应用长短期记忆循环神经网络的弱反射信号增强方法

Enhancement method of weak reflection signals with long short-term memory recurrent neural network

由于沉积环境的特殊性和复杂性,地下介质中不同反射界面的波阻抗差可能差异巨大。如果储层的有效反射信息较弱,在地震数据中极可能被强反射信息掩盖,不易被识别,影响了储层识别效果,因此亟需一种解释性处理技术突出弱反射信息。常规方法一般是先从地震数据中分离出强反射分量,再将它削弱或删除。但如果地震子波提取不准确,减去法中强反射残留会引入虚假信号。文中提出了一种“升弱降强”的新思路,通过构建幂次反射系数映射模型缩小弱反射信号与强反射信号的相对差异。首先计算测井反射系数的幂次反射系数,将弱反射系数相对增大、强反射系数相对减小,得到拟反射系数序列;再用原始反射系数序列和拟反射系数序列分别与地震子波进行褶积运算,得到合成地震记录和拟合成地震记录,生成训练样本集;然后用该样本集训练长短期记忆(LSTM)循环神经网络,建立合成地震记录与拟合成地震记录的映射关系;最后将该网络应用于地震数据,增强了地震弱反射信号。模型和实际数据应用结果表明,该方法能有效增强地层本身引起的弱反射信号,提高地震数据的储层识别能力。

Due to the particularity and complexity of the sedimentary environment,the wave impedance difference of different reflecting boundaries in underground media may differ greatly.Weak effective reflection information of a reservoir is highly likely to be shielded by strong reflection information in seismic data and is thus difficult to recognize,which affects the identification of the reservoir.Therefore,an explanatory processing technology is urgently needed to highlight weak reflection information.In conventional methods,the strong reflection component is separated from seismic data first and then weakened or deleted.However,strong reflection residue in the subtraction method would introduce false signals in the case of inaccurate seismic wavelet extraction.This paper proposes a new idea of enhancing the weak signal and weakening the strong signal and thereby narrows the relative difference by constructing a power reflection coefficient mapping model.Firstly,the paper calculates the power reflection coefficient of the log reflection coefficient.The weak reflection coefficient is increased relatively,and the strong reflection coefficient is decreased relatively to obtain the pseudo-reflection coefficient sequence.Then,the original reflection coefficient sequence and the pseudo-reflection coefficient sequence are used for convolution operation with seismic wavelets to obtain synthesized and pseudo-synthesized seismic records,with which a training sample set can be generated.The sample set is employed to train long short-term memory(LSTM)recurrent neural networks for establishing the mapping relationship between synthesized and pseudo-synthesized seismic records.Finally,the network is applied to seismic data to enhance weak seismic reflection signals.The application of the model and actual data shows that this method effectively enhances weak reflection signals of strata and improves the ability to identify reservoirs with seismic data.