The Precise Prediction of the Turbulence Coefficient Based on Neural Network Modeling

Abstract

Skin is the most challenging problem in oil and gas production resulting from near wellbore mechanical damage or non-Darcy effect due to gas turbulence. Mechanical skin is introduced to the pay zone during drilling and completion phase while rate-dependent skin (non-Darcy effect) comes into play as gas production commences. Also turbulence effect may cause a huge pressure drop for oil wells but it is more sensitive in gas production. Rate-dependent skin is caused by contravening basic Darcy's assumptions in gas reservoir and would be sensible as gas starts rushing to the wellbore. It can be used to have an idea about fluid compressibility and flow path tortuosity near wellbore caused by high gas flow rate. Lots of attempts have been directed for computation of rate-dependent skin. Most of them proposed a power mode equation for estimation of non-mechanical skin in gas reservoirs. This includes turbulence coefficient, squared drawdown pressure, flow rate, and deviation coefficient. The turbulence coefficient (D) to be determined needs running a gas well test or at least two pressure flow rate data points. Other conventional methods also can be used for prediction of this parameter. But as a matter of fact, every predictor model may ignore some effective parameters for simplicity, which might deviate the result from reality. Thus, using a better approach including as much as possible effective parameters such as artificial intelligence can result in more accurate results. The authors propose a new technique to estimate value of the turbulence coefficient (D) by using neural networks based on skin factor, reservoir rock, and fluid properties. It is easy to apply and evaluate. The proposed method is validated using field data under variety of conditions. A computed value of D from neural network matches real data pretty well in comparison with conventional correlations.