Roberto Bonfigli

Unsupervised algorithms for non-intrusive load monitoring: An up-to-date overview

Research on Smart Grids has recently focused on the energy monitoring issue, with the objective to maximize the user consumption awareness in building contexts on one hand, and to provide a detailed description of customer habits to the utilities on the other. One of the hottest topic in this field is represented by Non-Intrusive Load Monitoring (NILM): it refers to those techniques aimed at decomposing the consumption aggregated data acquired at a single point of measurement into the diverse consumption profiles of appliances operating in the electrical system under study. The focus here is on unsupervised algorithms, which are the most interesting and of practical use in real case scenarios. Indeed, these methods rely on a sustainable amount of a-priori knowledge related to the applicative context of interest, thus minimizing the user intervention to operate, and are targeted to extract all information to operate directly from the measured aggregate data. This paper reports and describes the most promising unsupervised NILM methods recently proposed in the literature, by dividing them into two main categories: load classification and source separation approaches. An overview of the public available dataset used on purpose and a comparative analysis of the algorithms performance is provided, together with a discussion of challenges and future research directions.

An integrated system for voice command recognition and emergency detection based on audio signals

Emergencies and home automation commands are recognised by means of acoustic signals.Distress calls are recognised to allow tele-assistance by a relative or a care giver.A novelty detection algorithm detects abnormal acoustic events.The entire system has been implemented in a low-consuming embedded platform. The recent reports on population ageing in the most advanced countries are driving governments and the scientific community to focus on technologies for providing assistance to people in their own homes. Particular attention has been devoted to solutions based on acoustic signals since they provide a convenient way to monitor people activities and they enable hands-free human-machine interfaces. In this context, this paper presents a complete solution for voice command recognition and emergency detection based on audio signals entirely integrated in a low-consuming embedded platform. The system combines an active operation mode were distress calls are captured and a vocal interface is enabled for controlling the home automation subsystem, and a pro-active mode, were a novelty detection algorithm detects abnormal acoustic events to alert the user of a possible emergency. In the first operation mode, a Voice Activity Detector captures voice segments of the audio signal, and a speech recogniser detects commands and distress calls. In the pro-active mode, an acoustic novelty detector is employed in order to be able to deal with unknown sounds, thus not requiring an explicit modelling of emergency sounds. In addition, the system integrates a VoIP infrastructure so that emergencies can be communicated to relatives or care centres. The monitoring unit is equipped with multiple microphones and it is connected to the home local area network to communicate with the home automation subsystem. The algorithms have been implemented in a low-consuming embedded platform based on a ARM Cortex-A8 CPU. The effectiveness of the adopted algorithms has been tested on two different databases: ITAAL and A3Novelty. The obtained results show that the adopted solutions are suitable for speech and audio event monitoring in a realistic scenario.

A Deep Neural Network approach for Voice Activity Detection in multi-room domestic scenarios

This paper presents a Voice Activity Detector (VAD) for multi-room domestic scenarios. A multi-room VAD (mVAD) simultaneously detects the time boundaries of a speech segment and determines the room where it was generated. The proposed approach is fully data-driven and is based on a Deep Neural Network (DNN) pre-trained as a Deep Belief Network (DBN) and fine-tuned by a standard error back-propagation method. Six different types of feature sets are extracted and combined from multiple microphone signals in order to perform the classification. The proposed DBN-DNN multi-room VAD (simply referred to as DBN-mVAD) is compared to other two NN based mVADs: a Multi-Layer Perceptron (MLP-mVAD) and a Bidirectional Long Short-Term Memory recurrent neural network (BLSTM-mVAD). A large multi-microphone dataset, recorded in a home, is used to assess the performance through a multi-stage analysis strategy comprising multiple feature selection stages alternated by network size and input microphones selections. The proposed approach notably outperforms the alternative algorithms in the first feature selection stage and in the network selection one. In terms of area under precision-recall curve (AUC), the absolute increment respect to the BLST-mVAD is 5.55%, while respect to the MLP-mVAD is 2.65%. Hence, solely the proposed approach undergoes the remaining selection stages. In particular, the DBN-mVAD achieves significant improvements: in terms of AUC and F-measure the absolute increments are equal to 10.41% and 8.56% with respect to the first stage of DBN-mVAD.

Advanced integration of multimedia assistive technologies: A prospective outlook

In the recent years several studies on population ageing in the most advanced countries argued that the share of people older than 65 years is steadily increasing. In order to tackle this phenomena, a significant effort has been devoted to the development of advanced technologies for supervising the domestic environments and their inhabitants to provide them assistance in their own home. In this context, the present paper aims to delineate a novel, highly-integrated system for advanced analysis of human behaviours. It is based on the fusion of the audio and vision frameworks, developed at the Multimedia Assistive Technology Laboratory (MATeLab) of the Università Politecnica delle Marche, in order to operate in the ambient assisted living context exploiting audio-visual domain features. The existing video framework exploits vertical RGB-D sensors for people tracking, interaction analysis and users activities detection in domestic scenarios. The depth information has been used to remove the affect of the appearance variation and to evaluate users activities inside the home and in front of the fixtures. In addition, group interactions are monitored and analysed. On the other side, the audio framework recognises voice commands by continuously monitoring the acoustic home environment. In addition, a hands-free communication to a relative or to a healthcare centre is automatically triggered when a distress call is detected. Echo and interference cancellation algorithms guarantee the high-quality communication and reliable speech recognition, respectively. The system we intend to delineate, thus, exploits multi-domain information, gathered from audio and video frameworks each, and stores them in a remote cloud for instant processing and analysis of the scene. Related actions are consequently performed.

Improving the performance of the AFAMAP algorithm for Non-Intrusive Load Monitoring

Among the many electrical load disaggregation methods, often referred to as Non-Intrusive Load Monitoring techniques, the Additive Factorial Approximate MAP (AFAMAP) algorithm has shown outstanding capabilities and, therefore, it is nowadays regarded as a reference model. In order to achieve more accurate disaggregation results, and to satisfy real life environment requirements, further improvements in the algorithm are needed. In this work, the AFAMAP algorithm has been extended, by means of a differential forward model, thus complementing the existing differential backward model. Furthermore, an aggregated data examination method has been employed, aimed to the detection of inadmissible working state combinations of appliances, as well as the constraints setting based on the reactive power disaggregation feedback. The new approach has been evaluated by means of a subset, spanning over 6 months, of the Almanac of Minutely Power dataset (AMPds). On purpose, a real life environment, accounting 6 appliances, has been modelled and the carried out experiments revealed a improvement up to 18% with respect to the baseline AFAMAP.

User-aided footprint extraction for appliance modelling in Non-Intrusive Load Monitoring

In the area of Non-Intrusive Load Monitoring (NILM), many approaches need a supervised procedure of appliance modelling, in order to provide the informations about the appliances to the disaggregation algorithm and to obtain the disaggregated consumptions related to each one of them. In many approaches, the appliance modelling relies on the consumption footprint, which is a typical working cycle of the appliance. Since the NILM system has only the aggregated power consumption available, the recorded footprint might be corrupted by other appliances, which can not be turned off during this period, i.e., the fridge and freezer in the household. Furthermore, the user needs a facilitated procedure, in order to obtain a clean footprint from the aggregated power signal in real scenario. Therefore, a user-aided footprint extraction procedure is needed. In this work, this procedure is defined as a NILM problem with two sources, i.e., the desired appliance and the fridge-freezer combination. One of the resulting disaggregated profiles of the algorithm corresponds to the extracted footprint. Then, this is used for the appliance modelling stage to create te corresponding Hidden Markov Model (HMM), suitable for the Additive Factorial Approximate Maximum a Posteriori (AFAMAP) algorithm. The effectiveness of the footprint extraction procedure is evaluated through the confidence of the disaggregation output of a real problem, using a span of 30 days data taken from two different datasets (AMPds, ECO). The experiments are conducted using the HMM from the extracted footprint, compared to the confidence of the same problem using the HMM from the true footprint, as appliance level consumption. The results show that the performance are comparable, with the worst relative F1 loss of 3.83%, demonstrating the effectiveness of the footprint extraction procedure.

Short-term load forecasting for smart water and gas grids: A comparative evaluation

Moving from a recent publication of Fagiani et al. [1], short-term predictions of water and natural gas consumption are performed exploiting state-of-the-art techniques. Specifically, for two datasets, the performance of Support Vector Regression (SVR), Extreme Learning Machine (ELM), Genetic Programming (GP), Artificial Neural Networks (ANNs), Echo State Networks (ESNs), and Deep Belief Networks (DBNs) are compared adopting common evaluation criteria. Concerning the datasets, the Almanac of Minutely Power Dataset (AMPds) is used to compute predictions with domestic consumption, 2 year of recordings, and to perform further evaluations with the available heterogeneous data, such as energy and temperature. Whereas, predictions of building consumption are performed with the datasets recorded at the Department for International Development (DFID). In addition, the results achieved for the previous release of the AMPds, 1 year of recordings, are also reported, in order to evaluate the impact of seasonality in forecasting performance. Finally, the achieved results validate the suitability of ANN, SVR and ELM approaches for prediction applications in small-grid scenario. Specifically, for the domestic consumption the best performance are achieved by SVR and ANN, for natural gas and water, respectively. Whereas, the ANN shows the best results for both water and natural gas forecasting in building scenario.

A real-time implementation of an acoustic novelty detector on the BeagleBoard-xM

Novelty detection consists in recognising events that deviate from normality. This paper presents the implementation of a real-time statistical novelty detector on the BeagleBoard-xM. The application processes an incoming audio signal, extracts Power Normalized Cepstral Coefficients and determines whether a novelty sound is present or not based on a statistical model of normality. The novelty detector has been implemented as a standalone graphical application capable of running in real-time on the BeagleBoard-xM platform. Experiments have been conducted to assess the performance of the solution in terms of both detection performance and of real-time capabilities. The results demonstrate that the system is able to operate in real-time on the BeagleBoard-xM with a real-time factor equal to 8.10%, and an F-Measure equal to 77.41%.

Exploiting temporal features and pressure data for automatic leakage detection in smart water grids

In this paper, the unsupervised approach recently proposed by the authors for automatic leakage detection in smart water grids is extended. First of all, the EPANET tool is adopted in order to simulate more realistic leakages. Also, with respect to the original work, an additional time resolution, of 30 minutes, is included, based on the water dataset of the Almanac of Minutely Power Dataset (AMPds). New experiments are performed, as well, to evaluate the results of the application of both temporal features and pressure data. The pressure data is obtained by means of the EPANEt tool, whereas the leakages are induced at run-time for a more realistic behaviour. Two alternative sets of temporal features are evaluated by combining them with the features extracted from both flow and pressure data. Gaussian Mixture Models (GMMs), Hidden Markov Models (HMMs), and One-Class Support Vector Machine (OC-SVM) are used to characterize the normal data behaviour, under a comparative perspective. A feature selection strategy is adopted in computer simulations and the resulting performance indices are evaluated in terms of Area Under Curve (AUC). The obtained results show that the introduction of the temporal information produces a slight performance improvement for both flow and pressure data, but, most importantly, the combination of flow and pressure features allows a significant improvement of leakage detection for both GMM and HMM at every resolution, up to 88% of AUC.