Suitability of Different Machine Learning Outlier Detection Algorithms to Improve Shale Gas Production Data for Effective Decline Curve Analysis

Shale gas reservoirs have huge amounts of reserves. Economically evaluating these reserves is challenging due to complex driving mechanisms, drilling and completion configurations, the complexity controlling producing conditions. Decline Curve Analysis (DCA) historically considered easiest method for production prediction unconventional as it only requires history. Besides uncertainties in selecting a suitable DCA model match behavior shale wells, data are usually noisy because changing choke size used control bottom hole flowing pressure multiple shut-ins remove associated water. Removing this noise from important effective prediction. In study, 12 machine learning outlier detection algorithms were investigated determine one most improving quality data. Five them found not suitable, they complete portions rather than scattered points. The other seven deeply investigated, assuming that 20% outliers. During work, eight models studied applied. Different recommendations stated regarding their sensitivity noise. results showed clustered based factor, k-nearest neighbor, angular factor DCA, while stochastic selection subspace be least effective. Additionally, models, such Arps, Duong, Wang less sensitive removing noise, even with different algorithms. Meanwhile, power law exponential, logistic growth model, stretched exponent decline more varying performance under outlier-removal This work introduces best combination outlier-detection algorithms, which could reduce related forecasting reserve estimation reservoirs.