Sistemi di Machine Learning innovativi per il monitoraggio del carico elettrico

Non-Intrusive Load Monitoring (NILM) is the process that allows obtaining information about the electrical loads powered by an electrical system through a single measurement performed in a single point of the system itself. Systems based on this process provide an alternative solution to the more traditional intrusive one. NILM requires a reduced number of equipment and less occupied space, even if it presents a greater complexity in terms of processing the acquired data. In fact, this solution is much simpler, from the hardware point of view, as it requires the measurement of a voltage and a current, or often even just the current. However, the complexity shifts to the processing section, which must identify the absorption of the individual devices through the use of appropriate algorithms. The information required from an electrical loads monitoring system may concern their status (ON/OFF) or the electrical quantities involved in their operation. This information must be made available in a more or less short time depending on the application in which the measuring system is used. The most common application is to monitor the electricity consumption of different devices within a residential home. In this case the information must be updated on time intervals of the order of days or weeks. Today, new NILM systems are used in numerous innovative applications in residential environments. For example, some human activity recognition (HAR) and ambient assisted living (AAL) systems are based on disaggregated appliance activity data. Innovative commercial and industrial applications are based on the NILM technique, such as to implement predictive maintenance. Energy disaggregation is also applied to manage the generation and storage of energy in smart grids. Therefore, the times in which it is necessary to have information about the state or the electrical quantities of a load are drastically reduced, down to a few seconds. In the first part of this thesis the current state-of-the-art of NILM systems will be defined, paying particular attention to the most significant contributions. Subsequently, the applications of these systems in industrial and residential contexts will be described in detail. In the second part, three different systems will be proposed having different characteristics both from the point of view of the electrical quantities measured, the sampling frequency and the signal processing section. More specifically, the experimental systems created, based on a microcontroller, use Machine Learning algorithms to process the signals obtained from the measurement section. For each of the proposed systems, a wide range of measurements on test systems were carried out, in order to effectively evaluate their performance in real conditions.