Extracting resilience metrics and load composition from utility data

Thumbnail Image
Carrington, Nichelle'Le
Major Professor
Dobson, Ian
Wang, Zhaoyu
Ajjarapu, Venkataramana
Niemi, Jarad
Hu, Chao
Committee Member
Journal Title
Journal ISSN
Volume Title
Research Projects
Organizational Units
Journal Issue
Is Version Of
Electrical and Computer Engineering
Over the years devices such as fuse cards and smart meters have been incorporated into the electric power distribution systems infrastructure to record the status of the system. These devices record the details of outages and load. This thesis shows how this utility data can be processed to offer insights into the resilience and load composition of electric power distribution systems. This thesis quantifies the resilience of power distribution systems using historical utility data. Resilience concerns the power system’s response to stress from an external disruption such as bad weather. A resilience curve describes the accumulated outages and restores that occur in the power system as time progresses in response to the disruption. Several works have used resilience curves to model the power system's response before, during, and after a disruption. We developed a method of systematically detecting and extracting resilience curves from utility data. This allows resilience events of all sizes to be analyzed. It is common to divide idealized resilience curves into distinct time-dependent phases, such as outage and restoration phases. We defined these phases for resilience curves extracted from the utility data and calculated metrics for each phase, such as restoration duration, outage duration, recovery rate, and outage duration. The resilience curves are grouped in small, medium, and large sizes to determine the characteristics of curves with similar sizes. Resilience metrics were extracted from the curves for each group size, giving the probability distribution for each metric and its mean, standard deviation, and percentiles. The quantified uncertainties in the metrics assist utilities with giving upper bound estimates on metrics such as restoration time and customers hours impacted. The extraction of resilience metrics from the resilience curve using phases does not address the issue of outages and restores overlapping in time, and in our data, these processes substantially overlapped. Our approach provides an innovatively simple and effective way to decompose resilience curves into a restore process and an outage process. Metrics for each process can then be calculated. Mathematical formulas were derived to fit the data and extract the metrics, such as outage duration and restore duration, as a function of the number of outages. The variability of duration resilience metrics was calculated from the decomposed processes. For each number of outages, the mean duration and standard deviation determined a gamma distribution and the upper bound of a 95% confidence interval was calculated. A function from that fitting was derived to estimate an 95% upper bound of the duration based on the number of outages. Similarly, we were able to extract the restore and outage processes from resilience curves and derived mathematical formulas for customer hours lost and risk as a function of the number of outages. These new approaches to deriving and analyzing metrics are a novel statistical analysis that works with practical utility data, avoids the complexities of modeling an individual repair process, and applies decomposition to solve the problems of overlapping processes. The thesis also processed some transmission system utility data, showing how to obtain useful detailed records from a public website, and examining the weather impact on cascading outages. We also developed software that processes and analyzes advance metering infrastructure (AMI) data for small utilities. AMI data is a recording of energy consumption for distribution customers at the building level and can record the energy consumption as finely as per minute. A deployable tool was developed to aid small utilities with processing AMI data. One analysis in the tool is the capability of classifying customers based on consumption. This analysis uses k-means clustering to group the customers based on load. A comprehensive breakdown of the load consumption is another analysis feature within the tool. The hourly load consumption is broken down into the contributions of each customer class. The tool was developed to provide small utilities with the capability to clean, analyze and export their AMI data.
Subject Categories