基于完备集合经验模态分解-归一化希尔伯特变换的神经网络储层流体识别

Reservoir Fluid Identification Using the Complete Ensemble Empirical Mode Decomposition-Normalized Hilbert Transform and Neural Network

地震资料的瞬时属性包含丰富的地质信息,可用于储层流体的识别。希尔伯特-黄变换目前在地震资料的瞬时属性提取中呈现出了很大的优势,但是该方法中存在模态混叠、频率误差等问题,限制了其应用。基于此,引入了基于完备集合经验模态分解和归一化希尔伯特变换的改进方法有效提取地震资料更具物理意义的瞬时属性。同时,为了提高储层含气性检测的准确性和精度,选取瞬时频率和瞬时振幅构建分频剖面模型,对目标区域的储层特性和含气特征进行描述。再结合测井资料,运用有监督的神经网络实现对储层含气性的自适应高精度识别。实例研究表明,基于完备集合经验模态分解-归一化希尔伯特变换的前向反馈神经网络方法能够根据某一区域地震数据有效预测该区域储层的含气状况。

The instantaneous attributes of seismic data contain abundant geological information and it can be used to identify reservoirs fluids. The Hilbert-Huang transform currently presents a significant advantage in the instantaneous attributes’ extraction of seismic data,but the existence of modal aliasing and frequency error in HilbertHuang transform limits its application. An improved method based on complete ensemble empirical mode decomposition and normalized Hilbert transform to effectively extract the more physical moments of seismic data was introduces. Meanwhile,in order to improve the accuracy of reservoir gas detection,the instantaneous frequency and instantaneous amplitude were selected to construct the frequency profile model to describe the reservoir and gas-bearing characteristics of the target area. Combined with the logging data,the supervised neural network was used to realize the adaptive high-precision identification of reservoir gas. The case study shows that the back propagation neural network method based on complete ensemble empirical mode decomposition-normalized Hilbert transform can identify the reservoir more effectively.