Identification of Optimal Number of Histogram Bars for Large Data Trends

摘要

A statistical method to identify probability density function was previously reported in the literature. Analyses there include constructing a histogram that allows the characteristics of an unknown statistical distribution to be evaluated. Selection of the necessary number of bars, with detailed discussion, is an important issue. As a logical extension of that activity, this paper proposes constructing two sets of symmetrized and nonsymmetrical histograms. Comparing entropies through parameters called entropy coefficients allows the histogram with the optimal number of bars to be found. Additionally, Pearson's criterion is evaluated for verification. Two propositions are formulated in order to generalize the obtained results. This paper presents three cases that summarize the classification, developed after comparing symmetrized and nonsymmetrical histograms, and each case is discussed. Two of the cases demonstrate that symmetrical histograms represent properties of unknown statistical distributions when certain requirements are met. The third case summarizes results from data-trend evaluation, where symmetrization is not appropriate. Conclusions that follow the results are drawn. The presented approach was implemented and tested in a developed set of computer programs for data-trend analysis at the "Kozloduy" nuclear power plant. The source information has been obtained from Units 1 through 4, which are equipped with WER-440/V230 reactors.