Robustness assessment and enhancement of power grids from a complex network's perspective using decision trees

In this brief, by examining the profile of the failure cascade of power systems, we identify a critical observable parameter, namely onset time, which is the time after which the propagation rate of a cascading failure increases rapidly. Based on the onset time and the scale of the failed grid in a cascading failure event, we categorize each component in a power network into three types, corresponding to three levels of severity of the failed grid upon the initial failure of that component. Moreover, to investigate robustness enhancement of power networks, we propose a decision-tree-based learning model to extract significant network-based features. By utilizing a number of power networks generated by means of edge re-arrangement targeting topology improvement of the original power system, a decision tree is generated. This tree identifies three network features, including average shortest path length, average clustering coefficient, and average effective resistance (distance) to the nearest generator, which exhibit strong correlation with the robustness of the power network. It is shown that using multiple network-based features can effectively enhance the robustness of power networks.