Abstract:
This paper is the result of oil and gas exploration engineering.
Objective Gas content is a crucial parameter for evaluating shale gas resources potential. Accurate prediction of gas content can guide shale gas exploration deployment.
Methods This study takes the shale of the Qiongzhusi Formation in the Sichuan Basin as an example. Based on measured gas content data and logging data of the shale, a model that combines the Kmeans clustering algorithm and the Light Gradient Boosting Machine (LightGBM) algorithm for predicting gas content in shale reservoirs is proposed, and its prediction results are compared with those of Extreme Gradient Boosting (XGBoost) and LightGBM algorithms.
Results The error rate of the XGBoost algorithm's predictions is 9.76%, with a root mean square error of 0.734 and a coefficient of determination of 0.8714. The prediction results of the LightGBM algorithm show an error rate of 9.48%, a root mean square error of 0.6478, and a coefficient of determination of 0.9427. The prediction results of the Kmeans–LightGBM algorithm show an error rate of 7.96%, a root mean square error of 0.5805, and a coefficient of determination of 0.96.
Conclusions The LightGBM prediction model enhanced by Kmeans clustering features can effectively improve the prediction accuracy of gas content in deep shale reservoirs. Based on the Kmeans–LightGBM algorithm prediction, the gas content of the Qiongzhusi shale ranges from 0.21 m3/t to 13.27 m3/t in a great difference, with the high gas content in the Qiong 2 member in the vertical direction.
下载: