基于随机森林算法的煤层气产能模式识别模型

    Pattern recognition model of coalbed methane productivity based on random forest algorithm

    • 摘要: 为探究煤层气井的排采产能特征,合理分配开发顺序,根据沁水盆地柿庄南地区煤层气生产井实际生产资料分析,提取出排采曲线4类特征值:平均日产气量、峰值日产气量、见气到峰值的时间以及生产时间;结合排采曲线形态和4个特征值建立了3种产能模式,分析了3种产能模式的生产特征;利用随机森林算法建立3种产能模式与对应3号煤层的地球物理测井资料之间的非线性关系,通过网格搜索结合交叉验证的方式确定了随机森林模型超参数,建立了以测井曲线为特征向量的产能模式分类预测模型。将预测类别与实际类别对比分析,预测正确率达到91.7%,说明基于随机森林算法的煤层气产能模式识别具有较高的预测精度。

       

      Abstract: In order to explore the productivity characteristics of coalbed methane(CBM) wells and reasonably allocate the development sequence, according to the actual production data analysis of CBM production wells in Shizhuang south area of Qinshui Basin, four types of characteristic values of the drainage curve were extracted: average daily gas production, peak daily gas production, the time to reach the peak and the production time. Combining the shape of the drainage curve and the 4 characteristic values, three production capacity modes were established, and the production characteristics of the three production capacity modes were analyzed. The random forest algorithm is used to establish the nonlinear relationship between the three productivity models and the geophysical logging data corresponding to the No.3 coal seam. The hyperparameters of the random forest model are determined by grid search combined with cross-validation, and the log curve is established. Classify the prediction model for the capacity pattern of the feature vector. Comparing the predicted category with the actual category, the accuracy rate of prediction reached 91.7%. This shows that the CBM productivity pattern recognition based on the random forest algorithm has high prediction accuracy.

       

    /

    返回文章
    返回