Overcoming data problems in predictive reliability distribution modeling

Creating an accurate predictive distribution reliability assessment model is a data intensive process. Even for a modest sized system, volumes of utility data must be carefully filtered, interpreted and manipulated before predicted values match historical results. This paper addresses several important data issues in reliability modeling. The first issue is how to extract and model various system components. This includes selecting which electrical components are necessary for a reliability model, determining how to best model various types of components, and identifying what critical data are needed. The second issue is how to appropriately reduce or de-couple the model. This includes "full" versus "reduced" (i.e. three-phase modeling), creation of plausible study areas, and representation of substations, transmission fines, and feeder inter-ties. The third issue is how to calibrate the model so that it conforms to historical observations. For each of these issues, strategies are presented to overcome the problems, the impact of various decisions are quantified, and the advantages, disadvantages and tradeoffs associated with different modeling approaches are discussed. Results and techniques are demonstrated on a real utility system using actual data.