傅立成臺灣大學：資訊工程學研究所陳立民Chen, Li-MingLi-MingChen2007-11-262018-07-052007-11-262018-07-052004http://ntur.lib.ntu.edu.tw//handle/246246/53663隨著科技的進步，電腦開始進入人類生活的每一個部分，嵌入式裝置、資訊家電、及家庭控制網路的出現讓人類生活中存在各種不同類型的電腦提供應用，人類跟電腦之間的互動不再侷限於坐在電腦前用螢幕、鍵盤、滑鼠，而是可以擴展到更多樣化的方式，也開啟了智慧型電子家庭系統的研究及應用，不論是資訊家電、健康照顧、生活協助等等，這些都是目前智慧型電子家庭系統中熱門的研究方向。 而智慧型電子家庭系統中最終的目的就是－讓使用者能感到更舒適。要達到這個目的，唯一方式的就是讓系統有能力去了解使用者的個人偏好行為，當系統了解了使用者的習慣後，自然就可以提供最適當的服務給使用者，讓使用者感到相當窩心，就像是使用者的好朋友一樣，總是知道使用者目前想要的服務。 因此，在這篇論文中，我們提出了一套『自動化個人偏好學習系統（Automatic Personal Preference Learning System）』，這個系統利用智慧型電子家庭系統中的感測器（Sensor）以及代理人程式（Agent）收集電子家庭中的各種訊息，包括使用者操作家電習慣以及當時的環境資料等。在經過長時間的資料收集後，『自動化個人偏好學習系統』會將這些收集的資料透過學習的方式去學習使用者在電子家庭中的偏好行為，並且利用學習後的結果建立起『個人偏好模型』。 最後，智慧型電子家庭系統可以透過這個模型去預測目前使用者所想要的服務有哪些，並且，智慧型電子家庭系統會以推薦的方式，將預測出來的結果推薦給使用者。In this thesis, the Automatic Personal Preference Learning System (APPLE system) is proposed. The APPLE system is able to learn personal preference (or inhabitant preference) and then feedback a proper service to the inhabitant. First of all, the complete database is needed before learning the personal preference. Thus we need the data collection mechanism to collect all these information. There are two kinds of data have close relationships with the personal preference; that is the environment data and the personal preference data, which we call “e-Home Data.” In addition, this thesis adopts Device Control Agent both to connect sensors with home electric appliances, and to collect environment data through different type of sensors. The other e-Home data is personal preference data, which records all electric appliances using condition operated by people. Personal preference data can support the APPLE system to find out personal preference precisely. This thesis proposes Interface Agent to collect personal preference data. However, the e-Home data collection mechanism sometimes receives dirty data during collection. Too much dirty data will seriously affect the accuracy of APPLE system. In order to avoid the situation, the APPLE system uses DCA (detection and correction algorithm) and RMA (removal and merging algorithm) to handle those dirty data. After collecting e-Home data, the APPLE system use Personal Preference Belief Network to predict the personal preference. The system is evaluated through experiments, and the accuracy of predicting personal preference is over 90% when the APPLE system has learned after a period of time.Chapter 1 Introduction 1 1.1 Motivation 1 1.2 Objectives 2 1.3 Related Works 3 1.4 Thesis Organization 5 Chapter 2 A Multi-Agent Architecture for Intelligent e-Home System 7 2.1 Overview 7 2.2 Agent Host 8 2.3 Mobile Agent 8 2.3.1 Database Agent 9 2.3.2 Device Control Agent 12 2.3.3 Personal Preference Agent 15 2.3.4 Interface Agent 19 Chapter 3 e-Home Data Definition and Collection 20 3.1 Overview 20 3.2 Definition of the e-Home Data 21 3.2.1 Definition of the Environment Data 21 3.2.2 Definition of the Personal Preference Data 24 3.3 Collection of the e-Home Data 27 3.3.1 Collection of the Environment Data 27 3.3.2 Collection of the Personal Preference Data 31 Chapter 4 e-Home Data Preprocessing 34 4.1 Overview 34 4.2 Missing Data Cleaning 35 4.2.1 What is “Missing Data”? 35 4.2.2 How to Clean “Missing Data”? 36 4.3 Noisy Data Cleaning 38 4.3.1 What is “Noisy Data” 38 4.3.2 How to Clean “Noisy Data” 39 4.4 Redundant Data Reduction 42 4.4.1 What is “Redundant Data”? 42 4.4.2 How to Reduce “Redundant Data”? 44 Chapter 5 Automatic Personal Preference Learning System 48 5.1 Overview 48 5.2 Personal Preference Belief Network 50 5.2.1 Overview of Personal Preference Belief Network 50 5.2.2 Bayesian Belief Updates 53 5.2.3 Probability of each environment data class 55 5.3 Service Recommendation 62 Chapter 6 Experiment and System Evaluation 64 6.1 Experiment Environment 64 6.2 System Implementation 66 6.3 System Evaluation 69 Chapter 7 Conclusion 723301333 bytesapplication/pdfen-US自動化個人偏好學習系統個人偏好學習蘋果系統家庭自動化環境資料收集系統電子家庭智慧型環境home automationintelligent environmentautomatic personal preference learning systemAPPLE systemintelligent e-Homee-Homepersonal preference learningenvironment data collection systemthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53663/1/ntu-93-R91922108-1.pdf