| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系 === 104 === According to the research, if we can provide immediate and fine-grained power information to users, a significant reduction in the energy wastage can be achieved. Non-Intrusive Appliance Load Monitoring is an approach to reach the goal, which is more practical and feasible for typical families. In previous studies, we can discover that there were some disadvantages. First, it usually used high frequency sensor to acquire information, which made the cost of hardware higher. Second, most studies focused on the high consumption or on/off type appliances. As a result, low consumption appliances, multi-state appliances and continuously variable appliances were ignored.
In this paper, we proposed a low cost and real-time approach. We use two-step detection in training phase and cluster detection in testing phase to confirm an event. Besides, we use a clustering algorithm-ISODATA to find an appropriate number of state for each appliance in the training set after feature extraction. Finally, we succeed to build the ensemble learning model integrating with dynamic time warping (DTW) model to identify appliances.
Experimental results implies that two-step detection and cluster detection method can avoid excessive unknown appliance events, which can improve the accuracy of event detection. In addition, we can solve the problem of tie vote by using ensemble learning model integrating with DTW predictive model, which results in better recognition accuracy than using a single predictive model.
|
|---|
