标题： 基于混合量子神经网络的地震反演 显示英文标题

标题： Seismic inversion using hybrid quantum neural networks

摘要： 量子计算利用量子比特，通过叠加和纠缠来解决经典计算机难以处理的问题，提供了显著的计算优势。 量子机器学习（QML）将量子计算与机器学习相结合，在各个领域具有巨大的潜力，但在地球科学中仍 largely 未被探索。 然而，其进展受到当前NISQ硬件限制的阻碍。 为解决这些挑战，混合量子神经网络（HQNNs）应运而生，将量子层结合到经典神经网络中，以利用两种范式的优点。 据我们所知，这项研究首次通过开发用于地震反演的混合量子物理信息神经网络（HQ-PINNs），将QML应用于地下成像。 我们将HQ-PINN框架应用于叠前和叠后地震数据集的反演，估计P波和S波阻抗。 提出的HQ-PINN架构采用编码器-解码器结构，其中编码器（HQNN）处理地震数据以估计弹性参数，而解码器则利用这些参数根据地球物理关系生成相应的地震数据。 HQ-PINN模型通过最小化输入和解码器生成的预测地震数据之间的不匹配来进行训练。 我们系统地评估了各种量子层配置、微分方法和量子设备模拟器对反演性能的影响，并通过Sleipner数据集的单独和同时反演案例展示了其实用性。 HQ-PINN框架在合成和现场案例研究中一致且高效地估计了准确的地下阻抗，证明了利用QML进行地震反演的可行性，从而为量子计算在地球科学中的更广泛应用铺平了道路。 显示英文摘要

摘要： Quantum computing leverages qubits, exploiting superposition and entanglement to solve problems intractable for classical computers, offering significant computational advantages. Quantum machine learning (QML), which integrates quantum computing with machine learning, holds immense potential across various fields but remains largely unexplored in geosciences. However, its progress is hindered by the limitations of current NISQ hardware. To address these challenges, hybrid quantum neural networks (HQNNs) have emerged, combining quantum layers within classical neural networks to leverage the strengths of both paradigms. To the best of our knowledge, this study presents the first application of QML to subsurface imaging through the development of hybrid quantum physics-informed neural networks (HQ-PINNs) for seismic inversion. We apply the HQ-PINN framework to invert pre-stack and post-stack seismic datasets, estimating P- and S-impedances. The proposed HQ-PINN architecture follows an encoder-decoder structure, where the encoder (HQNN), processes seismic data to estimate elastic parameters, while the decoder utilizes these parameters to generate the corresponding seismic data based on geophysical relationships. The HQ-PINN model is trained by minimizing the misfit between the input and predicted seismic data generated by the decoder. We systematically evaluate various quantum layer configurations, differentiation methods, and quantum device simulators on the inversion performance, and demonstrate real-world applicability through the individual and simultaneous inversion cases of the Sleipner dataset. The HQ-PINN framework consistently and efficiently estimated accurate subsurface impedances across the synthetic and field case studies, establishing the feasibility of leveraging QML for seismic inversion, thereby paving the way for broader applications of quantum computing in geosciences.