms,the main control factors of the three-stage production capacity are finally determined.BP neural network and support vector machine are used to predict the production capacity of the three declining stages respectively.By comparing with the data generated by the mathematical model of seepage in volumetric fracturing horizontal wells,the root mean square error(RMSE)and determination coefficient(R~2)are calculated,and an improved method combining the two algorithm models is proposed,i.e.,a combination prediction model.The results show that the combined prediction model is better than the single prediction models of BP neural network and support vector machine in terms of prediction accuracy and stability.An application example of predicting production of two tight reservoirs using the established combination prediction model method is provided.A volume fracturing horizontal well fracturing parameter optimization design method is established,and combined with actual data from a horizontal well in a certain tight reservoir,the fracturing parameters for a completed but unfractured well M are optimized.This study provides technical support for production prediction and fracturing parameter optimization design of tight reservoirs.