Use of a Big Data Mining Technique to Extract Relative Importance of Performance Shaping Factors from Event Investigation Reports

摘要

In this study, the relative importance of significant performance shaping factors (PSFs), which is critical for estimating the human error probability (HEP) of a given task environment is extracted from event investigation reports of domestic nuclear power plants (NPPs). Each event was caused by one or more human performance related problems (i.e., human errors), and its investigation report includes detailed information describing why the corresponding event has occurred. Based on 10 event reports, 47,220 data records were identified, which represent the task environment of 11 human errors in terms of significant PSFs. After that, the relative importance of the associated PSFs was analyzed by using a CART (Classification and Regression Tree) method that is one of the representative techniques to scrutinize the characteristics of big data.