Abstract

The focus of this thesis is the development of a residential end-use energy estimation model which is based on the disaggregation of the monitored whole-house electric load. The data used to develop the model involves a one-time intrusive monitoring period to collect rapid-sampling interval data of demand for each major appliance load and the whole-house for a period of one week for one dwelling. The data applied to test the model consists of only the whole-house demand for a period of two weeks. The model identified is a rule-based algorithm applying pattern recognition techniques. The model is developed to scan the whole-house electric demand profile and to detect a predefined appliance energy signature. The results are presented in terms of the appliance daily demand profile and energy use. The benefits of this approach are that the frequency and time of usage of an appliance can be estimated without regard to the energy-related habits of the household occupants, their social or demographic characteristics, or the thermal characteristics of the dwelling. Moreover, the model is conceptually simple and versatile because of its rule-base platform. Artificial neural networks are investigated as a possible alternative to some of the rules developed as part of the Pattern Recognition Algorithm. A neural network model is developed as a preprocessor to the Pattern Recognition Algorithm with the aim to detect the ON and OFF occurrences of an appliance from the whole-house demand profile based on one week's worth of training data. Commercially available neural network models with different architectures and training parameters are applied in this study. Both approaches show a promising potential for application in residential buildings. Both models developed are characterized by low cost, modest data collection needs, and no occupant-related information required