基于随机森林算法的页岩气多段压裂井产量预测

Yield Prediction of Shale Gas Multi-stage Fracturing Wells Based on Random Forest Algorithm

针对页岩气藏多段压裂水平井产量预测精度不高等问题,通过对储层物性和压裂施工参数分析,引入随机森林算法对页岩气多段压裂井产量进行预测。以涪陵页岩气田具有产气剖面的一期产区多段压裂井为研究对象,全面收集测井解释资料、压裂施工资料、钻井资料,采用二级策略降维法明确了影响产量的主要因素及响应关系,评价了几种产量回归预测模型的预测精度。研究结果表明,测深、垂深、延伸压力、脆性指数、密度测井值和总液量是多段压裂井产量的主要影响因素,且影响程度各异;与支持向量机算法和内核岭回归算法这2种机器学习算法比较,基于随机森林算法构建的产量回归模型具有最优的预测能力,模型决定系数为0.723,测试集均方根误差为0.319,表现了更好的预测效果和泛化能力。

Aiming at the problem that the accuracy of yield prediction for multi-stage fracturing horizontal wells in shale gas reservoirs is not high,through the analysis of reservoir physical properties and fracturing completion parameters,the random forest algorithm was introduced to predict the production of shale gas multi-stage fracturing wells.Taking multi-stage fracturing wells in the first stage of Fuling shale gas field with gas production profile as the research object,comprehensively collecting logging interpretation data,fracturing construction data and drilling data,the main factors affecting the output and their response relations were determined by two-stage strategy dimension reduction method.Finally,the prediction accuracy of several production regression prediction models was evaluated.The results show that depth sounding,vertical depth,extension pressure,brittleness index,density logging value and total fluid volume are the main factors influencing the production of multi-stage fracturing wells with different degrees of influence.Compared with the support vector machine algorithm and the kernel ridge regression algorithm,the yield regression model based on random forest algorithm has the optimal predictive ability,with the model determination coefficient of 0.723 and the test set root-mean-square error of 0.319,showing better prediction effect and generalization ability.