顾及时空异质性的可解释性集成插值方法研究

时空数据插值是应对时空数据稀疏分布的关键技术。相比于单一建模策略，集成多个模型的策略可以综合不同插值模型的优点来完成学习任务，在一定程度上避免插值模型选择的问题。然而，现有的集成策略受空间异质性和时间非平稳性的统计约束，且无法解释不同地理空间变化过程对集成过程的作用机制。因此，本项目拟结合时空统计和机器学习方法，研究空间异质性和时间非平稳性的统一表达方法、顾及时空异质性的集成插值策略、集成过程的作用强度和响应规则，以期构建顾及时空异质性的可解释性集成插值框架，提高时空分析与建模能力，增强地理过程的理解。本项目的实施有望丰富时空数据挖掘领域的方法体系，为环境监测、交通预测、应急管理等领域提供技术支撑，对于深入理解社会过程和地理现象具有重要的科学意义和现实意义。

Spatiotemporal data interpolation is a key technology to deal with the sparse distribution of spatiotemporal data. Compared with a single modeling strategy, the strategy of ensemble multiple models can synthesize the advantages of different interpolation models to complete the learning task, and avoid the problem of interpolation model selection to a certain extent. However, the existing ensemble strategies are statistically constrained by spatial heterogeneity and temporal non-stationarity and cannot explain the mechanism of different geospatial variation processes on the ensemble process. Therefore, this project intends to combine spatiotemporal statistics and machine learning methods to study the unified expression method of spatial heterogeneity and temporal non-stationarity, the ensemble interpolation strategy considering spatiotemporal heterogeneity, the action strength and response rules of the ensemble process, so as to build an interpretable ensemble interpolation framework considering spatiotemporal heterogeneity, improve spatiotemporal analysis and modeling capabilities, and enhance the understanding of geographic processes. The implementation of this project is expected to enrich the methodological system in the field of spatiotemporal data mining, provide technical support for environmental monitoring, traffic prediction, emergency management, and other fields, and is of great scientific and practical significance for the in-depth understanding of social processes and geographical phenomena.