基于复合机器算法的致密气井产能预测模型——以鄂尔多斯盆地SM区块为例

Prediction model of tight gas well productivity based on composite machine algorithm:Taking Block SM of Ordos Basin as an example

产能精准预测是致密气藏经济高效开发的关键环节。传统产能预测方法假设条件严苛,因而在实际现场应用存在精度较低的问题。为了进一步提高致密气产能预测的精度,以鄂尔多斯盆地SM致密砂岩气藏压裂直井作为研究对象,综合考虑地质参数、压裂施工参数,应用K-近邻、多元线性回归、支持向量机、随机森林、BP神经网络和复合机器学习算法搭建了气井产能预测模型,并应用机器学习模型分别预测了SM致密气藏的气井产能;测试集气井产能真实值和预测值的均方误差、均方根误差、平均绝对百分比误差和决定系数分别是1.2、1.1、9.5、0.96。实例分析结果表明,复合机器学习模型预测精度高于其他机器学习模型,决定系数为0.97,相较于气井产能公式提高了0.02,适用于对本区块进行气井产能预测,可以实现基于数据的快速预测,最终选择复合机器学习模型作为研究区气井产能预测模型。研究成果可为致密气藏的产能预测提供理论指导。

Accurate prediction of productivity is a key part of its economic and efficient development.Traditional productivity prediction method has strict assumptions,with problem of low accuracy in actual field application.In order to further increase prediction accuracy of tight gas productivity,fractured vertical wells of tight sandstone gas reservoir in Block SM of Ordos Basin is taken as research object.Considering geological parameters and fracturing operation parameters comprehensively,K-nearest neighbor,multiple linear regression,support vector machine,random forest,BP neural network and composite machine learning algorithm are used to build gas well productivity prediction model,respectively predicting gas well productivity of SM tight gas reservoir using machine learning model.The mean square error,root mean square error,mean absolute percentage error and determination coeffi-cient between real and predicted gas well productivity in test set are 1.2,1.1,9.5 and 0.96,respectively.Case anal-ysis shows that composite machine learning model has higher accuracy than other machine learning models,with determination coefficient of 0.97,increasing by 0.02 of gas well productivity formula,being suitable for gas well productivity prediction of this block to realize fast data-based prediction.Finally,composite machine learning mod-el is selected to predict gas well productivity of studied area.The research provides theoretical guidance for produc-tivity prediction of tight gas reservoirs.