The increasing demand for efficient energy management in smart grids has led to the development of various Non-Intrusive Load Monitoring (NILM) techniques. These aim to disaggregate energy consumption and classify appliances using data from a single-point smart meter at a household’s grid mains. With the use of machine learning methods, NILM solutions increasingly rely on datasets for training and validation. While datasets like WHITED, BLOND, and UK-DALE provide insights into consumption patterns, they face limitations such as lack of steady-state data, complicated ground-truth or low sampling rates, which hinder detecting low-power appliances. High sampling rates, however, improve classification accuracy and enable identifying these devices. This study introduces the HIgh Frequency household electrical signals DAtaset (HIFDA), a high-frequency dataset capturing steady-state signals from 14 household appliances at 100 kSPS, including the empty grid. Data, collected via a custom System-on-Chip (SoC) device, focuses on active consumption and includes multiple time windows. HIFDA, hosted on Zenodo, ensures its suitability for modern NILM research and other applications.

Electromyography (EMG) signals are fundamental in neurorehabilitation as they provide a non-invasive means of capturing the electrical activity of muscles, enabling precise detection of motor intentions. This capability is essential for controlling assistive devices, such as therapeutic exoskeletons, that aim to restore mobility and improve motor function in patients with neuromuscular impairments. The integration of EMG into neurorehabilitation systems allows for adaptive and patient-specific interventions, addressing the variability in motor recovery needs. However, achieving the high fidelity, low latency, and robustness required for real-time control of these devices remains a significant challenge. This paper introduces a novel multi-channel electromyography (EMG) acquisition system implemented on a System-on-Chip (SoC) architecture for robotic neurorehabilitation. The system employs the Zynq-7000 SoC, which integrates an Advanced RISC Machine (ARM) processor, for high-level control and an FPGA for real-time signal processing. The architecture enables precise synchronization of up to eight EMG channels, leveraging high-speed analog-to-digital conversion and advanced filtering techniques implemented directly at the measurement site. By performing filtering and initial signal processing locally, prior to transmission to other subsystems, the system minimizes noise both through optimized processing and by reducing the distance to the muscle, thereby significantly enhancing the signal-to-noise ratio (SNR). A dedicated communication interface ensures low-latency data transfer to external controllers, crucial for adaptive control loops in exoskeletal applications. Experimental results validate the system’s capability to deliver high signal fidelity and low processing delays, outperforming commercial alternatives in terms of flexibility and scalability. This implementation provides a robust foundation for real-time bio-signal processing, advancing the integration of EMG-based control in neurorehabilitation devices.

Energy disaggregation represents a fundamental approach in energy management, allowing total electrical consumption to be analyzed and broken down into individual appliances without the need for intrusive installations. Non-Intrusive Load Monitoring (NILM) emerges as an innovative methodology in this context, offering the possibility to monitor and optimize household energy consumption. This paper presents an advanced Convolutional Transformer-based model for NILM implemented under a sequence-to-point (S2P) setting for energy disaggregation. Compared to previous proposals, the model introduces a compact Convolutional Transformer architecture for S2P NILM by combining positional embeddings, multi-head attention, pointwise convolutions, and dilated convolutions. The research evaluates the generalization capability of the model using never-before-seen houses within the same domain through the REFIT dataset, demonstrating a higher performance compared to several state-of-the-art models. In addition, the study explores the generalization capability of the model by evaluating its adaptability to houses belonging to different contexts, specifically using the UK-DALE dataset, thus proving the alternative of deploying the trained model in different environments/houses. The results indicate that the proposed model not only excels in terms of disaggregation performance, achieving the lowest average MAE (13.37) and SAE (12.46) across all appliances, but also shows robust performance in diverse scenarios. Compared to the best-performing reference model ([33], with MAE = 40.24 and SAE = 16.95), the proposed method reduces the average error by approximately 67% in MAE and 21% in SAE.