General NILM Methodology for Algorithm Parametrization, Optimization and Performance Evaluation

The research area of NILM exhibits a high heterogeneity, regarding approaches and characteristics. Especially, in terms of the applied algorithms, measurement data, quantities and features used, as well as congruent appliance event and state definitions. Therefore, performance evaluation and the establishment of comparability is not straightforward. The aim of the presented work was to address these challenges, through the development of an application-oriented, general methodology for parametrization, optimization and performance evaluation of NILM algorithms. The methodology is based on the general NILM framework and applicable to a wide range of NILM approaches and measurement data. Temporary, individual appliance measurements were utilized to build an extended appliance database and for providing a reliable ground truth for common performance evaluation metrics. Therefore, also a congruent event and state definition was formulated. For the application of the methodology, the focus was set on event-based NILM algorithms and measurement data of a commercial building and for one significant appliance, regarding the building´s overall energy demand. The methodology proved to be suitable for the aimed purpose. Two different event detection algorithms could be optimized, regarding their input parameters to be able to identify the appliance operation behavior optimally.