HAL Portal UTC Université de Technologie de Compiègne
Not a member yet
11652 research outputs found
Sort by
Application of Jellyfish Search Algorithm for Reactive Power Planning-based Power Losses Minimization in Electrical Power Networks
No full textInternational audienceThis study investigates the application of the Jellyfish Search Optimization (JFSO) algorithm for Reactive Power Planning (RPP) to minimize power losses in electrical power networks. The RPP problem is formulated as a multiobjective optimization task that seeks to reduce active power losses and minimize the investment costs of reactive power compensators. The performance of the JFSO algorithm is benchmarked against Differential Evolution (DE) and Particle Swarm Optimization (PSO) algorithms on the IEEE 30-bus test system under two case studies: (1) minimizing active power losses alone and (2) minimizing both losses and investment costs. For minimizing the losses in Case 1, JFSO reduced power losses from an initial 5.6 MW with 18.2% reduction, compared to 16.8% for DE and 11.0% for PSO. In Case 2, JFSO minimized the total costs to 2,391,466dollars/year, outperforming DE and PSO, which resulted in 2,435,863dollars/year and 2,629,826dollars/year, respectively. The findings demonstrate that JFSO offers superior optimization performance with enhanced convergence properties. By effectively balancing investment and operational costs, the proposed algorithm provides a robust solution for efficient RPP
Two copper layer Insulated Metal Substrate PCB potential using Wide-Bandgap semiconductors
No full textInternational audienceThis paper evaluates two copper-layer Insulated Metal Substrate (IMS) PCBs for power converters, highlighting reduced parasitic inductance and strong thermal performance. It compares various layouts, considering parasitic capacitance, inductance, thermal resistance, and power density, demonstrating exceptional performance in a 4 kVA half-bridge with a 400 V dc-link design
The role of scientific research in the energy transition: Uncertainties, interference with values, and the difficulties of communication with decision-makers
No full textInternational audienceOn the one hand, science-based policies require to be based on up-to-date sets of knowledge developed by the scientific community. On the other, these sets of knowledge suffer from gaps and uncertainties, scientists are also citizens and defend political and ethical values that may interfere with their expertise, and decision-makers are sometimes unfamiliar with scientific vocabulary. In the case of energy, these difficulties overlap with the needs of democracy, i.e, the necessary public hearing of citizens whose preferences might run counter to scientific recommendations. How should scientists take these various difficulties into account? Should they allow their values overcome their expertise? The talk will address ways to address these questions coming from the social and behavioral sciences
Test allocation based on risk of infection from first and second order contact tracing
No full textUnder limited available resources, strategies for mitigating the propagation of an epidemic such as random testing and contact tracing become inefficient. Here, we propose to accurately allocate the resources by computing over time an individual risk of infection based on the partial observation of the epidemic spreading on a contact network; this risk is defined as the probability of getting infected from any possible transmission chain up to length two, originating from recently detected individuals. To evaluate the performance of our method and the effects of some key parameters, we perform comparative simulated experiments using data generated by an agent-based model
Passivity Analysis in Power Electronic Converters with SHE Modulation
No full textInternational audienceTo ensure the stability of modern power grids, particularly those dominated by power electronics, it is convenient for grid-connected converters to exhibit passivity across the entire frequency spectrum. Furthermore, it is interesting not only for the system to be passive, that is, for the converter system's resistance to be positive, but also for this resistance to have a high value, as it helps dampen potential grid resonances. This article presents a study of a grid-connected converter, achieving passivity and increasing the converter's resistance through control strategies. Given the growing use of Selective Harmonic Elimination (SHE) modulation techniques in high-power converters, the analysis will also focus on how to assess the passivity of a system employing these modulation techniques
A Multi-Agent Deep Reinforcement Learning Approach for Traffic Management in Complex Communication Networks
No full textInternational audienceModern communication networks like 5G and 6G are increasingly integrating Distributed Artificial Intelligence (DAI) to provide fast decision-making services like traffic management despite both the unpredictable patterns of network traffic and the intrinsic dynamism of the underlying communication network. In particular, Distributed Artificial Intelligence will enable optimal network resource usage and prevent network congestion, addressing the challenges posed by the dynamic patterns characterizing complex communication networks like 5G and 6G networks. This paper focuses on designing and assessing a new traffic management solution based on a Multi-Agent Deep Reinforcement Learning (MA-DRL). Our solution aims at adapting to network conditions to prevent network traffic congestion, while improving throughput, latency, and loss compared to existing traffic management methods
Context-Aware Multi-Criteria Recommender Systems Using Variable Selection Networks
No full textInternational audienceConventional recommender systems, which rely on a single criterion such as overall rating, often fail to capture the complexity of user preferences and the influence of contextual information. Context-aware multi-criteria recommender systems address these limitations by incorporating multiple dimensions of user preferences, item attributes, and contextual factors, leading to more accurate and relevant recommendations. This paper presents a context-aware multi-criteria recommender system using variable selection networks. Our approach dynamically selects the most relevant features from a variety of inputs, including user profiles, item characteristics, multiple criteria, and contextual factors, to enhance personalization. By leveraging deep learning-based variable selection networks, our model significantly improves recommendation accuracy and interpretability, outperforming several baseline models in experimental evaluations. This advancement underscores the importance of integrating both multi-criteria and context-aware methodologies in modern recommender systems
