- Title
- Intelligent recognition of electrical household appliances based on machine learning
- Creator
- Jiang, Lei
- Relation
- University of Newcastle Research Higher Degree Thesis
- Resource Type
- thesis
- Date
- 2016
- Description
- Research Doctorate - Doctor of Philosophy (PhD)
- Description
- The development of smart grids, especially the vast deployment of smart meters, enables users to access their energy data on a much more detailed scale. This has promoted considerable interest in nonintrusive load monitoring (NILM) research. It is hoped that NILM techniques can separate energy-consumption data into individual appliance levels, thereby aiding customers who want to save energy or cut back usage. For this dissertation, we implemented a novel NILM system based on a support vector machine (SVM), an edge symbol analysis method (ESA), and principal power component analysis (PPCA). Our research uses a set of power load disaggregation methods, which constitute the most important part of our NILM system. Furthermore, our system automatically monitors a household power consumption and the power consumption of individual devices. We have developed the means to automatically detect a power-load event and the means to classify appliances based on machine learning. Our methods involve a new transient-detection algorithm. By analysing turn-on and turn-off transients, it can accurately detect the points when a device is switched on or switched off. The proposed load classification technique can identify different power appliances with better recognition accuracy and computational speed. The load classification method comprises two items: frequency feature analysis and a support vector machine. Our results indicate that by incorporating the new edge detection and turn-on, turn-off transient signature analyses into our NILM, more information is revealed than by traditional NILM methods. The load classification method achieved more than 90 percent recognition rate. We also propose a multiple class support vector machine to recognize different appliances. The approach consists of two stages. In stage one, feature analysis is applied to power signals. In stage two, a trained classifier based on SVM was applied to identify different appliances. Our experiment results on real data give better performance compared to other studies that have used supervised classification for household power-appliance monitoring. Our contributions also focus on feature extraction and pattern analysis for whole-house power load classification and disaggregation. Specifically, after finding a specific power event (such as a switch-on or switch-off) and clustering data in terms of power factor, the optimized features can be obtained from our algorithm by looking at harmonics of active/reactive power and using eigenvalue feature analysis. The experiments, based on real world data, give higher recognition accuracy and faster computational speed, and represent a promising approach for distinguishing different loads effectively.
- Subject
- NILM; SVM; PCA; power decomposition
- Identifier
- http://hdl.handle.net/1959.13/1322134
- Identifier
- uon:24523
- Rights
- Copyright 2016 Lei Jiang
- Language
- eng
- Full Text
- Hits: 667
- Visitors: 966
- Downloads: 377
Thumbnail | File | Description | Size | Format | |||
---|---|---|---|---|---|---|---|
View Details Download | ATTACHMENT01 | Thesis | 5 MB | Adobe Acrobat PDF | View Details Download | ||
View Details Download | ATTACHMENT02 | Abstract | 723 KB | Adobe Acrobat PDF | View Details Download |