工业生产过程数据稳定性评价方法的适应性研究

A Study of Adaptability of Different Statistical Methods for Stability Assessment of Complex Industrial Data

在分析工业生产过程的多形态数据特征基础上,模拟具有不同形态特征、不同稳定性等级的工业生产过程数据,对比分析6种常用方法对不同形态工业生产数据稳定性分析的适应性。研究结果表明,极差法、偏差程度(z-分数)法、假设检验p值法、方差法、信息熵法仅适应于部分工业过程数据的稳定性分析,且信息熵法所得的量化指标值与生产过程时间有关;变异系数法具有更大范围的适应性,可定量评价具有典型结构特征的工业生产过程数据稳定性。卷烟制丝生产数据分析结果表明,变异系数法适用于卷烟制丝等工业生产过程的稳定性定量评价。

Six widely-used methods for stability analysis of simulated data with multi-morphological features and different levels of stability were theoretically compared,and the adaptability for statistical assessment was further studied to suggest optimal selection of these methods,which include range method,variance method,deviation degree method,p-value method,information entropy,and CV method.After summarizing the five different cases of multi-morphological data,an integrated strategy was proposed to simulate the variation of these data and then strictly evaluate the objectivity of these methods used for stability analysis.It shows that the range method,deviation degree method and p-value method are not suitable for the stability analysis of industrial production process data with high complexity.In contrast to these three methods,the performance of variance method and information entropy method is improved for the analysis of some types of data.Fortunately,the CV method has high robustness and wide adaptability to all the five cases of multi-morphological industrial data.The analysis of real data obtained from cigarette processing process further validate the results and conclusions introduced above.The results reported in this study provide an efficient way for stability analysis of industrial data with high complexity.