An Integrated Model Combining Complex Fracture Networks and Time-Varying Data Modeling Techniques for Production Performance Analysis in Tight Reservoirs

Abstract

Efficient development of tight reservoirs often relies on complex-hydraulic-fracture-network. Due to the time repeated iteration for simulation, the (semi)-analytical model or fully-numerical model often requires a trade-off especially for the accuracy of production analysis. Hence, a comprehensive model for accelerating production matching needs to be established. In this paper, a neighboring-long-short-term-memory (n-LSTM) model, integrated with a complex fracture semi-analytical flow model, can make production performance analysis with high efficiency. The interconnections between hydraulic and natural fractures with arbitrary angles and complex geometry were considered in flow model. Then, the reservoir flow derived from Laplace domain was coupled with fracture network flow numerically solved by finite difference method to obtain the semi-analytical flow solution. The specific distribution of flow solutions was obtained based on the range of reservoir properties, well information, and geological parameters. Thus datasets including production rate and date can be constructed, enlarged and split into training and testing dataset. The integrated model proposed in this paper adopted a non-orthogonal network with 4100 feet length and 53 segments for testing, and was applied for the characterization of complex fractures in the Changqing tight reservoir in the Ordos Basin, China. It is worth mentioning that 65 semi-analytical solutions are expanded to 1280 pairs of production-time data point using the n-LSTM model. With the strong power of capture and excavate the non-linear relationship between multitype data, it only takes a few minutes to forecast and match the daily production data with samples from actual oilfield. As a result, the mean square error of 0.31% in the training dataset and 2.63% in the testing dataset shows that the semi-analytical solution that accurately characterizes the complex fracture networks can be combined with improved LSTM for the prediction and analysis of oil production. In addition, it can be found that the prediction results of the integrated model can also identify the 1/4 slope and 1/2 slope straight lines in the log/log transient response curve. The interpreted results expand the application of semi-analytical solution assisted data-driven model and reduce the consumption of a large amount of repetition time. This paper provides an integrated data-driven model combined with semi-analytical model to make well performance analysis with more efficiency and high accuracy. This workflow, incorporated with fracture precise characterization, data generation and expansion, prediction and calibration, can be readily applied in oilfield to obtain fracture parameters with less time. In addition, time-series and small samples can be enlarged and excavated, especially for the in-proper records in production history.