基于异质集成的井漏预警模型

Lost circulation early warning model based on heterogeneous integration

钻井井漏事故具有突发性和难以控制的特点。因此,迫切需要建立一种有效的井漏预测方法。将随机森林、支持向量机和反向传播神经网络模型相结合的异质积分器Stacking应用于青海省柴达木盆地英西地区。首先对目标区块的数据集进行处理,运用灰色关联对数据进行相关性分析,选择其中10个相关性高的参数,后设置两层堆叠集成,第一层选择随机森林、支持向量机和BP神经网络模型作为基础学习器,第二层选择逻辑回归模型作为元学习器。结果表明,异质集成模型提高了预测精度(0.981的准确率、0.970的精确率、0.963的召回率和0.960的F 1分数),克服了同质分类器的局限性。强调了综合井漏预警预报中考虑多种地质因素的重要性。

Drilling lost circulation accidents are characterized by abruptness and difficulty in control.Therefore,it is urgent to establish an effective lost circulation prediction.In this study,Stacking,a heterogeneous integrator combined with stochastic forest support vector machine and back propagation neural network model,was applied to Yingxi area of Qaidam Basin,Qinghai Province.Firstly,the data set of the target block is processed,and ten parameters with high correlation are selected by grey correlation,and then two layers of stacking integration are set up.The first layer selects random forest,support vector machine and back propagation neural network model as the basic learning device,and the second layer selects logistic regression model as the meta-learning device.The results show that the heterogeneous ensemble model improves the prediction accuracy(0.981 accuracy,0.970 precision,0.963 recall,and 0.960 F 1 score)and overcomes the limitations of homogeneous classifiers.The importance of considering various geological factors in comprehensive lost circulation early warning and prediction is emphasized.