FPGA-based Architecture for Identification Algorithms in NILM Techniques

Abstract

During the last decades, smart grids have emerged as a general framework, where different systems and proposals have been developed for a wide range of applications. One of them is the so-called Non-Intrusive Load Monitoring (NILM), which tries to disaggregate the consumed energy on a household or building, based on the voltage, current and/or power measurements carried out in a single smart meter. NILM techniques often vary according to the variables measured and their corresponding sampling frequency, thus determining the final application. In recent years, these NILM techniques have become relevant in the field of Ambient Intelligence for Independent Living, where the identification of loads and the detection of the on/off switching in appliances can be used to infer the householders’ behavior in a non-intrusive way. This is the basis for further applications, such as diagnosis support tools, monitoring proposals for carers, or alert systems. In any case, after the energy disaggregation, the load identification is a key stage that is commonly implemented by different types of intelligent systems. This work defines a preliminary architecture for the real-time implementation of a classical neural network on a FPGA (Field-Programmable Gate Array) device. The proposal is focused on the load identification problem, as well as on the energy disaggregation previously mentioned. The architecture is thought to be integrated in a global system, also including the acquisition modules for voltages and currents in the mains, as well as the low-level processing of those variables and/or parameters involved to achieve the energy disaggregation.