【摘要】 页岩储层总有机碳含量(total organic carbon,TOC)是页岩生烃潜力及页岩气富集程度的重要参数,其精确预测对油气勘探开发具有重要意义。常规的线性回归方法受到地区以及测井资料之间复杂的非线性关系的影响,存在预测精度有限的问题。为此提出一种Adaboost-WOA-BP预测模型来进行TOC含量的预测,将WOA(whale optimization algorithm)算法优化过的BP(backpropagation)神经网络作为Adaboost(adaptive boosting)算法的弱学习器,集成多个弱学习器进而构建一个强的学习器。优选自然伽马、密度、声波时差等与计算TOC含量相关的敏感测井参数作为预测模型的输入,通过与常规线性回归方法、BP神经网络、WOA-BP神经网络这3种方法进行对比,Adaboost-WOA-BP模型具有更高的TOC含量预测精度,预测TOC与实测TOC符合率达到95%。更多还原

【Abstract】 The total organic carbon content in shale reservoirs is a crucial parameter for assessing hydrocarbon generation potential and shale gas enrichment. Accurate prediction of TOC(total organic carbon) is essential for oil and gas exploration and development.Conventional linear regression methods are limited in their predictive accuracy due to the complex nonlinear relationships among regional and well logging data. To address this issue, a prediction model based on Adaboost-WOA-BP was proposed for predicting TOC content. This model integrates WOA(whale optimization algorithm) optimized Backpropagation neural networks as weak learners within the Adaboost framework to construct a strong learner. Use of optimal natural gamma, density, acoustic time difference, and other sensitive logging parameters associated with TOC content calculation as inputs for the prediction model. Compared to conventional linear regression, BP neural networks and WOA-BP neural networks, the Adaboost-WOA-BP model demonstrates higher predictive accuracy,achieving a 95% match between predicted and measured TOC values.更多还原