Machine Learning-Based Production Prediction Model and Its Application in Duvernay Formation

Zekun Guo,Yuepeng Jia,Li Shen,Hongjun Wang,Xiangwen Kong

doi:10.3390/en14175509

Zekun Guo, Yuepeng Jia + Show 3 more

Open Access

https://doi.org/10.3390/en14175509

Copy DOI

Abstract

The production of a single gas well is influenced by many geological and completion factors. The aim of this paper is to build a production prediction model based on machine learning technique and identify the most important factor for production. Firstly, around 159 horizontal wells were collected, targeting the Duvernay Formation with detailed geological and completion records. Secondly, the key factors were selected using grey relation analysis and Pearson correlation. Then, three statistical models were built through multiple linear regression (MLR), support vector regression (SVR), gaussian process regression (GPR). The model inputs include fluid volume, proppant amount, cluster counts, stage counts, total horizontal lateral length, gas saturation, total organic carbon content, condensate-gas ratio. The model performance was assessed by root mean squared errors (RMSE) and R-squared value. Finally, sensitivity analysis was applied based on best performance model. The analysis shows following conclusions: (1) GPR model shows the best performance with the highest R-squared value and the lowest RMSE. In the testing set, the model shows a R-squared of 0.8 with a RMSE of 280.54 × 104 m3 in the prediction of cumulative gas production within 1st 6 producing months and gives a R-squared of 0.83 with a RMSE of 1884.3 t in the prediction of cumulative oil production within 1st 6 producing months (2) Sensitivity analysis based on GPR model indicates that condensate-gas ratio, fluid volume, and total organic carbon content are the most important features to cumulative oil production within 1st 6 producing months. Fluid volume, Stages, and total organic carbon content are the most significant factors to cumulative gas production within 1st 6 producing months. The analysis progress and results developed in this study will assist companies to build prediction models and figure out which factors control well performance.

Highlights

Support vector regression, and multiple linear regression were applied to model cumulative production
In testing set, gaussian process regression (GPR) model shows R2 at 0.8 and root mean squared errors (RMSE) at 280.54 × 104 m3, it means that the model is able to explain 80% of the gas production variance in the study area and on average there is around 280.54 × 104 m3 uncertainty in the prediction of gas production within 1st 6 producing months for each well
Based on data-driven methodology presented in this paper, it is concluded that: Based on grey relation analysis and Pearson Correlation analysis, 8 parameters are selected as input in predictive model, they are Fluid, PROP, Clusters, Stages, Lateral Length, Sg, total organic carbon content (TOC), CGR

Summary

Introduction

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. There are enormous shale resources distributed worldwide and it requires advanced exploration and development strategies to get economic production. Due to the application of horizontal wells and hydraulic fracturing technologies, the shale reservoirs have achieved an economic production, which plays an important role in world’s gas supply. The production performance is influenced by many factors such as geology, drilling, completion. A production model that contains a comprehensive set of variables is required for production prediction and optimization

Objectives

Methods

Findings

Conclusion