基於非侵入式負載監測之空間人數估計研究

Abstract

現今環境由於用電成本提高，家庭與企業對於節能議題越來越重視，然而過去的節能管理系統，多半提供家庭與企業其歷史用電資訊作為相對比較與節能的改善基礎，但這樣的資訊讓使用者不知該從何下手，往往無法有效幫助使用者制定節能策略，因此若能建立以使用者人數為基礎之用電模型，則可以提供使用者更具意義的節能建議，並幫助使用者制定節能策略。 本論文研究電器狀態與空間人數間的關係，利用非侵入式負載監測可辨識出電器的型態、狀態與耗電資訊，再以歷史人數資料搭配監督式與半監督式學習，建構人數預測模型。監督式學習方面採取類神經網路與支持狀態機建立人數預測模型，透過實際收集的電器與人數資訊做為驗證，再利用時間關聯性與皮爾森相關係數提升監督式學習的預測能力。另一部分的半監督式學習則使用支持狀態機作為基礎，將相同型態空間的電器資訊利用轉換與對應的方式，再利用主成分分析找出共同轉換最大數，產生出可用於相同型態空間的人數預測模型。基於上述研究，本論文提出依照空間人數之用電效率評估與節電建議。

The energy-saving issue is getting more and more important recently due to the increase of electricity cost. Conventional energy management system (EMS) provides the historical information of energy consumption of a building, but the information cannot indicate inefficiency of energy usage which is strongly related to the number of people in a space. In this paper, we investigate the relationship between appliances, appliance states and the number of people in a space. To gather the appliance type, state and power consumption, we rely on nonintrusive load monitoring (NILM) meter. Thus, we can develop energy consumption models based on the number of people in a space instead of showing only energy consumption of a space. To develop people counting schemes, we propose both supervised and semi- supervised training. For supervised training, we develop the model based on artificial neural network (ANN) and support vector machine (SVM). To improve the prediction accuracy, we further enhance the model with time relation and Pearson correlation coefficient. For semi-supervised training, we develop the model based on the combination of SVM and principal component analysis. Based on the proposed technologies and models, we are able to alert users the energy inefficiency according to the number of people in a space.