基于机器学习的致密气气井动态储量预测模型研究

Study on Dynamic Reserve Prediction Model for Tight Gas Wells Based on Machine Learning

摘要: 精确预测气井的动态储量是致密气高效开发的关键基础。本文针对致密气藏气井动态储量预测问题，基于X致密气藏某区块400口致密气井的实际数据，采用机器学习方法开展致密气井动态储量预测。首先，通过对原始数据进行预处理，其次，利用皮尔逊相关系数法分析，筛选出影响动态储量的主控因素。接着，运用机器学习的方法建立了动态储量的预测模型，最后，使用粒子群优化算法对超参数进行优化。结果表明利用随机森林算法预测结果良好，能够很好地预测气井的动态储量。该技术为提升气井动态储量的预测能力提供了新途径。

Abstract: Accurate prediction of gas wells’ dynamic reserves is a critical foundation for the efficient development of tight gas reservoirs. Aiming at the technical problems in predicting dynamic reserves of tight gas wells, this study uses machine learning methods to predict dynamic reserves of tight gas wells based on the actual data of 400 tight gas wells in a block of the X Tight Gas Reservoir. First, the original data is preprocessed. Second, the Pearson correlation coefficient method is used for analysis to screen out the main controlling factors affecting dynamic reserves. Then, machine learning methods are applied to establish a prediction model for dynamic reserves. Finally, the particle swarm optimization algorithm is used to optimize hyperparameters. The results show that the random forest algorithm provides good prediction results and can effectively predict the dynamic reserves of gas wells. This study provides a new technical means for predicting the dynamic reserves of gas wells.