An unsupervised approach in learning load patterns for non-intrusive load monitoring

This paper proposes a new novel way for non-intrusive load monitoring. The technique can be applied to develop a powerful framework for low-cost power monitoring in buildings, particularly in the small commercial and residential sector. The method proposes the construction of a data base of prior knowledge about load patterns and it provides a powerful platform which has the capacity to solve one of the major challenges in power monitoring and energy management, which has been the development of robust unsupervised learning algorithms that eliminate the need for costly human involvement. To do so, a proposal is made on the basis of forming Bayesian networks for the load classification problem. The method has shown to be computationally compatible with handling a large data set. Finally, a case is studied for some major loads obtained from a bank building to demonstrate a basic test case in the real world.