基于Triplet Loss和KNN的非侵入式未知负荷识别OA北大核心CSTPCD

In allusion to the problems of misclassification of traditional non-intrusive load identification algorithm when new load is connected,a non-intrusive unknown load identification algorithm based on Triplet Loss and KNN(K-nearest neighbor)is proposed.The multi-feature fusion color V-I trajectory image is constructed by means of the current and voltage of the load during steady state operation.The online Semi-Hard sample pairs are mined,and the Triplet Loss is used to training neural network,so as to obtain feature vectors for each sample.The PCA(principal component analysis)dimensionality reduction for feature vectors is conducted,the neighbourhood is constructed based on the class center,and the KNN algorithm is used for the load identification.The algorithm is tested by means of PLAID and COOLL datasets.The testing results show that the proposed load identification algorithm has high accuracy in both the classification of known loads and the recognition of unknown loads.