Non-invasive power monitoring of microgrid based on two-layer clustering and multi-task learning

A new non-invasive power monitoring method based on two-layer clustering and multi-task learning is proposed to solve the problems of power monitoring at microgrid level. Firstly, the historical operating data of microgrid users are pre-clustered based on density clustering algorithm, and then the membership matrix is obtained by fuzzy clustering. Then, a power identification model is constructed based on clustering results, and the shared layer of the model is used to extract the influence of external factors such as climate on the microgrid. Finally, the non-invasive power decomposition and monitoring of the microgrid are completed based on identification model and independent component analysis. The feasibility of the proposed method is verified by a case study of an actual integrated energy microgrid.