Search

In order to solve the problem that a single V-I track feature can not effectively identify similar track features and the extracted features are prone to redundacy or even noise features in non-invasive load identification,a hierarchical non-invasive load identification method based on K-means clustering and PSO feature optimization was proposed.Firstly,K-means algorithm was used to initially classify the HOG features of load V-I trajectories,and the appliances with similar trajectories were classified into one category.Then,multi-dimensional features are extracted from electrical current data in each category and the optimal feature subset is selected by PSO algorithm.Finally,KNN model was used for secondary load identification.The experimental results show that this method effectively improves the accuracy of load identification.Extracting the HOG feature of V-I trajectory solves the problem of fluctuation of the same electrical appliance.PSO feature optimized KNN secondary classification is carried out for each category after the first level classification,which solves the problem of poor adaptability of some electrical appliances to feature subset.The proposed method solves the influence of redundant features and even noise features on the identification accuracy to a certain extent,and provides a new idea for the selection of load features,which has important reference significance for the practical application of load identification.

Aiming at the problems of the current non-intrusive load identification,such as too long model training time and low identification accuracy of electrical appliances with similar load characteristics,a non-intrusive load identification method based on CF-MF-SE joint feature was proposed.Based on the steady-state current signal,the peak factor was extracted to represent the distortion degree of the waveform,the margin factor was extracted to represent the stability degree of the signal,the spectral entropy was extracted to represent the complexity degree of the spectrum structure,and PSO-SVM was combined to realize load identification.Experimental results show that this method can solve the problem that the electrical current waveform is too similar to identify successfully,reduce the training time,and improve the recognition accuracy and efficiency.This method introduces the vibration signal characteristics as load characteristics into the field of load identification,which provides a new idea for feature selection of non-invasive load identification technology.As a key feature sensitive to load,spectral entropy can significantly improve the identification rate when combined with other features,which provides reference for the flexible selection of load characteristics in practical application.[HQ]