Abstract
Early identification of failures is a critical task in predictive maintenance, preventing potential problems before they manifest and resulting in substantial time and cost savings for industries. We propose an approach that predicts failures in the near future. First, a deep learning model combining long short-term memory and convolutional neural network architectures predicts signals for a future time horizon using real-time data. In the second step, an autoencoder based on convolutional neural networks detects anomalies in these predicted signals. Finally, a verification step ensures that a fault is considered reliable only if it is corroborated by anomalies in multiple signals simultaneously. We validate our approach using publicly available Air Production Unit (APU) data from Porto metro trains. Two significant conclusions emerge from our study. Firstly, experimental results confirm the effectiveness of our approach, demonstrating a high fault detection rate and a reduced number of false positives. Secondly, the adaptability of this proposal allows for the customization of configuration of different time horizons and relationship between the signals to meet specific detection requirements. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Abstract
Renewable energy generation technologies, heat pumps or electric vehicle (EV) charging stations use active power electronics such as IGBT or MOSFET for AC to DC conversion with the consequence of emissions in the higher frequency range above 2 kHz (non-intentional supraharmonic emissions) and with an impact to the higher frequency grid impedance. In this study the impact of active power electronics on the higher frequency grid impedance in the range up to 150 kHz is analyzed. As existing grid modelling solutions do not consider these technologies sufficiently, this study analyzes the impact of a vehicle to grid (V2G) chargers to a representative distribution grid considering different grid topologies and different types of V2G chargers. The study shows that the additional capacitance and inductance (LCL filter, DC link capacitor) introduced in the electrical grid causes parallel and series resonances in a wide frequency range starting from 500 Hz up to 50 kHz. The grid topology and the number of V2G chargers connected determines the frequency range and characteristics of resonances. Finally, the major contribution of this study is outlining the importance of considering the higher frequency grid impedance for characterization of supraharmonic emissions (primary vs. secondary emissions) and their propagation.

Abstract
Self-consumption regulations are leading to the emergence of new business models proposed by new players and causing traditional players to make new proposals to take advantage of the new business opportunities. In this context, traditional retailers are assessing self-consumption business models, offering management services for self-consumption structures, or the installation of distributed resources, such as solar panels or batteries. Some of the new business models being proposed by electricity suppliers are related to virtual battery services. Indeed, suppliers can, in the free retail market, create innovative tariffs, and design them to make their customers believe they own and manage a battery, even if it does not correspond to a physical battery in the grid. This paper analyses the business model of a supplier offering a virtual battery service, comparing it to the installation of a physical battery, showing that it has no significant benefits compared to more simple approaches.

Abstract
Image segmentation for detecting illegal landfill waste in aerial images is essential for environmental crime monitoring. Despite advancements in segmentation models, the primary challenge in this domain is the lack of annotated data due to the unknown locations of illegal waste disposals. This work mainly focuses on evaluating segmentation models for identifying individual illegal landfill waste segments using limited annotations. This research seeks to lay the groundwork for a comprehensive model evaluation to contribute to environmental crime monitoring and sustainability efforts by proposing to harness the combination of agnostic segmentation and supervised classification approaches. We mainly explore different metrics and combinations to better understand how to measure the quality of this applied segmentation problem.

Abstract
In this paper, the performance analysis of Machine Learning (ML) algorithms for fault analysis in photovoltaic (PV) plants, is given for different algorithms. To make the comparison more relevant, this study is made based on a real dataset. The goal was to use electric and environmental data from a PV system to provide a framework for analysing, comparing, and discussing five ML algorithms, such as: Multilayer Perceptron (MLP), Decision Tree (DT), K-Nearest Neighbors (KNN), Support Vector Machine (SVM) and Light Gradient Boosting Machine (LightGBM). The research findings suggest that an algorithm from the Gradient Boosting family called LightGBM can offer comparable or better performance in fault diagnosis for PV system.

Abstract
Falls are one of the causes of severe hilliness among elders, and the COVID-19 pandemic increased the number of unattended cases because of the social distancing measures. This study aims to create a dataset that collects the data from a 3-axis acceleration sensor fixed on a hinged board apparatus that mimics a human fall event. The datalogging system uses off-the-shelf devices to measure, collect and store the data. The resulting dataset includes data from different angle positions and heights, corresponding to joints of the lower limbs of the human body (ankle, knee, and hip). We use the dataset with a threshold-based fall detection algorithm. The result from the Receiver Operating Characteristic curve shows a good behavior with a mean Area Under the Curve of 0.77 and allow to compute a best threshold value with False Positive Rate of 14.8% and True Positive rate of 89.1%. The optimal threshold value may vary depending on the specific population, activity patterns, and environmental conditions, which may require further customization and validation in real-world settings. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.