Using Sensor Fusion Techniques to Improve Voltage Instability Predictions Using Local Measurements

Abstract

Voltage collapse is a risk to power systems that arises when a system is loaded to near its power transfer limits. As grids are operated closer to their limits, the risk of this phenomenon occurring increases. This research is an effort to build upon the concepts of the Voltage Instability Device (VIP) to improve predictions about proximity to voltage collapse. The principle behind the VIP device is to linearize the power system using simplifying assumptions and current and voltage measurements local to a load. An issue encountered by using this method is that the predictions by VIP devices throughout the system tend to vary considerably. A sensor-fusion framework is presented that makes use of multiple inputs from a network of nearby sensors to attempt to improve prediction accuracy. The sensor-fusion framework employed is known as Dempster-Shafer Evidential (DSE) Theory. This theory relies on the assignment of probabilities to represent the support for the evidence presented by each “sensor” (i.e. VIP device) location. In this work, a consensus algorithm is developed using measurements nearby and a centrality algorithm is used to rate how central a VIP location is in the system. The consensus algorithm is shown to consistently improve the overall error in prediction by a moderate amount. The centrality algorithm improves the prediction in some cases, but not in larger systems. Overall the research presents a positive, albeit limited, improvement in VIP accuracy. The DSE framework employed shows promise as a method to combine data, however, improvements on the algorithms for assigning evidence would be needed for truly successful implementation.