Due to an exponential rise in energy consumption, it is imperative that buildings adopt sustainable energy consumption systems. A number of studies have shown that this can be achieved by providing real-time feedback on the energy consumption of each appliance to residents. It is possible to accomplish this through non-intrusive load monitoring (NILM) that disaggregates electricity consumption of individual appliances from the total energy consumption of a household. Research on NILM typically trains the inference model for a single house which cannot be generalized to other houses. In this Master thesis, a novel approach is proposed to tackle mentioned issue.This thesis proposes to use two finite mixture models namely generalized Gaussian mixture and Gamma mixture, to create a generalizable electrical signature model for each appliance type by training over labelled data and create various combinations of appliances together. By using this strategy, a model can be used on unseen houses, without extensive training on the new house. 
The issue of different measurement resolutions in the NILM area is also a considerable challenge. As a rule of thumb, state-of-the-art methods are studied using high-frequency data, which is rarely applicable in real-world situations due to smart meters' limited precision. To address this issue, the model is evaluated on three different datasets with different timestamps, AMPds, REDD and IRISE datasets.  To increase the aggregation level and compare with RNN and FHMM as two well-known methods in NILM, an extension that we called DNN-Mixtures, is proposed. The results show that the proposed model can compete with state of art techniques. For evaluation, accuracy, precision, recall and F-score metrics are used.