Analysis and prediction of produced water quantity and quality in the Permian Basin using machine learning techniques

Appropriate produced water (PW) management is critical for oil and gas industry. Understanding PW quantity and quality trends for one well or all similar wells in one region would significantly assist operators, regulators, and water treatment/disposal companies in optimizing PW management. In this research, historical PW quantity and quality data in the New Mexico portion (NM) of the Permian Basin from 1995 to 2019 was collected, pre-processed, and analyzed to understand the distribution, trend and characteristics of PW production for potential beneficial use. Various machine learning algorithms were applied to predict PW quantity for different types of oil and gas wells. Both linear and non-linear regression approaches were used to conduct the analysis. The prediction results from five-fold cross-validation showed that the Random Forest Regression model reported high prediction accuracy. The AutoRegressive Integrated Moving Average model showed good results for predicting PW volume in time series. The water quality analysis results showed that the PW samples from the Delaware and Artesia Formations (mostly from conventional wells) had the highest and the lowest average total dissolved solids concentrations of 194,535 mg/L and 100,036 mg/L, respectively. This study is the first research that comprehensively analyzed and predicted PW quantity and quality in the NM-Permian Basin. The results can be used to develop a geospatial metrics analysis or facilitate system modeling to identify the potential opportunities and challenges of PW management alternatives within and outside oil and gas industry. The machine learning techniques developed in this study are generic and can be applied to other basins to predict PW quantity and quality. [Display omitted] •First study analyzed and predicted PW quantity and quality in the NM-Permian Basin.•5-fold cross-validation showed that the RFR model reported high prediction accuracy.•The ARIMA model showed good results for predicting PW volume in time series.•Machine learning techniques are useful to PW data analysis and prediction.