相似度-遗传神经网络在储层物性预测中的应用

Application of Similarity-Genetic Neural Network to Reservoir Parameters Prediction

传统的测井解释需要建立精确的数学模型,并常伴有严格的条件限制,因此很难得到真实反映储层特性的结果。采用遗传算法与BP神经网络相结合,利用遗传算法的全局寻优特点,优化神经网络的连接权值和阀值,提高网络的训练精度和预测精度,避免了BP算法易陷入局部极小的缺点,提高运算速度。将相似度的概念引入到测井中,定义相似度在测井中的计算公式,提出相似度与遗传神经网络相结合的方法。实例研究表明,预测准确性较高,可以有效控制预测精度,避免因储层差别大而造成的预测精度降低的现象。

Due to the anisotropy of reservoir, using the linear method it is difficult to get actual reservoir characteristics. On the basis of a combine of genetic algorithm and BP neural network, the global random hunting function of the genetic algorithm is used to optimize neural network connection weights and threshold, which enhances the network tranining precision and parameters prediction accuracy, and as well, increases computing speed by avoiding its disadvantages that standard BP algorithm is apt to trap in local minimal solution. At the same time, we define the similarity in well logging and its calculation formula, and propose a predicting method to combine similarity and Genetic Neural Network. Compared with real samples, the predictive accuracy is higher and effectively controlled, enhance the neural network prediction accuracy.