基于灰色关联投影随机森林算法的水平井压后产能预测及压裂参数优化

Post-fracturing Production Prediction and Fracturing Parameter Optimization of Horizontal Wells Based on Grey Relational Projection Random Forest Algorithm

基于数据挖潜建立了压裂油井产量预测与合理压裂参数优选的灰色关联投影随机森林算法,采用加权的灰色关联投影方法筛选出数据集;采用网格搜索交叉验证方法对森林模型参数进行优化;在此基础上训练随机森林回归模型进行产量预测,结合响应面分析方法对压裂参数进行优化。产量预测实例表明该算法的决定系数0.9189高于随机森林算法、梯度提升决策树算法;三口井日产油实际值与预测值之间的误差为9.6%、1.2%、1.1%。结合新疆玛湖油田实际储层对M1井进行压裂优化:总压裂液量21000 m^(3)、裂缝簇数34条、总砂量1181.86 m^(3)为最佳压裂参数,此时日产油为44.356 m^(3)。

A gray correlation projection random forest algorithm for the production prediction and fracturing parameter optimization of oil wells is establishedbased on data tapping,and the data set is screened out using the weighted gray correlation projection method.On this basis,the stochastic forest regression model is trained to predict the production,and the fracturing parameters are optimized by response surface analysis.The examples of production prediction show that the decision coefficient score of the algorithm is 0.9189,which is higher than that of random forest algorithm and gradient lifting decision tree algorithm;the error between the actual production and the predicted production of three oil wells is 9.6%,1.2%and 1.1%separately.Combined with the actual reservoir of Mahu oilfield in Xinjiang,the fracturing parameters of well M1 are optimized:the total fracturing fluid volume is 21000 m^(3),the number of fracture clusters is 34,the total sand volume is 1181.86 m^(3).The daily oil production under these conditions is 44.356 m^(3).