论文部分内容阅读
针对现有核磁共振测井渗透率模型对孔隙结构复杂的致密砂岩储层预测精度不高的问题,在分析误差反向传播神经网络的缺陷后,提出了一种利用集成正则化改进神经网络(BPNN)算法与核磁共振T_2全谱预测致密砂岩储层渗透率的方法。该方法采用自构形算法自动确定隐层神经元的个数,采用自适应雨林优化算法避免BP神经网络迭代陷入局部极小值,利用L_2正则化算子保证算法的稳定性,采用Adaboost集成算法串联若干BP神经网络以提高模型泛化能力。提取某区致密砂岩储层192块岩样的核磁共振T_2全谱数据进行建模,并应用于非建模井的渗透率评价,认为基于集成正则化改进BPNN算法评价储层渗透率精度较高,均方误差仅有0.286。
Aiming at the problem that the existing permeability model of NMR logging is not accurate in the prediction of tight sandstone reservoirs with complex pore structure, after analyzing the defect of neural network with error backpropagation, this paper proposes an improved method using integrated regularization neural network ( BPNN) algorithm and the full spectrum of NMR T 2 to predict the permeability of tight sandstone reservoirs. In this method, self-configuration algorithm is used to automatically determine the number of hidden neurons. The adaptive rainforest optimization algorithm is used to avoid the BP neural network iteratively falling into local minima. The L_2 regularization operator is used to ensure the stability of the algorithm. Adaboost integration algorithm Several BP neural networks are connected in series to improve the generalization ability of the model. The T 2 NMR full-spectrum data of 192 samples from a tight sandstone reservoir in a certain area were extracted and used to evaluate the permeability of non-modeling wells. The results show that the accuracy of reservoir permeability estimation based on integrated regularization improved BPNN algorithm is high , The mean square error is only 0.286.