多波地震油气储层的自组织神经网络学习与预测

如何充分挖掘出地震数据的有效信息以更有效预测出地震油气储层的分布情况,一直是业内所追求的目标。多波地震数据所包含的纵、横波数据不仅含有大量对油气敏感的特征信息,而且二者对油气响应存在差异,利用好这种差异可有效降低地震反演中的多解性问题。为此,设计了一种基于自组织神经网络的多波地震油气储层分布预测方案。首先,通过聚类分析优选出对油气响应比较敏感的地震属性,然后,对优选优化后的属性进行多波复合运算提取油气特征信息。最后,根据输入样本属性数据集设计自组织神经网络结构,计算神经元与样本的距离确定最佳匹配单元,更新调整网络权值,完成网络训练,得到预测结果。应用结果表明,基于本方案所预测的地震油气藏分布范围与实际情况基本吻合,有效地降低了反演结果的不确定性,从而验证了自组织神经网络应用于地震油气储层预测的有效性和可行性。

Prediction of seismic oil and gas reservoir using self-organizing neural network from multi-component seismic data

How to use seismic data to effectively predict the distribution of seismic oil and gas reservoirs has always been the goal of the industry.. The P-wave and S-wave attributes in multicomponent seismic data not only contain a lot of information sensitive to oil and gas, but also have different responses to oil and gas. The problem of multiple solutions in seismic inversion can be effectively reduced using of these differences. Therefore, a multicomponent seismic reservoir prediction scheme based on self-organizing neural network was designed. Firstly, the seismic attributes sensitive to oil and gas are optimizated by cluster analysis. Then, the oil and gas characteristic information is extracted by composite operation. Finally, according to the input sample attribute data set, the network structure of self-organizing neural network is designed. The distance between neuron and sample is calculated to determine the best matching unit. The network weights are updated and adjusted to complete the network training. We assess the proposed method on real seismic data. The results show that the predicted results are basically consistent with the actual situation, which reduces differences and decrease in-terpretation ambiguity.