CUSTOMER SEGMENTATION AND VIRTUAL METER DEVELOPMENT BASED ON PROBABILISTIC MODELING OF RESIDENTIAL LOAD PROFILES THROUGH AMI

点击翻译按钮获取中文摘要

Along with the growing inclusion of smart technologies into the electrical power grids, benefits, which can beoriginated from advanced metering infrastructure (AMI), have grabbed noticeable attention from distributionutilities. AMI is an architecture for automated, two-way communication between customers' meters and theutility; from the utility company point of view, AMI can provide real-time data of power consumption whichcan be efficiently used for several tasks such as network monitoring. Saskatoon Light & Power (SL&P) whichis a distribution utility in Saskatchewan has been operating the AMI system since July 2016. As the number ofmeters are severely ample in practical systems, SL&P, similar to other utilities, creates virtual meter data byaggregating loads served by distribution transformers. Although this process helps the operators to analyze thegrid with ease, it sacrifices valuable information provided by the AMI and confines their applications to thebilling process. Such an important deficiency can be considered as the main factor which motivates thisresearch program to delve more into the residential load profile modeling and find approaches which can yieldto the enhanced development of virtual meters. The long-term goal of collaboration between University ofSaskatchewan and SL&P is to find a novel framework that combines customer segmentation and virtual meterdevelopment processes together so that the end result not only can meet the operators expectations, but also canform a cutting-edge knowledge motivating other researchers to focus on this problem. To achieve this ultimategoal, the short-term goals which triggers starting a collaboration in terms of an engage grant are: (i) to assesscurrent probabilistic methods for modeling residential load profiles and select a method which can address theissue of highly volatile characteristics with respect to the practical needs; and (ii) to develop a framework forcustomer segmentation and virtual meter development based on both machine learning techniques and theprobabilistic modeling conducted in (i). The outcomes of this research are expected to constitute milestones inpower system monitoring, operation and control, and contribute to the development of a more reliable grid.