The massive deployment of smart meters in most Western countries in recent decades has allowed the creation and development of a significant variety of applications, mainly related to efficient energy management. The information provided about energy consumption has also been dedicated to the areas of social work and health. In this context, smart meters are considered single-point non-intrusive sensors that might be used to monitor the behaviour and activity patterns of people living in a household. This work describes the design of a short-term behavioural alarm generator based on the processing of energy consumption data coming from a commercial smart meter. The device captured data from a household for a period of six months, thus providing the consumption disaggregated per appliance at an interval of one hour. These data were used to train different intelligent systems, capable of estimating the predicted consumption for the next one-hour interval. Four different approaches have been considered and compared when designing the prediction system: a recurrent neural network, a convolutional neural network, a random forest, and a decision tree. By statistically analysing these predictions and the actual final energy consumption measurements, anomalies can be detected in the undertaking of three different daily activities: sleeping, breakfast, and lunch. The recurrent neural network achieves an F1-score of 0.8 in the detection of these anomalies for the household under analysis, outperforming other approaches. The proposal might be applied to the generation of a short-term alarm, which can be involved in future deployments and developments in the field of ambient assisted living.

Non-Intrusive Load Monitoring (NILM) techniques are commonly used to measure and identify the power consumption of different types of household appliances, starting from an aggregated signal obtained from a single measurement point. Currently, they are often based on the extended deployment of smart meters (SM) carried out in most developed countries in the last decade. Based on the measurements acquired by SMs, it is possible to disaggregate energy consumption, and then to identity the corresponding loads plugged to the mains of a house or building. NILM techniques can be applied in different application fields, such as energy efficiency, active demand response management, or even as a way to infer the behaviour patterns of the people living in a certain household under monitoring, in the context of Ambient Intelligence for Independent Living (AIIL). This paper presents a new approach for energy disaggregation through the use of Recurrent Neural Networks (RNN) based on measurements from a single point at low sampling rates. The proposed framework takes the power signals acquired by an SM as inputs, then pre-processes and detects the on/off switching events of the different appliances considered, and finally classifies them using two different Long Short Term Memory (LSTM)-based topologies. The proposal validation is carried out through the use of the well-known public dataset Building Level fUlly labelled for Electricity Disaggregation (BLUED). Several configurations of classification topologies have been compared, obtaining an average classification accuracy in the experimental results that exceeds 85%.

Non-Intrusive Load Monitoring (NILM) includes a set of methods orientated to disaggregating the power consumption of a household per appliance. It is commonly based on a single metering point, typically a smart meter at the entry of the electrical grid of the building, where signals of interest, such as voltage or current, can be measured and analyzed in order to disaggregate and identify which appliance is turned on/off at any time. Although this information is key for further applications linked to energy efficiency and management, it may also be applied to social and health contexts. Since the activation of the appliances in a household is related to certain daily activities carried out by the corresponding tenants, NILM techniques are also interesting in the design of remote monitoring systems that can enhance the development of novel feasible healthcare models. Therefore, these techniques may foster the independent living of elderly and/or cognitively impaired people in their own homes, while relatives and caregivers may have access to additional information about a person’s routines. In this context, this work describes an intelligent solution based on deep neural networks, which is able to identify the daily activities carried out in a household, starting from the disaggregated consumption per appliance provided by a commercial smart meter. With the daily activities identified, the usage patterns of the appliances and the corresponding behaviour can be monitored in the long term after a training period. In this way, every new day may be assessed statistically, thus providing a score about how similar this day is to the routines learned during the training interval. The proposal has been experimentally validated by means of two commercially available smart monitors installed in real houses where tenants followed their daily routines, as well as by using the well-known database UK-DALE.

Most Non-Intrusive Load Monitoring (NILM) techniques often require a correct detection of the events that occur in the mains, in order to carry out a correct identification of the corresponding appliances. For that performance, event-detectors are normally based on signals, such as voltages and currents, acquired by a smart meter at the entrance of the household. In this work, a finite precision analysis is performed for an event detector implemented on a System-on-Chip (SoC) based on a Field-Programmable Gate Array (FPGA) for NILM applications. The proposal employs an integrated circuit (ADE9153A) to perform the signal acquisition at 4 ksamples/s, higher than those used in common smart meters installed at homes nowadays. The purpose of the finite precision analysis is to reduce the hardware resource consumption of the FPGA, while the precision of the event detector is maintained. To validate the design, a comparison is carried out in terms of accuracy and specificity against previous works, as well as a comparison of the event detector using finite precision (fixed-point) and single-precision floating-point resolution. Finally, the hardware resource consumption obtained for the proposed architecture is discussed