应用时域卷积神经网络的地震波阻抗反演方法

地震波阻抗反演是储层预测研究的一种重要手段,线性地震波阻抗反演方法求解精度依赖于初始地质模型,而完全非线性方法可望得到高精度求解结果。有鉴于此,首先利用全卷积神经网络、因果卷积、膨胀卷积和残差块构建一个时域卷积神经网络(TCN),以建立地震数据与波阻抗之间的非线性映射关系;然后通过该网络对样本进行训练得到反演映射模型,进一步将地震数据输入该模型得到地震波阻抗。正演数据及实际数据测试结果表明,所提方法实现了地震数据到地震波阻抗间的映射,为地震波阻抗反演提供了具有并行计算能力和自适应结构的智能化方法,并在港2025区块砂泥岩储层预测中得到成功应用。

Seismic impedance inversion method based on temporal convolutional neural network

Seismic impedance inversion is an important method for reservoir prediction. The accuracy of linear seismic impedance inversion methods depends on the quality of the initial geological model. To get a high-accuracy solution, one can adopt a completely nonlinear method. In view of this, a temporal convolutional neural network (TCN) is first constructed by using a fully convolutional neural network, dilated convolution, causal convolution and a residual block. On this basis, a nonlinear mapping relationship is established between seismic data and wave impedance. Then, samples are trained by the network to yield an inverse mapping model. Further, seismic impedance is obtained by inputting seismic data into the model. According to the test results of forward data and actual data, the method realizes the mapping between seismic data and seismic impedance. It provides an intelligent method with parallel computing power and adaptive structure for seismic impedance inversion and has been applied in sandstone and mudstone reservoir prediction of Gang 2025 Block.