Dynamic Prediction of Shale Gas Drilling Costs Based on Machine Learning

Shale gas, a significant recoverable natural gas resource trapped in shale formations, represents a significant energy reservoir. Although China has significant recoverable shale gas reserves, the challenge of controlling drilling costs remains a critical barrier to efficient development. This study presents a novel stacked ensemble learning model that integrates support vector machine (SVM) and long short-term memory (LSTM) networks to improve the accuracy of shale gas drilling cost prediction. The methodology consists of three main phases. First, we constructed a comprehensive, multidimensional spatiotemporal dataset of shale gas drilling costs. Second, we used Gradient Boosting Decision Tree (GBDT) modelling to rank the importance of various factors influencing drilling costs. Finally, we developed a stacked ensemble learning model combining SVM and LSTM architectures to achieve superior cost prediction accuracy. Experimental results demonstrate the effectiveness of the model, with the coefficient of determination (R2) improving from 0.25189/0.33834 (traditional SVM/LSTM models) to 0.55934. Model validation using selected well investment data from the Changning Block shows promising performance, achieving a Mean Absolute Percentage Error (MAPE) of 6.41%, with optimal prediction accuracy in the medium investment range (60–70 million yuan). This innovative approach provides a reliable tool for predicting shale gas drilling costs and offers new methodological perspectives for cost reduction strategies. The results contribute significantly to the sustainable development of shale gas resources and provide valuable insights for industry practitioners and researchers in the fields of energy economics and resource management.