基于卷积神经网络的油气地震储层预测研究

Research on Oil and Gas Seismic Reservoir Prediction Based on Convolutional Neural Network

采用目前方法对油气地震储层进行预测时,没有提取油气地震储层数据的主成分,无法准确地在油气地震储层预测过程中预测砂厚度和储层厚度,导致方法存在泛化能力差的问题。提出基于卷积神经网络的油气地震储层预测方法,在莱特准则的基础上对油气地震储层数据进行取均值与剔除异常值处理,并采用主成分分析方法提取预处理后油气地震储层数据的主成分,在卷积神经网络中输入油气地震储层数据的主成分,实现油气地震储层的预测。仿真结果表明,所提方法在油气地震储层预测过程中预测的砂厚度与储层厚度与实际厚度相符,表明方法在油气地震储层预测过程中的泛化能力较好。

When the current method is used to predict oil and gas seismic reservoir, the principal components of oil and gas seismic reservoir data are not extracted, and the sand thickness and reservoir thickness cannot be accurately predicted in the process of oil and gas seismic reservoir prediction, resulting in the problem of poor generalization ability of the method. In this regard, a prediction method of oil and gas seismic reservoir based on convolutional neural network was reported. According to Wright criterion, the oil and gas seismic reservoir data were taken as the mean value, and the abnormal value was deleted. The principal component analysis method was used to extract the principal components of preprocessed oil and gas seismic reservoir data. In the convolution neural network, the principal components of oil and gas seismic reservoir data were input to realize the prediction of oil and gas seismic reservoir. The simulation results show that the error between the predicted sand thickness and reservoir thickness and the actual thickness is thin, indicating that this method has good pan China ability and application prospect in the field of oil and gas seismic reservoir prediction.