LIBRARIES
    • Login
    Research Exchange
    Share your work
    View Item 
    •   Research Exchange
    • Electronic Dissertations and Theses
    • Electronic Dissertations
    • View Item
    •   Research Exchange
    • Electronic Dissertations and Theses
    • Electronic Dissertations
    • View Item
    JavaScript is disabled for your browser. Some features of this site may not work without it.

    Browse

    All of Research ExchangeCommunities & CollectionsBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

    My Account

    LoginRegister

    The Science of Home Automation

    Thumbnail
    View/Open
    Thomas_wsu_0251E_12106.pdf (18.69Mb)
    Date
    2017
    Author
    Thomas, Brian Louis
    Metadata
    Show full item record
    Abstract
    Smart home technologies and the concept of home automation have become more popular in recent years. This popularity has been accompanied by social acceptance of passive sensors installed throughout the home. The subsequent increase in smart homes facilitates the creation of home automation strategies. We believe that home automation strategies can be generated intelligently by utilizing smart home sensors and activity learning. In this dissertation, we hypothesize that home automation can benefit from activity awareness. To test this, we develop our activity-aware smart automation system, CARL (CASAS Activity-aware Resource Learning). CARL learns the associations between activities and device usage from historical data and utilizes the activity-aware capabilities to control the devices. To help validate CARL we deploy and test three different versions of the automation system in a real-world smart environment. To provide a foundation of activity learning, we integrate existing activity recognition and activity forecasting into CARL home automation. We also explore two alternatives to using human-labeled data to train the activity learning models. The first unsupervised method is Activity Detection, and the second is a modified DBSCAN algorithm that utilizes Dynamic Time Warping (DTW) as a distance metric. We compare the performance of activity learning with human-defined labels and with automatically-discovered activity categories. To provide evidence in support of our hypothesis, we evaluate CARL automation in a smart home testbed. Our results indicate that home automation can be boosted through activity awareness. We also find that the resulting automation has a high degree of usability and comfort for the smart home resident.
    URI
    http://hdl.handle.net/2376/13002
    Collections
    • Electronic Dissertations and Theses - Electrical Engineering and Computer Science
    • Electronic Dissertations