Data-driven approach for spatiotemporal distribution prediction of fault events in power transmission systems

The spatiotemporal distribution of future fault events in a power transmission system assists in operational planning and maintenance scheduling. To this end, this paper proposes an environmental attributes-based framework for the spatiotemporal distribution prediction of potential fault events in the system. In this framework, the distribution of future fault events is predicted via the forecasted information of the environmental attributes rather than the electrical attributes. An extensive investigation covering all environmental attributes including the fault causes is presented, and the underlying fault-attribute relationships are explored. Notably, the rare association rule mining is employed to cope with the rare occurred elements in each environmental attribute through five new significance measurements. Next, to distinguish the diverse influence of each environmental element on the reliability of the whole system, the relative weights are developed. Also, the impact of the latent erroneous predictions of the events caused by one fault cause on the overall prediction performance is assessed via an extended definition of the component importance measurement. Ultimately, the efficiency of the modified significance measurements, the prediction performance in the two test cases, and the impact of each single fault cause are validated by an empirical study. The flexibility and the robustness of this framework in real applications are therefore demonstrated.