A comparative study on pretreatment methods and dimensionality reduction techniques for energy data disaggregation in home appliances

Abstract

Energy meters provide valuable information that can be used to determine important features such as energy consumption of electrical devices and consumption habits in corporate, residential or public institutions. This information is crucial to establish energy saving strategies. With this aim, different approaches have been proposed in the literature, including non-intrusive load monitoring techniques, which enable the energy disaggregation of appliances and devices through a centralized measurement taken at panel level using a metering infrastructure. Generally, the accuracy of these techniques increases as more information is available on the analyzed signals or through subsequent post-computed values. Active power, reactive power, or even current harmonics measurements can be used for this task. However, the use of these and other recently proposed power and current features increases the dimensionality and, therefore, the complexity of the algorithms involved in the disaggregation process. Therefore, it is necessary to apply advanced techniques to reduce the dimensionality of the data, as well as the possible linear dependence between variables. This paper compares the performance of 8 data pretreatment methods and 6 dimensionality reduction techniques to data retrieved by an advanced metering infrastructure in a real environment consisting of 10 different home appliances. Results obtained from the comparative analysis show that the information provided by raw data can be enhanced by using pretreatment techniques and dimensionality reduction methods, especially when a custom combination of active power and current harmonics measures is considered.