时空关联数据的网络化建模分析、因果推理及其应用研究

Spatio-temporal big data refers to those spatial information flows which have the spatio-temporal correlation with each other and change with time. The research on the inference of spatio-temporal big data has drawn much attention in the world and has been widely used in the fields of national defense, industry, medical treatment, transportation, disaster relief and meteorology. At present, the mining of spatio-temporal big data is often aimed at the static spatio-temporal model and can only infer the correlation of spatio-temporal objects locally from the system. They cannot accurately describe a series of dynamic changes such as the material transfer, energy exchange and information transfer in the dynamic system containing spatio-temporal big data. In order to solve the above problems, from the point of view of system theory, this project is fused data mining with complex network methods, based on the dynamic evolution and data-driven, explored the laws of spatio-temporal correlation data and the underlying mechanism. It is proposed a causal relation discovery and network modeling method of spatio-temporal data in order to accurately describe the complex relationship between spatio-temporal big data; Based on the network structure, it put forward the dynamic inference method considering the causal relationship between the objects, construct inference process, obtain the trend of the dynamic systems, and make attribution analysis on the basis of system-generated mutations. Finally, the research methods are applied to real spatio-temporal data such as seismic data, climate data and traffic data for improving the causal inference methodology.

时空大数据是指那些彼此之间具有时空关联、并随时间推移产生变化的空间特征信息流。时空大数据的推理研究在国际上备受瞩目，并在国防、工业、医疗、交通、救灾、气象等领域得到广泛应用。目前对于时空大数据的挖掘往往是针对静态时空模型，并仅能从系统局部推理出时空对象的影响关系，无法准确描述含有时空大数据的动态系统中物质的迁移、能量的交换、信息的传递等一系列动态变化。为解决上述问题，本课题从系统论的角度出发，融合数据挖掘和复杂网络方法，基于动态演化的观点，以数据为驱动，探索时空关联数据的规律及深层次内在机理，提出时空数据的因果关系发现与网络化建模方法，从而准确描述时空大数据之间的复杂关系；依据时空网络结构，提出考虑时空对象因果关系的动态推理方法，构建推理过程，得到动态系统的时空发展趋势，并依据系统产生的突变做归因分析。最后，将研究方法在地震、大气、交通等真实时空数据中得以应用，完善因果推理方法体系。

时空大数据的挖掘与推理研究在国际上备受瞩目，并在国防、工业、医疗、交通、救灾、气象等多个领域得到广泛应用。本项目着眼于时空关联数据的特征，对时空关联数据进行网络化建模、因果推理及其应用研究。首先，研究了时空关联数据的建网方法，并充分考虑了三维空间、尺度等多种因素对于时空网络建模的影响。提出了一种结合时空网络结构特征的综合性相似度计算方法，从时间维和空间维两个维度研究了时空网络的相似性和周期性问题。其次，以数据为驱动，探究了时空关联数据的规律及深层次内在机理，提出基于贝叶斯网络的时空数据因果关系发现与网络化建模方法，从而准确描述时空大数据之间的复杂关系；最后，将上述方法应用在地震、生物、医学、互联网等真实时空数据中，并根据真实时空数据的应用场景，提出相关模型，完善了因果推理方法体系，这对进一步利用时空关联数据进行推理的研究具有重要的理论研究意义和实际应用价值。