随机森林算法在吉木萨尔页岩油藏中的应用

Application of random forest algorithm in Jimsar shale reservoir

针对页岩油水平井压裂改造评价方法适应性不强,以吉木萨尔页岩油藏为研究对象,将储层物性参数和压裂完井参数相结合,采用随机森林算法对页岩油藏进行压裂水平井应用。基于已压裂水平井的压后评估,提出待压井压裂优化参数并预测压裂井产能水平。首先以吉木萨尔页岩油藏压裂油井三年累产油量为评价指标,基于二级降维法优选5组主控因素确定模型的有效性并采用决策树构建随机森林压后评估模型,然后基于偏依赖图分析多因素影响的页岩油藏压裂水平井产量状况,最后对待压裂井的产能水平进行预测。研究结果表明:单储系数、改造体积、簇间距、压裂段长、加砂量是页岩油水平井累计产量的主控参数,且各主控因素存在最优解区间指导待压井选取压裂参数,数据覆盖面足够大预测的产能模型精度越高。随机森林算法对筛选地质、工程参数以及对后续的水平井大规模压裂改造开发页岩油起到指导作用。

In view of the weak adaptability of the evaluation method for fracturing reconstruction of shale oil horizontal wells,taking Jimsar shale oil reservoir as the research object,the random forest algorithm was applied to fracturing horizontal wells in shale oil reservoir by combining reservoir physical parameters with fracturing completion parameters.Based on the post-fracturing evaluation of fractured horizontal wells,the fracturing optimization parameters of the wells to be killed were proposed and the productivity level of the fractured wells was predicted.First,the three-year cumulative oil production of fracturing oil wells in Jimsar shale oil reservoir was taken as the evaluation index,five groups of dominant factors were optimized to determine the validity of the model based on the two-stage dimension reduction method,and a random forest post-fracturing evaluation model was constructed by using the decision tree.Second,the production status of fracturing horizontal wells in shale oil reservoir under the influence of multiple factors was analyzed based on the partial dependence diagram,and finally the productivity level of the fracturing wells to be killed was predicted.The research results show that the single reservoir coefficient,reconstruction volume,cluster spacing,fracturing section length,and amount of sand addition are the main control parameters for the cumulative production of shale oil horizontal wells,and there is an optimal solution interval for each main control factor to guide the selection of fracturing parameters for the wells to be killed.The greater the data coverage is,the higher the accuracy of the productivity model predicted is.The random forest algorithm plays a guiding role in screening geological and engineering parameters and in the subsequent large-scale fracturing of horizontal wells to develop shale oil.