基於電器使用模型之建築耗電預測

Abstract

近年來在一個小範圍如家庭、大樓或社區等建築物耗能管理逐漸成為大家所關心的議題之一，其中如何預測未來的數分鐘至小時的時間的需量(Deamon)耗電來避免尖峰用電過載更是為重要的問題。在傳統的需量預測方法如類神經模型法等往往有時會因為負載耗電行為的劇烈變化造成學習不易甚至造成演算法發散，尤其在微電網中加入再生能源及不同的配電策略等原因可能會造成加入更多不確定的因素使預測精準度下降，為了彌補傳統預測方法的缺失本研究提出一基於非侵入式負載辨識技術(NILM)之需量預測系統，透過NILM技術我們可以利用單一電錶量判別出每一個電器的種類及使用狀態切換的時間，進而建立電器運轉模型用於預測未來某個時間點的能源需量。最後透過模擬實驗的分析本研究所提出的預測方法在一般在庭環境中準確率可達3至5%。

Nowadays, more and more people concern the energy and environmental issues and would like to manage the use of electric power from a small-scale area such as a house, a building, or community. One of the most important tasks is to forecast the power consumption in next several minutes and/or hours so that people may cooperatively use their appliances in an asynchronous manner to alleviate the peak power consumption of an area. Different from conventional large-scale power consumption forecast schemes which are mainly based on artificial intelligence methods such as artificial neural network, this paper proposes a new approach to predict the energy consumption of a house and building based on appliance state transition models which can be gathered from a nonintrusive load monitoring (NILM) meter. First, the appliance usage patterns of a house or a building are obtained from the NILM meter. Then, appliance state transition models can be established and they can be used to predict the energy consumption of a house or building efficiently. Simulation results indicate that only 3% to 5% prediction error is introduced for a typical house environment.