單一感測器智慧電表預測用電模型演算法

Abstract

本研究的目的在於設計一套適用於單一感測器智慧電表的用電解析預測模型演算法，能夠用於預測機器即時用電開關情況與長期用電占比。單一感測器智慧電表和多感測器智慧電表的差異在於感測器數量，多感測器智慧電表可以直觀地透過點對點的方式回報待測功率值，單一感測器智慧電表因為只能偵測到總功率，則需要用預測的方式才能達到用電解析目的。 單一感測器智慧電表用電解析預測的輸入項有1. 總功率 2. 用電特徵。在該類型的預測中常見三種問題：1. 預測目標機器之間的輸出視在功率過於相似2. 預測目標機器的輸出視在功率過小以及3. 預測目標機器為變頻機器(變載機器)。針對這三種問題，本研究中在整體預測模型建構的各個步驟中提出有效的解決或改善方法如分組、MUST & NEVER關聯性、仿效法以及誤差歸咎。將此預測模型演算法應用於某汙水廠的實際案例最高可以得到開關預測準確率96%左右。

The aim of the research is developing an algorithm for single-sensor smart meter which is able to do power disaggregation. Because single-sensor smart meter only has one sensor, it is impossible to monitor all machines by one sensor without other auxiliary methods. Therefore, it is necessary to utilize a model construction algorithm for prediction purpose. There are two inputs in the algorithm: 1. Total power detected by single-sensor smart meter 2. Machine signatures. There are three common problems with prediction of single-sensor smart meter. 1. Similar apparent power values among machines 2. Machine with little apparent power 3. Machine with variable power output. To overcome the problems above, several effective solutions are presented in the paper. The prediction algorithm has been applied to an actual waste water power plant case and the accuracy is around 96%.