Investigating curve smoothing techniques for enhanced shale gas production data analysis

Evaluating shale gas reservoir economic viability remains challenging due to different factors such as long transient flow period and liquid loading resulting in successful shut-ins. Such factors cause fluctuations in production data, with inherent noise impacting analysis methods like decline curve analysis (DCA). In this research, we investigated data smoothing techniques as an alternative to noise removal methods. By applying these techniques, the essential characteristics of the periodic events and signals are retained while reducing the influence of noise making identifying and analyzing patterns easier. Applying seven smoothing techniques to three shale gas datasets with different noise levels to investigate their performance, then, utilizing the cluster-based local outlier factor (CBLOF) algorithm to remove noise from the production data, then, applying seven different DCA models to the original, smoothed, and processed data with CBLOF, the study found that smoothing the data facilitated the extraction of the well's signals. Different smoothing techniques exhibited varying spike levels. The goodness of fit was superior using LOWESS and Fast Fourier Transform (FFT) methods compared to Binomial Smoothing. Moreover, each smoothing technique yielded variations in prediction using the same DCA model. Applying the DCA models that commonly underestimate the reserve to the smoothed data led to further underestimations; however, the DCA models that commonly reserve overestimating reserves also leaned towards underestimations. The Duong's DCA model achieved the highest correlation coefficient (R²), whereas the Wang's DCA model recorded the lowest. In conclusion, this research highlights the benefits of smoothing shale gas production data for better analysis.