| dc.description.abstract | The electrical grid is among the most critical of infrastructures, as it both assures high quality of life and promotes economic growth. Loss of power leads to major economic, social, and environmental impacts. Annual impacts in the U.S. from weather-related outages in the electrical network result in more than $150 billion in lost revenue.
Due to the high level of environmental exposure of the electric utility overhead infrastructure, the most dominant cause of electricity outages is weather impact. About 75% of power outages are either directly caused by weather-inflicted faults (e.g., lightning), or indirectly due to weather-caused increases in equipment deterioration rates (e.g. insulation) leading to subsequent failures. While it is not feasible to prevent severe weather conditions, the impact of severe weather can be significantly reduced, and in some cases even eliminated, by accurate prediction of where faults may occur and what equipment may be vulnerable so that adequate maintenance or replacement mitigation approaches can be deployed.
The assessment of weather impact on electrical grids falls into a group of problems referred as "the Big Data problems". The electric power industry has been deploying a smart grid infrastructure containing anywhere from thousands to millions of measurement points throughout the network. In addition, comprehensive analysis of data not coming from utility infrastructure, such as weather, lightning, vegetation, and geographical base data, which also comes in great volumes, is necessary. This brings out new challenges in dealing with extremely large data sets and using them to improve decision-making. Efficient, predictive, condition-based asset management and outage mitigation requires real-time processing of large volumes of multi-domain data.
The goal of this research is to provide a comprehensive framework for the use of Big Data to assess weather impacts on utility assets. This is accomplished in four primary steps: 1) identifying the relevant weather parameters in relation to the electric power outages and asset deterioration; 2) evaluating electrical grid vulnerability to hazardous weather conditions using advanced data analytics; 3) predicting the risk imposed on the electrical grid by severe weather based on weather forecasts and network vulnerability estimates; and 4) developing optimal mitigation strategies that minimize the risk of weather-related power outages.
In this study a unified data framework is developed that enables collection and spatiotemporal correlation of variety of data sets. The Gaussian Conditional Random Fields (GCRF) algorithm is used for predicting the probability of future outages in the network, given weather forecast data. The temporal and spatial interdependencies between components and events in the network are leveraged for the improvement of prediction algorithm accuracy, and its capability to deal with bad and missing data. The algorithm shows high accuracy of prediction by predicting risk of 64% or higher for all the cases of outages in distribution, and over 74% for all cases in transmission. The prediction results are presented on a geographical map in the form of the Risk Maps. A dynamic asset management system based on optimization is built to reduce the predicted risk of outages and component failure while maintaining predetermined economic investment in periodic asset maintenance. The study approach is tested on real utility data for multiple applications. Two scenarios are observed in this dissertation to demonstrate the benefits of this approach: 1) lightning strikes on or in the vicinity of power lines; 2) the combination of high-speed wind and heavy precipitation causing lines to come in contact with surrounding vegetation. | |
