Prediction of shale gas horizontal wells productivity after volume fracturing using machine learning – an LSTM approach

Abstract

The exploration and development of shale gas is becoming more important owing to the increasing of world energy demand. However, calculating the productivity of horizontal wells after shale gas volume fracturing is always difficult due to various complicated factors. In this study, the long short-term memory (LSTM) neural network was establised and demonstrated to be successful in China complex shale gas production time series prediction. Firstly, the geological characteristics of shale gas and fracturing technology was briefly introduced. Then, a shale gas horizontal well volume fracturing productivity prediction model was established based on a long short-term memory (LSTM) neural network and using actual production data for two shale gas models. The mean absolute percentage error between the predicted results and the actual production data is less than 5%, which indicates a good performance in terms of the prediction of values and trends. Based on this model, sensitivity analysis of the effect of the stimulated reservoir volume (SRV), fracture parameters, permeability, and other factors on the productivity of shale gas wells was carried out. The newly developed LSTM time series productivity prediction method and the insights it provides can be used by reservoir engineers to optimize shale gas field development plans.

Highlights

• A new machine learning (LSTM) shale gas production prediction model is proposed.

• The new machine learning model is better than the traditional RTA or DCA methods.

• The example calculation results show that the LSTM can predict the future production capacity value with a certain accuracy.

• The new model is useful for optimization in shale gas field development.

KEYWORDS:

• horizontal wells
• machine learning
• productivity prediction
• LSTM
• Shale gas

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was mainly supported by the Open Fund (PLC 20180705) of the State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation (Chengdu University of Technology). Some support was provided by the Sichuan Province Education Department (Grant No.18ZB0072). This work was also partly sponsored by the National Natural Science Foundation (Grant No. 51804048) of China.