On the effect of the meter uncertainty on the Multi Object Optimization in NILM applications: a preliminary analysis

In literature the topic of energy disaggregation is known as NILM (Non–Intrusive Load Monitoring). [1]. In some areas, specifically in the industrial ones, monitoring the electrical consumption of each individual device is necessary to provide indications on the energy optimization of the entire system, providing indications on the areas in which to intervene primarily, containing the costs associated with the installation of sensors, thanks to the process of non-intrusive monitoring. In fact, using a single measuring device placed upstream of the system, allows to obtain the consumption of individual devices through mathematical algorithms applied to the overall measurements. The advantages are various, among which is the reduction of the number of measurement devices, which otherwise would have been necessary to install on individual devices, consequently reducing the measurement error related to the sensors. There are different approaches to learning such algorithms, in particular they are divided into unsupervised and supervised. In unsupervised approaches, the devices must be identified and extracted from the aggregate power signal and their models added to the device database. Thus, quality of load disaggregation is dependent on the ability of the system to correctly identify the devices. When adopting optimization algorithms, the device characteristics are compared with a reference database. When the deviation between the ground truth and the predicted state is minimized, the solution is derived. However, the performance of these algorithms, decreases when the number of loads increases and when similar appliances are involved.