HAL-MINES ParisTech
Not a member yet
27348 research outputs found
Sort by
On the optimal selection of meteorological variables as inputs to machine learning models for solar irradiance forecasting
No full textInternational audienceVariables output from numerical weather models are essential and widely used as inputs to machine learning models for forecasting the solar irradiance available at ground level. However, meteorological models generate hundreds of variables, many of which may not contribute meaningfully to the performance of the predictive model. This abundance of data necessitates an effective optimization strategy to identify the most relevant features and eliminate those with limited impact in order to reduce computational resources.The strategy described here includes the introduction of novel meteorological features and the removal of less relevant or redundant variables. Feature selection is guided by statistical metrics such as Pearson and Spearman correlation coefficients. It is complemented by a permutation feature importance method using on-the-shelf machine learning models such as XGboost and random forest regressor. This enables linear and non-linear evaluation of the influence of each variable on the target variable.This optimization process shows that several variables often used as inputs to forecast models, including wind velocity at 10 m or precipitation, have a negligible impact on the forecasting accuracy and may be removed. Concurrently, novel features are introduced, such as cloud cover, atmospheric aerosol optical depth, and ground albedo, which exhibit a strong correlation with solar irradiance and contribute positively to the predictive capability of the model.The optimized feature set is tested against our commercial model for short-term and day-ahead solar irradiance forecasting. Post-optimization results show a relative improvement greater than 10 % for each forecast horizon, as measured by standard error metrics like RMSE and MAE, across diverse weather conditions and geographic locations.This study highlights the importance of a data-driven approach to feature optimization in solar forecasting models. By systematically evaluating and refining input variables, significant enhancements in model performance may be achieved
Ternary Choline Chloride-Based Deep Eutectic Solvents: A Review: Review
No full textInternational audienceTernary choline chloride-based deep eutectic solvents (TDESs) exhibit unique physicochemical properties, including lower viscosities, lower melting points, higher thermal stabilities, and enhanced solvations compared to binary deep eutectic solvents (BDESs). Although BDESs have been widely studied, the addition of a third component in TDESs offers opportunities to further optimize their performance. This review aims to evaluate the physicochemical properties of TDESs and highlight their potential applications in sustainable industrial processes compared to BDESs. A comprehensive analysis of the existing literature was conducted, focusing on TDES properties, such as phase behavior, density, viscosity, pH, conductivity, and the effect of water, along with their applications in various fields. TDESs demonstrated superior physicochemical characteristics compared to BDESs, including improved solvation and thermal stability. Their applications in biomass conversion, CO 2 capture, heavy oil upgrading, refrigeration gases, and as solvents/catalysts in organic reactions show significant promise for enhancing process efficiency and sustainability. Despite their advantages, TDESs face challenges including limited predictive models, potential instability under certain conditions, and scalability hurdles. Overall, TDESs offer significant potential for advancing sustainable and efficient chemical processes for industrial applications.</div
EXISTENCE OF TWO SPONTANEOUS NERVE INFLUX OF DIFFERENT VELOCITIES IN A GIVEN AXON, NUMERICAL EVIDENCE AND SENSITIVITY ANALYSIS
No full textWe prove in this paper, numerically, that there exists at leas two propagation velocities in the original Hodgkin-Huxley system, obtained by a numerical method coming from the analysis of the unstable manifold of the associated ODE. None of the velocities is the one identified by Hodgkin and Huxley. We prove numerically that a sensitivity analysis can be performed on these two velocities and indicate small variations when the constants varies
Embedding sufficiency: A case study of sufficiency-oriented tools and strategies in the cleaning industry
No full textInternational audienceThis study explores the case of Clinitex, a professional cleaning services provider, which has been engaged in a seven-year transition to a sufficiency-oriented organisational model. Through a detailed case study, we explore the experimental tools, strategies, and the overall dispositif that enabled this shift
Improving Clinical Decision-Making in Radiotherapy: A Comparative Analysis of LQ and LQL Dose Models
No full textInternational audienceRadiotherapy is an essential component of cancer treatment, requiring accurate dose planning to optimize tumor control while sparing healthy tissues. This study, originating from a radiobiology workshop held during the 27 th Congrès National de Cancérologie et de Radiothérapie-2024 in Sousse, Tunisia, aims to investigate advanced dose modeling approaches, focusing on the Linear Quadratic (LQ) and Linear Quadratic Linear (LQL) models, to refine the calculation of biologically effective doses (BED) and improve treatment personalization. The workshop brought together experts in the field to discuss and evaluate the latest advancements in dose modeling, providing a comprehensive overview of current best practices and emerging trends. Using tools such as LQL-equiv and other BED calculators, we integrated patient-specific data (e.g., fractionation schedules and organ-at-risk (OAR) constraints) to predict outcomes such as normal tissue complication probabilities (NTCP). Unlike many theoretical studies, our approach embeds these models within a unified interface tailored to real clinical scenarios, enabling practitioners to simulate and adjust treatment plans based on complex, practical constraints. Through a series of clinical case studies (including treatment interruptions, palliative boosts, and reirradiation scenarios), participant responses were analyzed using the Jaccard similarity index, revealing a significant lack of consensus in treatment planning decisions (mean agreement of 25.83%). This variation illustrates the current ambiguity among clinicians regarding which model to use and how to apply it, despite access to advanced tools. This heterogeneity in decision-making could lead to divergent treatment recommendations for patients with clinically similar profiles. While the LQ and LQL models offer promising tools for personalized radiotherapy, their interpretation and implementation remain highly variable. In addition, the question of professional responsibility in dose equivalence calculations emerged as a key issue, as many departments lack clearly defined accountability frameworks. This study emphasizes the need for standardized guidelines, enhanced training programs, and decision-support systems to reduce inter-observer variability and ensure effective clinical adoption, ultimately improving patient care. The findings underscore the importance of harmonizing predictive modeling practices to achieve more consistent and effective radiotherapy outcomes.</div
Investigating the "blues" of safety professionals
No full textInternational audienceThe aim of this article is twofold. First, it introduces a phenomenon of discontent among safety professionals, termed the "blues," expressed through several recent critical publications about their profession. Although different in tone and geographical origin, these publications share similar analyses about the profession's dysfunctions. The main criticisms focus on excessive bureaucratisation, simplistic approaches to safety, disconnection from field reality, and lack of professional recognition. The authors of these books particularly denounce the excessive time spent on reports and indicators at the expense of more concrete prevention actions. Second, the article outlines a research project to understand better the representativeness of this "blues" and its implications for the profession, particularly by exploring links with the quest for meaningful work and mechanisms of dissonance between professional ideals and field reality
Préface: l'importance du "non" pour l'innovation
No full textInternational audienceL’innovation doit être mise en accusation, doit être condamnée. Ce qu’elle cause, ce qu’elle fait subir, sa domination n’est plus acceptable. L’injustice qu’elle peut créer comme les inégalités sont à mettre à l’index. « L’innovation standard » a fait son temps tant elle est nuisible aussi bien à l’environnement qu’aux systèmes politiques, aussi bien à l’humain qu’à son développement. Pourtant, formidable fille du progrès, l’innovation est celle qui nous a permis d’être ce que nous sommes, d’avoir fait en sorte que la santé, l’éducation, la sécurité, le vivre-ensemble soient possibles et ne cessent de progresser. C’est donc son détournement qui est mis sur le banc des accusés et l’enjeu de ce Manifeste n’est pas de faire cesser l’innovation mais de la (re)penser. Ce Manifeste est l’occasion de formuler un vœu : celui d’innover plus que jamais, car nous n’avons jamais eu autant besoin d’innovations. Toutefois, sous une forme inédite : non plus guidée par le seul profit, mais par le mieux vivre ; non plus soumise à la rentabilité exclusive, mais orientée vers le préférable; non plus administrée par une hiérarchie illégitime, mais par une communauté de choix. Il s’agit de faire sortir l’innovation du mondain, de l’arracher à ses griffes d’immanence et de l’inciter à s’élever et s’élargir à partir d’une posture de transcendance. La sortie du mondain est donc une ambition singulière pour l’innovation, car cela l’oblige à prendre un chemin inédit où elle risque de se perdre, de ne plus retrouver ses réflexes. Pourtant, c’est dans cet inconfort qu’émergera l’innovation non-standard
Des minéraux pour la transition. Comment les études de criticité posent la question environnementale
No full textInternational audienceThe article focuses on criticality studies, a form of expert knowledge aimed at anticipating the need for minerals or metals (referred to as “critical resources”) for a given economy. These studies saw significant growth in the 20th century in many countries, particularly during wartime, to guide public action and ensure supply security. By examining two recent criticality studies conducted by the European Commission in 2023 and the French agency in charge of supervising the electric grid (Réseau du transport d’électricité, or RTE) in 2021, we analyse how both institutions frame the environmental issue when considering criticality. We demonstrate that, for the European Commission, the environment is viewed as an external factor influencing the European economy, with potential impacts on metal market prices. In contrast, for the RTE, the environment is framed as a set of material, social, and technical constraints that shape the ability to secure new materials. This second version of criticality raises the question of the desirability of metal extraction in relation to use, which should be debated collectively.L’article analyse les études de criticité, forme de savoir expert qui vise à anticiper des besoins en minerais ou métaux (qualifiés alors de « ressources critiques ») pour une économie donnée. Ces études ont connu un essor important au XXe siècle dans de nombreux pays pour guider l’action publique, notamment en périodes de guerre, afin de sécuriser les approvisionnements. En s’intéressant à la façon dont deux études récentes ont été menées par la Commission européenne en 2023 et le Réseau de transport d’électricité en France (RTE) en 2021, nous analysons la façon dont les deux institutions cadrent la question environnementale lorsqu’elles raisonnent en termes de criticité. Nous montrons que, pour la Commission européenne, l’environnement est considéré comme un contexte susceptible d’avoir des conséquences sur les prix de marchés des métaux, alors que pour RTE, l’environnement est un ensemble de contraintes matérielles, sociales et techniques qui jouent sur les possibilités d’obtenir de nouveaux matériaux. Cette seconde version de la criticité pose la question de la désirabilité de l’extraction de métaux en lien avec leurs usages à débattre collectivement
Information Management and Big Data
No full textInternational audienceThis book constitutes the proceedings of the 11th Annual International Conference on Information Management and Big Data, SIMBig 2024, held in Ilo, Peru, during November 20–22, 2024.The 27 full papers and 1 short paper included in this book were carefully reviewed and selected from 102 submissions. They were organized in topical sections as follows: machine learning and deep learning; natural language processing; mining of social networks and online platforms; and signal and image processing