基于样本优化的神经网络方法在储层裂缝识别中的应用

常规测井资料解释应用于非常规储层裂缝识别时,存在裂缝识别率低,储层评价不准确等问题;而成像测井方法(FMI)识别效果好,但成本过高。为了提高常规测井裂缝识别的准确率,首先采用BP(back propagation)神经网络方法,建立常规测井参数与裂缝发育程度之间的非线性关系。在神经网络样本选取上,引入K-means聚类算法,依据不同样本特征对其进行优化分类。最后,利用聚类结果分别建立更为精细的神经网络模型,并用于实际裂缝预测。将该方法应用于塔河油田碳酸盐岩储层A探井,识别结果表明:基于样本优化方法的裂缝密度曲线拟合效果(相关系数R分别为0.84、0.89、0.76)明显优于未考虑样本优化方法(R为0.58),验证了本文方法的优越性,可以将其作为一种识别储层裂缝发育程度的新方法。

Application of Neural Network Based on Sample Optimization in Reservoir Fracture Identification

The interpretation of conventional well logging data is a mainstream method of current unconventional reservoir fracture identification, but it is easily affected by factors such as fillings, mud and dissolution, resulting in low fracture identification rate and inaccurate reservoir evaluation. Identification result of formation micro-resistivity scanning imaging logging(FMI) method is better, but the cost of that is too high. Therefore, this paper used back propagation (BP) neural network method to establish the nonlinear relationship between conventional well logging parameters and the degree of fracture development. Aiming at the problem of low fracture identification rate caused by sample selection, K-means clustering algorithm was introduced to optimize samples. The neural network models were established according to the clustering results and used for actual fracture prediction. Taking the prospecting well A of the carbonate reservoir in Tahe Oilfield as an example, compared with the method without sample optimization, this method has greatly improved the identification of fractures and can be used as a new method to identify the degree of fracture development.