Improvement of Hydraulic Fracture Predictions by Real-Time History Matching on Observed Pressures

Abstract

Summary A new, fully integrated model of hydraulic fracturing can be used to compare measured and calculated pressures for parameter determination during the fracture treatment to improve the prediction of hydraulic fracture geometry. Sensor data obtained during the course of the treatment - such as wellhead flow rates, fracturing-fluid viscosity, and proppant concentration - are received directly by this model, which takes into account all the essential physical phenomena that influence the pressure response associated with hydraulic fracture growth. At any point during the treatment, the model can be rerun, faster than real time, changing reservoir and treatment parameters until the difference between calculated and measured wellhead or bottomhole pressure (BHP) histories is minimized. Updated predictions of final propped fracture geometry can then be made by running the model faster than real time, using the remaining treatment scheduled as input and adopting the parameters that correspond to the best history match; these predictions may differ substantially from that of the job design, thus providing field personnel with an improved estimate of the final fracture geometry before the treatment is completed, while remedies can still be implemented. Results show history matches of the pressure response and associated fracture geometries for three treatments performed in the Travis Peak formation of east Texas and one performed on a coal seam in the Piceance Creek basin near Collbran, in western Colorado. In addition, this work is contrasted with previous efforts to deduce fracture geometry from pressure-response profiles. Introduction A great number and variety of hydraulic fracturing models have been developed over the past 3 decades.1–7 Many have been applied in various ways to the design and analysis of treatments carried out on a commercial basis by the industry. Nonetheless, routine optimization of hydraulic fracture treatments - namely, the achievement of the greatest production possible for the smallest investment - remains a very elusive and desirable goal. Shortcomings in several areas have hindered the improvement of the fracturing process:the models used for design and analysis in many cases lack adequate descriptions of the physical phenomena;reservoir characteristics and fracturing-fluid rheology, which strongly influence fracture geometry, are often unknown or uncertain;inadequate monitoring of the fracturing treatment may diminish quality control; andinformation that becomes available during the course of the treatment is generally not used for updating the design prediction of final fracture geometry, which is usually generated from limited prefracture information. It is certainly true that substantial progress has been made over the past decade or so: fracture modeling has evolved from the early specialized constant-height formulations (Khristianovich-Geertsma-de Klerk1,2 and Perkins-Kern-Nordgren3,4) to the more general, fully three-dimensional (3D) simulators,5–7 as described in Ref. 8; minifracture and microfracture tests are occasionally performed to determine reservoir characteristics, such as in-situ stress distributions and the extent of fluid leakoff9–11; service companies have updated their on-site monitoring capabilities to ensure better execution of treatment designs11,12; and certain aspects of fracture creation (e.g., fracture-height containment) are determined from the pressure response during the treatment.9 These efforts have doubtless produced some improvements in treatment effectiveness. But even the most modern comprehensive simulators cannot accurately predict fracture geometry if pertinent reservoir characteristics are unknown; nor can the extent of fracture containment be inferred from the pressure response if rheological changes and sand staging are not taken into account. The optimization of hydraulic fracture treatments, therefore, requires substantive across-the-board improvements in which all the relevant capabilities in fracture modeling, data acquisition and interpretation, and field operations are synthesized into one coordinated system. Therefore, we propose a comprehensive methodology to improve hydraulic fracture prediction and to provide the basis for intelligent decision-making during the treatment. This methodology essentially involves detailing monitoring and real-time simulation and analysis of the fracturing process.13 It is based on the premise that the actual treatment record, and the information inferred primarily from the pressure response during the treatment, can be used to improve estimates of fracture geometry significantly over those derived from prefracture data and schedules. The essential aspects of this methodology can be summarized as follows.Detailed treatment monitoring (e.g., pressures, flows, rheology, and sand scheduling) and accounting of deviations from job design.Use of monitored sensor data as input to real-time hydraulic fracture models.Determination of unknown reservoir/treatment parameters and affirmation of known quantities by the history matching of observed and predicted response pressures.Best estimation of current fracture geometry using prefracture information, the history-matched parameters, and the real-time data flows.Updated predictions of future job status based on the best current fracture estimates and the remaining (or alternative) treatment schedule.Identification of any treatment pathologies and recommendation of possible remedies. Central to the analysis of the fracturing process during treatment is the real-time model of hydraulic fracturing and the history-matching procedure to determine unknown parameters. A summary of the fracture model is provided; its technical details are thoroughly described in Ref. 8. The major purpose of this paper is to describe the history-matching procedures and to show comparisons with actual field data, emphasizing the application of the model to the improvement of field fracturing operations.