High-performance power converter design for a domestic power flow management using NILM smart sensors

Conventional energy meters provide very limited information about the electricity consumption. In order to generate more valuable data about energy consumption in electric networks, NonIntrusive Load Monitoring (NILM) is studied. This approach is based on the disaggregation of the overall electric consumption signal from the mains of an electric network, reducing material and installation costs in front of sub-metering approaches. The performance of NILM has been arising in recent years due to machine learning techniques and Artificial Intelligence (AI) applications. Another key factor on the development of NILM is the cost-effectiveness of measuring devices and computational power, which has been improving year by year. The advance of information and communication technologies, along with microcontroller units (MCU) with the wide introduction of IoT, leads to more interconnected appliances both in household and industrial environment. The aim of this work is to create a supervised NILM system capable of classify four different appliances in each of its combinations of operation based on its steady state current signal. The system core is composed by a STM32F4 MCU which operates a previously trained k-NN classifier algorithm. After validating the NILM system, from the data extracted and a maximum current level allowed in the mains, a current set point is extracted to feed an AC/DC converter current control, which could be considered as an EV charger. Demonstrating the viability of relating both technologies through an Ethernet communication such Modbus TCP/IP. The results show that NILM machine learning algorithms can be performed by a cost-effective MCU with an acceptable accuracy. Also, it is demonstrated, in a simple way, the viability of NILM and power converters direct communication