Abstract
An all-passive, multipoint, and multiparameter optical monitoring system was developed and deployed in an industrial environment for the simultaneous measurement of methane concentration and other physical parameters. Methane is detected via rapid wavelength modulation spectroscopy (WMS) at 1648.2 nm and 4 MHz frequency. An attenuation invariant quantity defined by the peaks at 0, 4, and 8 MHz of the fast Fourier transform (FFT) of temporal signal is employed, characterized, and validated. Other parameters can concomitantly be measured by fiber Bragg grating (FBG) sensors operating in the 1520-1590 nm range. In the deployed system, the tested parameter was the temperature, which is an important quantity for gas monitoring. The system features a modular architecture that enables scalability up to 16 384 sensing points with an estimated less than 20-min acquisition cycle. In its current deployment, it monitors methane and temperature at eight locations using a single optical network. The system is intended to be used onshore and offshore platforms where the usual monitoring protocol consists of manual measurements usually performed three to four times a year and involves personal displacement and risky situations. Field tests at an onshore natural gas treatment unit (NGTU) demonstrated reliable performance and effective event detection, including undocumented nocturnal emissions, maneuvers at main shut-off valve, and partial plant shutdowns and restarts.

Abstract
Understanding user behavior in search systems is essential for improving retrieval effectiveness and user satisfaction. While prior research has extensively examined general-purpose web search engines, domain-specific contexts-such as sports information-remain comparatively underexplored. In this study, we analyze over 400,000 interaction log entries from a sports-oriented search engine collected over a two-week period. Our analysis combines classic query-level metrics (e.g., frequency distributions, query lengths) with a detailed examination of click behavior, including entropy-based intent variability and a custom query quality scoring model. Compared to established baselines from general and specialized search environments, we observe a high proportion of new and single-term queries, as well as a notable lack of representativeness among top queries. These findings reveal patterns shaped by the event-driven and entity-centric nature of sports content, offering actionable insights for the design of domain-specific retrieval systems.

Abstract
The aim of this work is to explore normalising flows to detect anomalous behaviours which is an essential task mainly for surveillance systems-related applications. To accomplish that, a series of ablation studies were performed by varying the parameters of the Spatio-Temporal Graph Normalising Flows (STG-NF) model [3] and combining it with attention mechanisms. Out of all these experiments, it was only possible to improve the state-of-the-art result for the UBnormal dataset by 3.4 percentual points (pp), for the Avenue by 4.7 pp and for the Avenue-HR by 3.2 pp. However, further research remains urgent to find a model that can give the best performance across different scenarios. The inaccuracies of the pose tracking and estimation algorithm seems to be the main factor limiting the models' performance. The code is available at https://github.com/AnaFilipaNogueira/Abnormal-Human-Behaviour-Detection- using-Normalising-Flows-and- Attention-Mechanisms.

Abstract
Maritime transport faces increasing decarbonisation requirements, placing new demands on port energy systems. Yet most existing studies analyse isolated components or short time horizons, limiting their usefulness for long-term planning. This work develops a holistic, least-cost optimisation model of the Port of Sines energy system using OSeMOSYS, integrating electricity and fuel consumption across port operations and fuel-management processes from 2020 to 2050.The study evaluates alternative technology pathways and policy measures, including carbon taxation, national emission-reduction targets, and the adoption of an innovative ocean-going vessel fleet. Results show that electrification, driven by onshore power supply and renewable expansion, is the most cost-effective decarbonisation route, while its performance depends on local generation capacity and the carbon intensity of the electricity mix. Policy mechanisms and fleet innovation further influence the timing and depth of emissions reductions. Overall, the model provides a replicable framework to support strategic port decarbonisation planning.

Abstract
The limited availability and high cost of acquiring real-world image data impacts the creation of high-quality datasets, hindering the development of robust machine learning models, particularly in complex visual domains. This paper investigates the feasibility of enhancing image classification performance by incorporating balanced synthetic data into existing datasets. Three distinct machine learning tasks—image classification, instance detection, and image segmentation—were explored across diverse image domains. Synthetic images were generated to complement real-world data, and various testing scenarios were conducted, adjusting the relative weights of real and synthetic samples. The results demonstrate that balanced datasets, comprising an equitable mix of real and synthetic images, consistently yielded the highest performance metrics across all tasks. It was also observed that even a small introduction of synthetic data can improve performance over real data alone. The 50–50 split showed to optimally balance the realism of real data and the variability of synthetic data. Real data ensures that the model learns accurate representations of objects, while synthetic data enriches the training process with additional variations, reducing overfitting to specific real-world examples. The proposed approach highlights the potential of strategically integrating synthetic data to improve model accuracy and robustness, particularly in scenarios where real-world data is limited or challenging to acquire. © 2025 Elsevier B.V., All rights reserved.

Abstract
The lack of open-source platforms capable of integrated operational modeling and multi-scenario decarbonization analysis, often hinders data-driven decision-making in the maritime sector. To address this gap, this paper presents an open-source, multi-agent, discrete-event simulator capable of accurately forecasting the energy consumption associated with the diverse assets and activities within a container terminal. The tool's modular architecture enables transparent evaluations of operational strategies and decarbonization alternatives by allowing users to systematically modify inputs or alter embedded energy modules. The tool's capabilities were validated through a case study of a medium-sized Portuguese container terminal. For this particular port, findings indicate that installing three onshore power supply (OPS) units and fully electrifying the internal truck fleet yields the most substantial emission reductions. However, these interventions result in a two-fold increase in daily electricity demand, potentially straining grid capacity. This finding underscores that the effectiveness of terminal electrification as a decarbonization strategy ultimately depends on a simultaneous transition to a decarbonized and secure energy supply.