Abstract
This paper explores the potential of Large Language Models (LLM) to optimize various stages of the software development lifecycle, including requirements elicitation, architecture design, diagram creation, and implementation. The study is grounded in a real-world case, where development time and result quality are compared with and without LLM assistance. This research underscores the possibility of applying prompt patterns in LLM to support and enhance software development activities, focusing on a B2C digital commerce platform centered on fashion retail, designated LUNA. The methodology adopted is Design Science, which follows a practical and iterative approach. Requirements, design suggestions, and code samples are analyzed before and after the application of language models. The results indicate substantial advantages in the development process, such as improved task efficiency, faster identification of requirement gaps, and enhanced code readability. Nevertheless, challenges were observed in interpreting complex business logic. Future work should explore the integration of LLM with domain-specific ontologies and business rule engines to improve contextual accuracy in code and model generation. Additionally, refining prompt engineering strategies and combining LLM with interactive development environments could further enhance code quality, traceability, and explainability.

Abstract
A fast response to faults in large-scale photovoltaic power plants (PVPPs), which can occur on hundreds of components like photovoltaic panels and inverters, is fundamental for maximizing energy generation and reliable system operation. This work proposes using a Graph Neural Network (GNN) combined with a digital twin for synthetic fault data scenario generation for fault location in PVPPs. It shows that GNN can adapt to system changes without requiring model retraining, thus offering a scalable solution for the real operating PVPPs, where some parts of the system may be disconnected for maintenance. The results for a real PVPP show the GNN outperforms baseline models, especially in larger topologies, achieving up to twice the accuracy in a fault location task. The GNN's adaptability to topology changes was tested on the simulated reconfigured systems. A decrease in performance was observed, and its value depends on the complexity of the original training topology. It can be mitigated by using several system reconfigurations in the training set.

Abstract
The HyStorization project aims to advance the modelling and operational understanding of hybrid electrochemical energy storage systems, focusing on Lithium-ion (Li-ion) and Vanadium Redox Flow Batteries (VRFBs). These technologies are key enablers of flexible, reliable, and scalable grid-scale energy storage. While Li-ion batteries are well-established for high-power applications, VRFBs offer promising advantages for medium- to long-duration storage due to their durability and decoupled energy and power capacities. The primary objective is to develop linearized battery models for both technologies, derived from experimental data, that accurately capture efficiency and power limits as functions of the State of Charge (SoC). These models are intended for integration into Mixed-Integer Linear Programming (MILP) tools to optimize energy dispatch in hybrid storage systems. A comprehensive testing campaign was conducted on three BYD stationary Li-ion battery systems. Due to a malfunction in one unit, the remaining three—of similar age and usage—were treated as a single representative system. A Python-based controller was developed to automate cycling and collect high-resolution data (1-second intervals) via HTTP. The testing protocol included: • Constant power cycles for initial validation and degradation screening. • Constant current cycles for parameter extraction. Key findings include: • A slight but consistent improvement in SoC estimation accuracy using a linear model over a bucket model (~2% reduction in MAE and MSE). • Shorter resampling intervals (e.g., 1-minute vs. 15-minute) improved accuracy, but the most significant reduction in error came from refreshing the SoC with real measurements rather than relying on estimated values. • SoC limits, while useful for safety, were found to be overly restrictive and may not reflect the battery’s full operational flexibility. • Attempts to assess cyclic degradation were inconclusive due to the limited number of cycles and short observation window. The final linear model includes parameters for nominal charge/discharge voltages, inverter efficiencies, and dynamic SoC limits as functions of DC power. These were validated against real operational data and compared with manufacturer-based models. Concerning the VRFB, the project originally planned to conduct targeted tests on the VRFB to: • Evaluate energy efficiency across different SoC levels and operational ranges. • Determine maximum and minimum effective power ratings as functions of SoC. • Support the development of non-linear models that will be linearized for MILP integration. However, due to a malfunction, the VRFB could not be tested as planned. Instead, the projectrelied on previously collected characterization data, which did not fully cover the intended test scope. Despite these limitations, the available data was used to: • Analyse energy efficiency trends across selected states of charge (SoC) and operational conditions. • Estimate effective power ratings within the constraints of the existing dataset. • Support the preliminary development of non-linear models, with the aim of future linearization for MILP integration. While these efforts provided valuable insights, the absence of new experimental data limited the ability to fully capture the unique operational characteristics of VRFBs, such as their decoupled energy and power capabilities and their suitability for long-duration storage. The project is expected to deliver: • Validated, MILP-compatible models for both Li-ion and VRFB technologies. • Enhanced dispatch strategies for hybrid storage systems. • Improved integration of real-time SoC measurements to reduce estimation error. • Recommendations for longer-term testing to better assess degradation and refine model accuracy. In conclusion, the HyStorization project provides a foundational step toward more accurate, data-driven modelling of hybrid storage systems. It highlights the importance of real-time data, flexible modelling approaches, and the need for continued testing to support the evolving role of batteries in grid operations.

Abstract
The detection of glucose is crucial for diagnosing diseases such as diabetes and enables timely medical intervention. In this study, a disposable enzymatic screen-printed electrode electrochemical biosensor enhanced with machine learning (ML) for quantifying glucose in serum is presented. The platinum working surface was modified by chemical adsorption with biographene (BGr) and glucose oxidase, and the enzyme was encapsulated in polydopamine (PDP) by electropolymerisation. Electrochemical characterisation and morphological analysis (scanning and transmission electron microscopy) confirmed the modifications. Calibration curves in Cormay serum (CS) and selectivity tests with chronoamperometry were used to evaluate the biosensor's performance. Non-linear ML regression algorithms for modelling glucose concentration and calibration parameters were tested to find the best-fit model for accurate predictions. The biosensor with BGr and enzyme encapsulation showed excellent performance with a linear range of 0.75-40 mM, a correlation of 0.988, and a detection limit of 0.078 mM. Of the algorithms tested, the decision tree accurately predicted calibration parameters and achieved a coefficient of determination above 0.9 for most metrics. Multilayer perceptron models effectively predicted glucose concentration with a coefficient of determination of 0.828, demonstrating the synergy of biosensor technology and ML for reliable glucose detection.

Abstract
The knowledge of the precise 3D position of a target in tracking applications is a fundamental requirement. The lack of a low-cost single sensor capable of providing the three-dimensional position (of a target) makes it necessary to use complementary sensors together. This research presents a Local Positioning System (LPS) for outdoor scenarios, based on a data fusion approach for unmodified UAV tracking, combining a vision sensor and mmWave radar. The proposed solution takes advantage of the radar's depth observation ability and the potential of a neural network for image processing. We have evaluated five data association approaches for radar data cluttered to get a reliable set of radar observations. The results demonstrated that the estimated target position is close to an exogenous ground truth obtained from a Visual Inertial Odometry (VIO) algorithm executed onboard the target UAV. Moreover, the developed system's architecture is prepared to be scalable, allowing the addition of other observation stations. It will increase the accuracy of the estimation and extend the actuation area. To the best of our knowledge, this is the first work that uses a mmWave radar combined with a camera and a machine learning algorithm to track a UAV in an outdoor scenario.

Abstract