Summary: | 碩士 === 國立臺北科技大學 === 自動化科技研究所 === 105 === The most challenge part of the NILM system is to identify the loads with fewer numbers of features and sampled values while maintaining the high accuracy. The load event detection is used to find the features that can be used to identify the loads. kNN and Gaussian mixture model with K-Means and EM are used in this paper as classifiers.
The experimental results show that the general household appliance has high recognition percentage. The 100% recognition rate can be achieved by using the single feature. However, when different types of loads are used, due to the feature confusion, these loads cannot be identified correctly. As a result, new features are needed to improve the recognition rate. In the process of multi-feature identification, it is found that different results are obtained when different samples are used. To alleviate the problem, an adaptive algorithm is proposed to improve the recognition rate. The results show that the proposed approach is capable of identify different types of loads even some of the features are similar.
|