论文部分内容阅读
本文绘出基于时间延迟神经网络模型的地震油气预测方法及其初步应用结果，不同于通常的孤立模式识别方法．在特征提取阶段，不仅提取地震道中相应目的层单时窗的特征，同时也提取时窗滑动时的特征，这些多时窗的特征信息反映出地层层序的变化．时间延迟神经网络模型通过井旁道特征串的训练，用于表达特征信息与地层含油气情况的复杂关系和特征信息的变化与地层油气聚集的联系．初步应用表明，这种基于时间延迟网络模型的油气预测方法的结果要好于BP网络方法的结果．
This paper presents a method of predicting seismic oil and gas based on time-delayed neural network model and its preliminary application results, which is different from the usual isolated pattern recognition method. In the feature extraction stage, not only the features of single-time window in the corresponding target layer of seismic trace are extracted, but also the features of sliding window are extracted. The feature information of these multi-time windows reflects the change of stratigraphic sequence. Time delay neural network model is used to express the complex relationship between the characteristic information and the hydrocarbon-bearing formation in the strata and the change of the characteristic information and the stratigraphic hydrocarbon accumulation through the training of the string of side-wells. Preliminary application shows that the results of this oil-gas prediction method based on the time-delay network model are better than those of the BP network method.