A modified-distance-based minimum spanning tree method for analyzing hierarchical structure of power generation system

In this paper, a modified-distance-based minimum spanning tree (MST) method is utilized in order to analyze the complexity and component interactions of the power generation system. The advanced method improved the traditional Mantegna-Sornette distance via considering the time lag effect and substituting the cross-correlation coefficient with its absolute value. The industrial big data investigated in this paper contain 19 groups of the two-month time series monitoring indices with the sampling time interval of one minute, which are originated from industrial sensors embedded within the power generation equipments in a power plant. The results shows the time lag effects are obvious between the generation load index and some other indices belonging to auxiliary equipments. Besides, from the generated MST using the modified distance, it can be seen that the apparent clustering feature exists in the spanning tree for the indices belonging to the same equipments. Moreover, such advanced MST can overcome the weakness associated with the conventional MST due to its capability of manifesting the connections between anticorrelated indices properly.