近年來,由於能源過度使用造成的環境與經濟問題,讓節能成為產業界最為關切的焦點技術議題之一。節能可以從每人都可做到的家庭用電做起,大多的家庭節能研究在家庭自動化的層次做考量,雖然有少數使用活動辨識技術來輔助節能系統,但他們仍忽略了如何為每個家庭提供更貼切的節能服務,以及系統是否滿足使用者對舒適度的要求,因此考量的層次仍不夠實際。 智慧家庭中透過各類感測器收集環境資訊,並利用全監督式學習法的活動辨識技術效果已逐漸成熟。由於節能系統的效果與其能提供服務的時間有絕對的關係,但現有的以活動辨識為基礎的節能系統通常依照活動辨識的結果來決定系統的後續行為,造成系統在準確率不高時無法提供使用者服務。因此此類節能系統的實用性,往往和系統是否提供額外的學習機制來輔助辨識模型有很大關係,而現有的活動辨識節能系統往往忽略此項議題。此外,現有節能系統的服務,往往都是提供通用的節能服務,沒有進一步為每個家庭分析可以節省的內容,可能導致節能效果不佳或造成過度調控。 本研究的主要貢獻有以下三點: 第一,節能系統的學習階段中,如何從現有行為模型找出相似的群體模型,並透過此類模型使系統提供服務的時間增加,進一步增加節能效果。第二,當進行特定活動時,每個家庭都會有不同的電器偏好,因此系統提供的不只是通用的節能服務,我們會根據前述的群體模型及電器偏好來找出相對應可調節的電器,替每個家庭提供更全面的節能服務,並且為此種偏好建立一個衡量的舒適指標。第三,提供節能服務時,我們會同時考量到使用者的整體舒適度來進行調控,希望替使用者找到節能且滿足使用者舒適度的節能服務。
In recent years, energy saving has become an important issue due to environmental and economic problems, so energy saving for household is a challenging and important issue for smart homes. Most of the prior works focus on home automation to achieve energy saving. Even few works focus on using context-aware technology to assist it; they often ignore those appliances which are operating indirectly or implicitly related to the context. On the other hand, the performance of context-aware based energy saving system is highly dependent on the accuracy of activity recognition rate; especially in multi-user environment, the accuracy is often lower than that for the single-user envi-ronment. The main contribution of this thesis is three-fold. Firstly, we will cluster similar single “activity contexts” into a “group activity context” in the learning phase, which can help us to make multi-user activity recognition far more tractable. Secondly, we propose a comprehensive comfort index to evaluate how the user feels under certain en-vironment conditions, including thermal, illumination and appliance-usage preference, which make the system to provide human-centric energy saving service to user. Thirdly, we formulate the energy saving into an optimization problem, which tries to minimize the total energy consumption while maintain the user comfort constraints.