資料探勘應用於臺灣低壓AMI用戶用電行為分析之研究

Abstract

本論文研究目的為提出一資料庫架構，以有效地利用智慧電表基礎建設(Advanced Metering Infrastructure, AMI)所收集之用電資料，並進行資料探勘以瞭解目標用戶群之用電行為。 研究對象為台電低壓AMI用戶約1萬戶，資料庫建構所使用資料為2015年9月至2016年10月期間之用電資料，資料探勘所使用資料為目標用戶群於2016年6月至2016年9月夏季期間經饋線別篩選之用電資料。 研究方法首先將建構資料庫以進行資料聚合與資料預處理，針對目標用戶群以階層式集群分析進行分群，分析群集之用電趨勢，以輸出之結果辨識用戶用電尖峰。 研究所提出的資料庫架構，能夠將原先零散的xml格式用電資料進行聚合，並搭配資料庫索引、固態硬碟(Solid-State Drive, SSD) 等技術有效地提升資料預處理效率。集群分析則能夠根據用戶每日24小時的用電趨勢辨識其用電尖峰與用電行為，並可歸納出典型尖峰與特定時段尖峰，作為發展需量反應相關措施的重要參考依據。

The purpose of this thesis is to propose a database architecture to effectively utilize the power usage data which collected by Advanced Metering Infrastructure (AMI) and to proceed data mining to understand the power usage behavior of target user group. Among approximately 10,000 Low-Voltage AMI users' power usage data from September 2015 to October 2016 is collected to construct the database. The data from June 2016 to September 2016 is later filtered by selecting certain feeder line as target user group for the data mining procedure. Hierarchical clustering is used to analyze power usage trend and recognize peak user load of target user group. The result shows that the proposed database architecture is able to aggregate power usage data, which is originally scattered in xml format. Database indexing and Solid-State Drive (SSD) techniques are also used to increase efficiency of data pre-processing. The clustering analysis recognizes peak user load and power usage behavior of target user group according to daily 24-hour data. Two peak load types can be summarized as typical peak load and peak load at certain period of time, which is also an important reference for developing Demand Response (DR) and other relevant programs.