A non-intrusive load identification method considering unknown loads

The current non-intrusive recognition model has complex input features and model structure. In addition, it usually can only identify known loads in training dataset and cannot identify unknown loads that are outside the training dataset, which makes it difficult to be applied to dynamic user-side monitoring environment. To this end, we propose a non-intrusive load identification method considering unknown loads—One Class Load Identification Network(OC-LIN), which can accurately identify both known and unknown loads. The proposed method consists of two main parts: (1) a load identification network that accurately identify loads using fusion V-I trajectory feature and odd harmonic amplitude feature; (2) an one class method in deep support vector data description (DSVDD) to identify unknown loads. Through the validation of the PLAID 2018 dataset, the proposed method not only identifies the unknown loads efficiently, but also improves the identification precision among the known loads.