顾及置信概率稳定的中误差归一化算法研究

Research on normalization algorithm of mean square error considering the stability of confidence probability

随着地理信息领域软硬件快速发展,地理信息大数据成为支撑多项重大工程的主要数据源。对海量数据成果进行高效、可靠的质量评价一直是研究热点。位置精度作为地理信息数据质量的核心指标之一,表征数据坐标与真实坐标相似度量程度,中误差是目前位置精度的主要量化指标。本文对检测点样本量、中误差及置信概率间关系开展对比研究,基于t分布-卡方分布尺度转换提出面向同一置信概率的中误差归一化算法。通过试验验证,在68.3%置信概率下对不同检测点样本量的位置精度尺度进行了稳定表征,对不同尺度下地理信息数据位置精度表征稳定性进行了探索,为不同检测点样本量下位置精度的高可信比较统一了尺度,有利于增强地理信息数据位置精度比较的公平性。

With the rapid development of hardware and software in the field of geographic information,big data of geographic information has become the main data source supporting multiple major projects.Efficient and reliable quality evaluation of results for massive data has always been a research hotspot.The positional accuracy of geographic information data is one of the core indicators of geographic information data quality,which characterizes the degree of similarity between data coordinates and real coordinates.Mean square error is currently the main quantitative indicator of positional accuracy.This article conducts a comparative study on the relationship between sample size,mean square error,and confidence probability of detection points,and proposes a mean square error normalization algorithm for the same confidence probability based on the t-distribution-chi-distribution scale transformation.Through experimental verification,stable representation of the positional accuracy scale for different sample sizes of detection points was achieved at a confidence probability of 68.3%,and the main issue of the stability of positional accuracy representation for geographic information data at different scales was explored.Unified the scale for high reliability comparison of location accuracy under different sample sizes of detection points,which is conducive to enhancing the fairness of location accuracy comparison in geographic information data.