浅谈测量平差到空间数据分析的可靠性理论延伸

可靠性作为分析结果评价的质量指标，同时也是分析模型构建的优化准则。在传统测绘领域，李德仁于20世纪80年代中期提出了两个多维备选假设的模型误差可区分理论和验后方差选权迭代的粗差剔除方法，发展了测量平差可靠性理论及摄影测量应用。在电磁物理和普适计算支持下，传统测绘技术已经进化为现代地球空间信息科技。随之，可靠性理论内容从测量平差的粗差处理、空间数据分析的异常处理延伸到空间信息服务的可信计算，可靠性分析方法也从统计推断、优化计算延伸到逻辑推理。类比语言学，测量平差的粗差处理为语法分析，空间分析的异常处理为语义分析，空间信息服务的可信计算则为语用分析。简要地给出了空间数据分析的可靠性指标计算方法，分析了优化目标（成本或代价）函数构造的一般准则（物理系统的能量最小化、数据系统的信息量最大化和用户系统的决策风险最小化），指出可靠性理论方法呈现整体模拟趋势（正常和异常的相对整体性，外在数据和内在状态的相对整体性，静态结构和动态行为的相对整体性，人机地系统的相对整体性）。

Extension of Reliability Theory of Surveying Adjustment into Spatial Data Analytics

Reliability is one quality element of analysis results, and also one optimization criteria of analytical models construction. In surveying and mapping, Li Deren proposes two multi-dimensional alternative hypothesis of model errors discrimination and the gross error elimination of posterior error variance-based weight iterative algorithm. With the promotion of electromagnetic physics and ubiquitous computing, the surveying and mapping technologies have evolved into the geospatial information technologies. Accordingly, the reliability theory is extended from gross error processing, outlier analysis until credible service computation, and its analytical method is developed from statistical inference, optimization computation to logical reasoning. With the metaphor of linguistics, gross error proce-ssing can be considered as grammar analysis, outlier analysis considered as semantic analysis, and credi-ble service computing considered as pragmatic analysis. Roughly, here the reliability of spatial data analysis is formalized. Generally, the criterions of object function in optimization are energy minimization of physic systems, information maximization of data systems, and risk minimization of user decision systems. Nowadays, reliability modeling is holistic by considering the dialectical properties of the normal and anomaly, the interior and exterior, the static and dynamic, the subjective and objective. Upon the requirement of eco-social developments, internet-related big data is mostly related to human mental models, which brings us the great difficulties of modeling human irrationality and spatial cognition.