Building on the previous exploration into deciphering electricity consumption patterns within households, the next project focused on testing various NILM (non-intrusive load monitoring) algorithms for accuracy. Leveraging datasets from multiple households spanning approximately 3 months, the project aimed to evaluate neural network-based model architectures.
However, the journey was not without its challenges. Limited data due to short recording periods, sensor failures, and varying device types among households required creative solutions to achieve meaningful results. Experiment setups were devised to navigate these challenges:
While the quest for precision in predicting individual appliance signals proved challenging, the analysis of weekly and monthly aggregated consumption data yielded promising results. Notably, in some instances, experiments faced an obstacle where devices, such as the pool pump, were used during the training period (summer) but not during the three-month testing period in autumn, adding seasonal factors and parameters to the mix.
The insights gained during the project paved the way for optimizing similar experiments and mapping best practices for potential future business applications. An example was the idea of aligning the usage patterns of devices across households. A clustering approach was suggested, wherein customers are grouped, each cluster containing similar households, and are assigned their own distinct model.
This collaborative venture between Energie Steiermark and the Know Center not only enriches the understanding of household energy consumption but also underscores the potential of AI-driven solutions in refining energy management for informed energy usage and consumption and a sustainable future.