论文部分内容阅读
在利用地震信息属性参数进行砂体预测时,仅用单一参数预测的结果往往精度很差,而盲目使用多参数作为神经网络的输入,又会使网络的学习过程不收敛。为克服上述问题,本文通过理论模型研究,并结合实际地震资料,从时间域和频率域中提取了目的层的19个地震信息属性参数。然后,选取与薄砂层厚度最密切的8种参数进行砂体预测。文中对几种常用的预测方法进行了分析和对比。应用结果表明,多参数的神经网络预测方法的精度最高;主频线性预测和振幅频率反演预测方法次之;振幅线性预测方法的精度最差。另外,当目的层最大砂岩厚度超过λ/4时,用频率域参数预测砂岩厚度比用时间域参数预测的精度高。
In the prediction of sandbody using seismic information attribute parameters, the results predicted by only a single parameter often have poor accuracy, while blindly using multiple parameters as the input of neural network will make the learning process of network not converge. In order to overcome the above problems, this paper extracts 19 seismic attribute attributes of the target layer from the time domain and the frequency domain through the theoretical model and the actual seismic data. Then, the eight parameters with the thinnest sand thickness are selected to predict the sand body. In this paper, several commonly used prediction methods were analyzed and compared. The application results show that the multi-parameter neural network prediction method has the highest accuracy. The main methods of linear frequency prediction and amplitude frequency inversion are the second, and the amplitude linear prediction method is the worst. In addition, when the maximum sandstone thickness of the target layer exceeds λ / 4, the prediction of sandstone thickness by the frequency domain parameters is more accurate than the prediction by the time domain parameters.