Maintaining Transformer Equilibrium In The Presence Of Electric Vehicles Using Incremental Learning: Predictive Analytics
No Thumbnail Available
Links to Files
Permanent Link
Collections
Author/Creator
Author/Creator ORCID
Date
2015
Type of Work
Department
Electrical and Computer Engineering
Program
Doctor of Engineering
Citation of Original Publication
Rights
This item is made available by Morgan State University for personal, educational, and research purposes in accordance with Title 17 of the U.S. Copyright Law. Other uses may require permission from the copyright owner.
Abstract
Maximum utilization of transformers has become the norm in deregulated electric markets. As electric vehicles are integrated into the distribution system the demand footprint grows, inducing significant fluctuations in power supply profile, besides degrading quality of service, or instigating frequent local power failures. While traditional forecasting tools have been deployed to control some of these problems, they have low early detection and intervention potential and no data-sensing capabilities under real-time conditions or the abilities to protect against spurious power surges. As electric loading in distribution sector approaches utilization thresholds, the ability to accurately predict local transformer states is a concern requiring immediate attention and research. This work investigates a more effective and a novel approach, 3-stage framework dubbed Smart Learner of Evolving Estimates of Power Events and Reliability, which abstracts electric vehicles' impact on local transformer equilibrium and provides real-time pathways that apply a sliding-window mechanism to manage data streams and knowledge discovery via modeling, state prediction and classification. Automatic power-aware management is embedded in the novel 3-stage framework. The 3-stage framework's input imputes missing data, detects concept drift and samples continuous data to support a Predictive Analytics Engine (PAE). PAE applies a Smart Online Available Power Predictor algorithm that incrementally learns from segmented data in order to classify the transformer states and demonstrate the framework performance. Finally, in a given constant time, when demand violates transformer utilization one-way messages are communicated from the Progressive Control Signaling module to utility agent that modulates load levels. The results of this work show the effectiveness and accuracy of local transformer behavior prediction and classification algorithms using real-world data sets and utility error tolerance. The work highlights the potential for interactive power utilization schemes with intelligent and real-time performance tracking. In summary, it makes a practical advance toward general, but direct, online control with potential to significantly improve local equilibrium issues in any distribution system.