【摘要】 针对地震综合数据因格式、来源不同导致数据分析效率低的问题，文中开展了基于异构数据融合的地震综合数据分析系统设计研究。通过构建无监督多模态、非负相关特征融合算法，以解决多模态数据共享空间内部特征的融合规律学习和聚类分析；同时针对多模态数据的相关和不相关特征，构建共享学习机制，将私有特征分离后得到具有可靠鲁棒性的模态共享特征；利用深度置信网络在网络平滑约束下将融合后的特征进行学习与分类，以提高数据分析能力。通过设置对照组进行对比测试，使用基于无监督多模态、非负相关特征融合算法的地震综合数据分析模型可以显著提高预测精度和纯度，纯度与精度分别提高了0.05%和0.06%，具有良好的可行性及优越性。更多还原

【Abstract】 Aiming at the problem of low data analysis efficiency of seismic comprehensive data due to different formats and sources,the design and research of seismic comprehensive data analysis system based on heterogeneous data fusion has been carried out. By constructing an unsupervised multi-modal and non-negatively related feature fusion algorithm to solve the fusion law learning and clustering analysis of the internal features of the multi-modal data sharing space;At the same time,the shared learning mechanism is constructed for the related and unrelated features of the multi-modal data,after separating the private features,a reliable and robust modal shared feature is obtained;The deep belief network is used to learn and classify the fused features under the constraints of network smoothness,and improve the data analysis ability. By setting a control group for comparative testing,the use of an integrated seismic data analysis model based on unsupervised multimodal and non-negative correlation feature fusion algorithms can significantly improve prediction accuracy and purity,increasing the purity and accuracy by 0.05% and 0.06%,respectively. It has good feasibility and superiority.更多还原