On the applicability of correlation filters for appliance detection in smart meter readings

Communication systems utilize correlation filters to detect waveforms. In a broader sense, these filters examine the amount of resemblance between a template pattern and the input pattern. In the domain of smart grids, many applications require the detection of active electrical appliances, their condition as well as their current state of operation. Furthermore, the identification of power eaters, the recognition of ageing effects, and the forecast of required maintenance represent important challenges in (home) energy management systems. In this paper, we examine the applicability of correlation filters as a possible solution to meet such challenges. First, we introduce the concept of predictability to power consumption patterns of electrical appliances. Second, we present our concept and the implementation of correlation filters for this kind of application. The correlation filters utilize a particular consumption pattern of an electrical appliance to detect the respective appliance in energy readings from smart meters and smart plugs. Lastly, we assess the performance of the correlation filters on the real-world energy consumption dataset GREEND, which provides readings from smart meter data as well as appliance-level measurement equipment. As the results approve, the correlation filters show a good performance for appliances with predictable consumption patterns such as refrigerators, dishwashers, or washing machines. Thus, we propose that future work should evaluate the applicability of correlation filters in appliance diagnosis systems.