A dynamic risk-early-warning methodology of distribution system faults incorporating spatiotemporal imbalanced data distributions

Abstract

Power distribution systems are susceptible to external environmental disturbances. The early warning of potential fault risks in both spatial and temporal scales can assist in maintenance planning and overhaul scheduling for distribution systems, thus their overall reliability consequently. To achieve it, this paper proposes a self-adaptive prediction model for future failure risks in distribution systems, namely the association rules exploration with conditional filter and fitness regulation (AREcffr). In this approach, electrical attributes along with surrounding condition factors are both implemented as inputs. Then, to cope with the commonly-occurred imbalanced data distribution in both two scales when distinguishing risky factors, the conditional importance diagnostic threshold setting and importance diagnostic standard calculation methods set are designed. In that case, the included high-risk-low-probability (HRLP) time series and condition factors in the sparsely distributed input data can be taken into the assessment. Next, to conduct more reasonable measurements of risk levels for those selected risky components, a component importance measure (CIM)-based relative weight analysis model, which is according to the variation of total system risks caused by each risky component rather than its appearance frequency or database share, is established. Finally, an empirical research is presented, and the flexibility and advantages of this risk-early-warning model can be validated and demonstrated in consequence.