Lin, Huan (2016)
Softwarekonzept für die automatisierte Disaggregation von haushaltsbezogenen Stromverbrauchsdaten.
Technische Universität Darmstadt
Masterarbeit, Bibliographie

Kurzbeschreibung (Abstract)

In the context of energy conservation and environment protection, a more efficient, more environment friendly, and safer energy supply management system has become an indispensable approach to support and facilitate energy efficiency improvement and better prepare the energy infrastructure and public policy for energy transformation. However, on the demand side of the energy system, instead of looking for solutions for more sufficient energy supply, exploiting potential for energy conservation and economical and ecological energy usage pattern has become focus of the attention. In the residential sector, with the development of smart household appliances and smart meter system, the wealth of data delivered by such interfaces and components is growing exponentially, which has provided a vast space for application of different data processing and machine learning algorithms and model. The Non-intrusive load monitoring system (NILM) depends on aggregate power readings from each household to explore and learning different usage pattern of consumers, which might serve to discover energy saving potential and supporting public policy making. A more specific implementation is power disaggregation, which provides power usage information on devise level based on aggregate power consumption data. In this study, different approaches to power disaggregation are overviewed and analyzed. especially unsupervised learning algorithms including Hidden Markov Model and its extensions. Based on comprehensive theoretical validation, a factorial framework of Hidden Semi-Markov Model is established and implemented. In order to evaluate the efficiency and performance of the factorial model, data from the public data set REDD is utilized to perform experimentation on the model. Before the experimentation several data preparation operation including data set diagnostic, data preprocessing should be carried out. Experimentations are designed to expose the model to different application scenarios with different complexity, so as to evaluate the robustness and applicability of the model. The result of the trials shows that, given information on major appliance type of the monitored households, the factorial model can accurately reproduce the power use behavior of each major target appliance in the household based on the input aggregate data. Although the robustness, adaptability and scalability of the model have been validated, there are still improvement potentials for further studies, such as integrating the factorial model with some sort of appliance type detection model or algorithms to enhance applicability of the model.

Frage zum Eintrag

Redaktionelle Details anzeigen