建築物耗電預估與評比工具之設計與實作

Abstract

建築物的耗電模型與預測近年來因為電費的高漲而逐漸浮上檯面，然而人們卻無法從每月的電費單中獲得更詳細的住家用電資訊。建築物耗能可能來自非常多地方如電器使用行為、房屋大小到建築物結構、座向與當地天氣等等，多樣的參數總和出整體耗電行為。如此多樣的建築物參數讓建立建築物耗電模型，無論在最初的收集參數到建築物耗電預測都顯得複雜許多。本篇論文針對建築物參數收集與建立模型提出一套建築物智慧專家系統，在收集資料方面我們利用人人都有的手機，利用上面的感應器如相機、Wi-Fi定位等收集資料，並與開放的網路資源結合如天氣資料與電器使用統計，讓一台手機能夠收集到絕大多數所需的建築物參數，降低了參數收集的難度；在耗電模型上，我們結合了理論模型(熱平衡理論)與統計模型，將理論模型簡化後的建築物參數集結合統計模型資料訓練的特性，達到簡化參數以及針對環境因地制宜的目的，期望利用此模型預測的結果當作真實建築物耗電的參考比較依據；我們也利用資料參數的靈敏度分析，在使用者收集建築物參數時給予建議，讓使用者在收集的同時，也能了解各種參數的重要程度。最後實驗得出我們的手機工具簡化了參數收集的難度且量測誤差約8%尚可接受；透過校準EnergyPlus，建築物模型能提供約80%準確的預測能力；參數靈敏度不僅受環境參數值影響，也與歷史資料分布有關，使用者可透過此建議了解參數的重要程度與其變異的走向。透過我們提供的建築物專家系統，希望使用者能夠較容易地了解他們的住家，進而發現潛藏的不正常耗電原因或是尚可節電之處。

Building energy consumption becomes important in the age of high electricity bills. However, people cannot understand detail information about their energy consumption by reading their monthly bill. Energy consumption of a building depends on many factors such as building structure, human activities and weather. It is hard to understand the relation between these building factors and energy consumption, and also hard to gather these data. In this paper, we purpose a building expert system that can help user realize their houses in an easier and systematic way. First, we apply sensors on the smart phone and open data API to help the building factor collection so that we can finish most collection works by just one phone rather than many tools. Second, we combine reduced building factor set from EnergyPlus, which is an engineering building modeling tool, and support vector regression as our building energy consumption model method. The reduced factor set can achieve about 70% accuracy of the prediction of energy consumption [2] . Support vector regression is a statistical model that has a data-training step to make the model more fitting the dataset. We use the integrated model to predict an energy consumption and benchmark the real energy consumption. Third, we find that in different environment, the importance of the building factor will be different. Therefore, we apply sensitivity analysis on the historical data in order to suggest different building factor when user in the factor collection step. This helps user focus on the critical building factor and help accurate the prediction of energy consumption. Finally, we collect all users’ data including the values of all building factors and real energy consumption. We can have more information and based dataset for next user. The experiment shows app tool can achieve about 90% measurement accuracy and building model can achieve up to about 80% prediction accuracy.