Title Page
Contents
ABSTRACT 6
Chapter 1. Introduction 8
Chapter 2. Previous Works 10
Chapter 3. Proposed Framework of Load Identification 13
3.1. Data set used for the experimental analysis 15
3.2. Features used for analysis 19
3.3. Algorithms used for load classification 24
3.4. Multi-feature/Multi-algorithm Method 26
Chapter 4. Experimental Results & Evaluation 28
4.1. Experimental settings 28
4.2. Results with K-NN classifier 30
4.3. Results with NB classifier 31
4.4. Results with proposed framework of load signature analysis 32
4.5. Analysis proposed framework result of load signature 35
Chapter 5. Conclusion 38
References 39
Table 1. Comparison of previous works 12
Table 2. Appliances monitored at each home 16
Table 3. Data used in this paper 17
Table 4. Number of Segments for Classification 17
Table 5. Number of Segments for Training 17
Table 6. Performance of k-Nearest Neighbor (k=1,2,3,4,5) algorithms 30
Table 7. Performance of Naive Bayesian algorithm 31
Table 8. Multi-feature multi-algorithm method (a) 32
Table 9. Multi-feature multi-algorithm method (b) 33
Table 10. Multi-feature multi-algorithm method (c) 34
Table 11. Multi-feature multi-algorithm result 35
Table 12. Confusion matrix of multi-feature multi-algorithm decision method 35
Table 13. TV's wrong classifier results 36
Table 14. The proportion of wrong classifier results with TV 36
Figure 1. Complex power space and appliance clusters. 10
Figure 2. The proposed load identification framework 14
Figure 3. Relationship between test data set and training data set 18
Figure 4. Four features used 19
Figure 5. An example of feature DC 20
Figure 6. An example of feature SO 21
Figure 7. Examples of feature VO 22
Figure 8. An example of feature ZC 23
Figure 9. Flow of Multi-Feature/Multi-Algorithm applied for a test data set 27
Figure 10. A current waveform collected from PC 28