基于神经网络的测井曲线砂体解释与应用研究

Sands Interpretation and Generalization with Well Log Data Based on Neural Network

为解决测井曲线人工砂体解释工作的繁琐性与主观性影响,提出一种基于测井曲线中砂体统计参数的自动解释方法,用于测井曲线的砂体分类工作。以工区内已解释砂体的统计参数作为神经网络训练数据,选择孔隙度、渗透率、含气饱和度、泥质含量、砂厚和深度作为输入,输出是对应砂体的解释结果。使用鄂尔多斯盆地某区块1 000余组测井曲线砂体样本数据对神经网络进行训练,选择2组不同的砂体数据测试神经网络的自动解释及其应用效果,结果表明该方法对测井曲线砂体自动分类的正确率分别为92.4%和89.8%。实验数据测试证明,该方法能够自动、快速地实现测井曲线砂体解释,从而协助工程人员进行可靠、高效的砂体识别与分类工作。

Log interpretations for sand reservoir are usually time consuming and subjective because most of their works depend on interpreters.To solve this problem,an automatic numerical method is proposed based on neural network.Training data with reservoir properties as input and log interpretation result from same or nearby region will be used to train the neural network which can interpret new log data for sand reservoir automatically.Over 1 000 sets of log data from Ordos basin are used to test the accuracy and reliability of this method.The generalization effect is also evaluated by two sets of data from different regions.The accuracies of test results in two different regions are 92.4% and 89.8% respectively.So,this automatic sand interpretation method can help interpreter to interpret sand reservoir with log data efficiently and reliably.