Abstract
Retailers depend on accurate forecasts of product sales at the Store x SKU level to efficiently manage their inventory. Consequently, there has been increasing interest in identifying more advanced statistical techniques that lead to accuracy improvements. However, the inclusion of multiple drivers affecting demand into commonly used ARIMA and ETS models is not straightforward, particularly when many explanatory variables are available. Moreover, regularization regression models that shrink the model's parameters allow for the inclusion of a lot of relevant information but do not intrinsically handle the dynamics of the demand. These problems have not been addressed by previous studies. Nevertheless, multiple simultaneous effects interacting are common in retailing. To be successful, any approach needs to be automatic, robust and efficiently scaleable. In this study, we design novel approaches to forecast retailer product sales taking into account the main drivers which affect SKU demand at store level. To address the variable selection challenge, the use of dimensionality reduction via principal components analysis (PCA) and shrinkage estimators was investigated. The empirical results, using a case study of supermarket sales in Portugal, show that both PCA and shrinkage are useful and result in gains in forecast accuracy in the order of 10% over benchmarks while offering insights on the impact of promotions. Focusing on the promotional periods, PCA-based models perform strongly, while shrinkage estimators over-shrink. For the non-promotional periods, shrinkage estimators significantly outperform the alternatives.

Abstract
The increasing integration of renewable energy sources (RES) at different voltage levels of the distribution grid has led to technical challenges, namely voltage and congestion problems. Conversely, the integration of new Distributed Energy Resources (DER) provides the necessary flexibility to accommodate higher RES integration levels. This work describes the development of innovative functional modules, based on optimal power flow calculations and grid forecasting, dedicated to the predictive management of the distribution grid considering DER flexibility, which are integrated into a commercial SCADA/DMS solution. © The Institution of Engineering and Technology 2023.

Abstract
A versatile, miniaturized, cost-effective, low-power wave profile and tide monitoring system, capable of long-term and scalable deployment, was developed to integrate pressure and temperature sensors in an RS485 network, for standalone operation with organized memory or real-time shared data monitoring. The pressure and temperature sensors are controlled by low-power microcontrollers, that communicate the data periodically to a datalogger, that depending on the application, store it in a removable SD card or send it to a server via Wi-Fi. The data is then analyzed to compensate for the loss in amplitude sensitivity according to the sensor's depth. The wave profile can be sampled at a maximum rate of 100 Hz, with a 1 cm resolution. The system was tested successfully in real-life conditions, in rivers Douro and Cavado, and off the coast of Viana do Castelo.

Abstract
Building evacuation simulation allows for a better assessment of fire safety conditions in existing buildings, which is why it is of interest to develop an easyto-use-Web platform that helps fire safety technicians in this assessment. To achieve this goal, the geometric and physical representation of the building and installed fire safety devices are necessary, as well as the modelling of occupant movement. Although these are widely studied areas, in this paper, we present two new model approaches, either for the physical and geometric representation of a building or for the occupant's movement simulation, during a building evacuation process. To test both models, we develop a multi-agentWeb simulator platform. The tests carried out show the suitability of the model approaches herein presented.

Abstract
Visual representations have the potential to improve information understanding. We explore this idea in the development of NewsLine, an open-source web-based prototype that focuses on narrative visualizations of news content. Having structured data as input, the prototype produces a storyline which showcases the narrative's events and participants, allowing the user to interact with the visualization in a number of ways. We built an information hub around the storyline to allow for multiple levels of exploration, specifically the main visualization, the event information module, and the sidebar. The visualization depicts the sequence of events that make up a news story, as well as the interactions between the involved parties in each event. The event information module presents additional information on a particular event. The sidebar is the “control center” of the visualization, unlocking a number of interactions and configurations. The prototype was evaluated with a user study with journalists and also with an online survey which gathered feedback from 178 potential end users. From these, 106 participants (60.6%) provided a rating of four or above (one to five scale) when asked to quantify their interest in using the application. Moreover, participants were asked to rank the importance of the visualization elements used. The results highlight that two elements stand out as the most important, the events and the entities. Overall, the participants generally found the application to be useful, but in need of some work in order for it to be made available to a broader public. © 2023 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).

Abstract
Identifying stress in people is not a trivial or straightforward task, as several factors are involved in detecting the presence or absence of stress. The problem of detect stress has attracted much attention in the last decade and is mainly addressed with physiological signals and in a controlled ambience with specific tasks. However, the widespread use of video cameras permitted the creation of a new non-invasive data collection techniques. The goal of this work is to provide an alternative way to detect stress in the workplace without the need of specific laboratory conditions. For that, a stress detection model based on images analysed with deep learning neural networks was developed. The trained model achieved a F1 = 79.9% on a binary dataset, of stress/non-stress, with an imbalanced ratio of 0.49. This model can be used in a non-invasive application to detect stress and provide recommendations to the collaborators in the workplace in order to help them to control their stress condition. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.