Development of a Method for Automatic Extraction of Know -How from Historical Information on Operation

我们开发了一种从有关工厂运行的历史信息中自动提取操作诀窍的方法。在该方法中，使用数据挖掘工具之一C4.5将操作知识提取为IF-THEN规则。然后，由于C4.5处理的事件或记录自然是彼此独立的，记录的顺序是没有意义的，因此需要对这样的时间序列数据进行特殊的预处理，以处理与工厂操作有关的历史信息。在这项研究中，我们注意到，操作者不仅关注变量的现值，还关注变量的趋势，并通过判断趋势是定性的增加还是减少来决定他们的操作。因此，我们提出了用不同采样周期之间变量差异的定性值。因此，如果可以对变量进行适当的选择和预处理，肯定可以从有关工厂操作的历史信息中提取操作诀窍。然而，C4.5有时提取了错误的规则，这些规则被因果关系相反的关系解释。在未来的工作中，我们将把C4.5与一个可以解释因果关系的概念结合起来，例如相关系数。

We have developed a method for automatic extraction of operational know -how from historical information concerning plant operations. In this method, an operational know-how is extracted as an if-then rule by using C4.5, which is a one of tools for data mining. Then, special preprocessing should be necessary to deal with such a time series data as historical information concerning plant operations because events or records treated by C4.5 are naturally independent each other and the 'order of records is meaningless. In this study, we pay attention that operators watch not only present values of variables but also trends of variables and make a decision about their operations by judging whether trends are qualitatively increased or decreased. So, we have proposed to use qualitative values of differences of variables between sampling periods. As a result, it has been definitely that operational know-how can be extracted from historical information concerning plant operations if adequate selection and preprocessing of variables may be performed. However, C4.5 sometimes extracted wrong rules that were explained by inverse relations between cause and effect. In future work, we will combine C4.5 with a concept that can explain cause and effect relationship such as correlation coefficients.