New Technique to Evaluate the Performance of Hydraulically Fractured Horizontal Wells

Abdalla, Mohamed (KFUPM) | Hassan, Amjed (KFUPM) | Abdulraheem, Abdulazeez (KFUPM) | Elkatatny, Salaheldin (KFUPM) | Mohamed, Abdelmjeed (KFUPM)

OnePetro 

Abstract

The horizontal wells are applied to enhance the production rate by increasing the contact area between the wellbore and reservoir, it has been also used to access the highly heterogeneous and unconventional formations. One horizontal well can produce the same amount of 5 vertical wells with a very competitive cost and operational time. Further improvement for the productivity of horizontal well can be achieved by conducting hydraulic fracture operations, especially for low permeable or unconventional formations. This paper shows a new technique to estimate the performance of hydraulically fractured horizontal wells, without a need for using downhole valves or smart completion.

In the literature, few empirical models have been proposed to evaluate the inflow performance of such wells. However, most of these models assume constant pressure drop in the horizontal section, therefore, significant errors were reported from those models. In this work, a reliable model will be presented to predict the well deliverability for hydraulically fractured horizontal well producing from heterogeneous and anisotropic formation. Different artificial intelligence (AI) methods were investigated to evaluate the well performance using a wide range of reservoir/wellbore conditions. The significant of several parameters on the well productivity were investigated including; permeability ratio (kh/kv), number of fracture stages and the length of horizontal section.

The AI model was developed and validated using more than 300 data sets. Artificial neural network (ANN) model is built to determine the production rate with an acceptable error of 8.4%. The model requires the wellbore configurations and reservoir parameters to quantify the flow rate. No numerical approaches or downhole well completions were involved in this ANN model, which reduce the running time by avoiding such complexity. Moreover, a mathematical relation was extracted from the optimized artificial neural network model. In conclusion, this work would afford an effective tool to determine the performance of complex wells, and reduce the differences between the actual production data and the outputs of commercial well performance software.