First Advisor

Robert Bass

Date of Publication

Spring 5-26-2016

Document Type


Degree Name

Master of Science (M.S.) in Electrical and Computer Engineering


Electrical and Computer Engineering




Electric power systems -- Mathematical models -- Analysis, Electric power systems -- Mathematical models -- Design



Physical Description

1 online resource (ix, 93 pages)


This thesis provides a synchrophasor data analysis methodology that leverages both statistical correlation techniques and a statistical distribution in order to identify data inconsistencies, as well as power system contingencies. This research utilizes archived Phasor Measurement Unit (PMU) data obtained from the Bonneville Power Administration in order to show that this methodology is not only feasible, but extremely useful for power systems monitoring, decision support, and planning purposes.

By analyzing positive sequence voltage angles between a pair of PMUs at two different substation locations, an historic record of correlation is established. From this record, a Rayleigh distribution of correlation coefficients is calculated. The statistical parameters of this Rayleigh distribution are used to infer occurrences of power system and data events.

To monitor an entire system, a simple solution would be observing each of these parameters for every PMU combination. One issue with this approach is that correlation of some PMU pairs may be redundant or yield little value to monitoring capabilities. Additionally, this approach quickly encounters scalability issues as each additional PMU adds considerably to computation - for example, if the system contains n PMUs the amount of computations will be n(n-1)/2. System-wide monitoring of these parameters in this fashion is cumbersome and inefficient.

To address these issues, an alternative scheme is proposed which involves monitoring only a subset of PMUs characterized by electrically coupled zones, or clusters, of PMUs. These clusters include both electrically-distant and electrically-near PMU sites. When monitored over an event, these yield statistical parameters sufficient for detecting event occurrences. This clustering scheme can be utilized to significantly decrease computation time and allocation of resources while maintaining optimal system observability.

Results from the statistical methods are presented for a select few case studies for both data and power system event detection. In addition, determination of cluster size and content is discussed in detail. Lastly, the viability of monitoring pertinent statistical parameters over various clustering schemes is demonstrated.


In Copyright. URI: This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

Persistent Identifier