基于随机森林算法的叠前流体识别

The pre-stack fluid identification method based on random forest algorithm

虽然叠前反演技术能够获得多种流体识别因子,但是仅利用单一的流体识别因子进行储层预测通常会带来多解性问题。目前根据多种流体识别因子对储层进行综合解释已成为一种新的趋势,但大部分方法对专家及其经验存在较强的依赖。鉴于此,将随机森林算法引入储层流体识别。首先基于测井数据优选输入特征(流体识别因子),并分别研究输入特征数量和不同特征组合对算法预测结果的影响;然后利用该算法对输入特征与井中储层信息之间的非线性关系进行学习;最后根据学习结果对储层进行综合判别,实现多种流体识别因子的综合利用。该算法削弱了单一流体识别因子所引起的多解性,提高了储层流体识别的精度与可靠性。应用实例表明,通过随机森林算法对5种流体识别因子与井中储层信息进行综合学习,达到了对含气储层和含水储层进行准确识别的目的。

Although pre-stack inversion technology can be used to obtain multiple fluid identification factors, multi-solutions are generally caused by reservoir prediction using a single fluid identification factor. At present, the comprehensive interpretation of reservoir based on various fluid identification factors has become a new trend, of which most methods rely on experts and their experience. Therefore, the random forest algorithm is introduced into reservoir fluid identification. Firstly, the input features(fluid identification factors) are selected based on logging data, and the effects of the number of input features and different feature combinations on the prediction results are studied. Then the algorithm is used to learn the nonlinear relationship between the input features and the reservoir information of the well. Finally, according to the obtained results, a comprehensive analysis is performed on the reservoir, thus realizing the integrated utilization of a large number of fluid identification factors. The algorithm weakens the multi-solution caused by a single fluid identification factor, and improves the accuracy and reliability of reservoir fluid identification. The application examples show that five kinds of fluid identification factors and reservoir information in wells are comprehensively explored using random forest algorithm, achieving the accurate identification of gas and water in underground reservoirs.