Non-Intrusive Load Monitoring (NILM) refers to the analysis of the aggregated current and voltage measurements of Home Electrical Appliances (HEAs) recorded by the house electrical panel. Such methods aim to identify each HEA for a better control of the energy consumption and for future smart grid applications. Here, we are interested in an event-based NILM pipeline, and particularly in the HEAs’ recognition step. This paper focuses on the selection of relevant and understandable features for efficiently discriminating distinct HEAs. Our contributions are manifold. First, we introduce a new publicly available annotated dataset of individual HEAs described by a large set of electrical features computed from current and voltage measurements in steady-state conditions. Second, we investigate through a comparative evaluation a large number of new methods resulting from the combination of different feature selection techniques with several classification algorithms. To this end, we also investigate an original feature selection method based on a deep neural network architecture. Then, through a machine learning framework, we study the benefits of these methods for improving Home Electrical Appliance (HEA) identification in a supervised classification scenario. Finally, we introduce new transfer learning results, which confirm the relevance and the robustness of the selected features learned from our proposed dataset when they are transferred to a larger dataset. As a result, the best investigated methods outperform the previous state-of-the-art results and reach a maximum recognition accuracy above 99% on the PLAID evaluation dataset.

Since decades past, time–frequency (TF) analysis has demonstrated its capability to efficiently handle non-stationary multi-component signals which are ubiquitous in a large number of applications. TF analysis us allows to estimate physics-related meaningful parameters (e.g., $F_0$, group delay, etc.) and can provide sparse signal representations when a suitable tuning of the method parameters is used. On another hand, deep learning with Convolutional Neural Networks (CNN) is the current state-of-the-art approach for pattern recognition and allows us to automatically extract relevant signal features despite the fact that the trained models can suffer from a lack of interpretability. Hence, this paper proposes to combine together these two approaches to take benefit of their respective advantages and addresses non-intrusive load monitoring (NILM) which consists of identifying a home electrical appliance (HEA) from its measured energy consumption signal as a “toy” problem. This study investigates the role of the TF representation when synchrosqueezed or not, used as the input of a 2D CNN applied to a pattern recognition task. We also propose a solution for interpreting the information conveyed by the trained CNN through different neural architecture by establishing a link with our previously proposed “handcrafted” interpretable features thanks to the layer-wise relevant propagation (LRP) method. Our experiments on the publicly available PLAID dataset show excellent appliance recognition results (accuracy above 97%) using the suitable TF representation and allow an interpretation of the trained model.

This letter extends our recently introduced method which was designed to estimate instantaneous frequency and chirp rate of linearly modulated signals. Indeed, we derive several new estimators related to our previous ones which provide in the time-frequency plane all the signal parameters of the investigated model: amplitude, frequency, and their local modulations (AM/FM). Our estimators are first introduced and compared in terms of statistical efficiency with theoretical bounds and with other state-of-the-art estimators. Then, they are used to improve spectral analysis applied to audio sinusoidal modeling. Finally, they lead to a new source separation technique based on coherent amplitude and frequency modulation that is evaluated on real-world music signals.

This paper addresses the problem of estimating the instantaneous frequency (IF) and amplitude of the modes composing a non-stationary multi-component signal in the presence of noise. A novel observation model for the signal spectrogram is developed within a Bayesian framework to handle intricate configurations involving noise or overlapping components. The model parameters are estimated using a stochastic variant of the Expectation-Maximization algorithm, bypassing the computationally expensive joint parameter estimation from the posterior distribution. We then design an algorithm for instantaneous amplitude and frequency estimation that accounts for overlap and amplitude variations of the components. To assess the performance of the proposed method, we conduct experiments on both real-world and simulated signals, involving separated or crossing modes. The benefits of our method in terms of efficiency compared with several state-of-the art techniques appear to be significant in that latter case, but also when the amplitude of the components are varying across time.

This paper addresses the problem of disentangling nonoverlapping multicomponent signals from their observation being possibly contaminated by external additive noise. We aim to extract and to retrieve the elementary components (also called modes) present in an observed nonstationary mixture signal. To this end, we propose a new pseudo-Bayesian algorithm to perform the estimation of the instantaneous frequency of the signal modes from their time-frequency representation. In a second time, a detection algorithm is developed to restrict the time region where each signal component behaves, to enhance quality of the reconstructed signal. We finally deal with the presence of noise in the vicinity of the estimated instantaneous frequency by introducing a new reconstruction approach relying on nonbinary band-pass synthesis filters. We validate our methods by comparing their reconstruction performance to state-of-the-art approaches through several experiments involving both synthetic and real-world data under different experimental conditions.