La forte variabilité de la vitesse du vent fait que l'énergie produite par un parc éolien n'est pas constante dans le temps. Le gestionnaire ne peut donc pas dimensionner son réseau électrique en prenant intégralement ce type de production en compte. L' une des solutions préconisées pour permettre le développement de l' éolien et son intégration avec une plus grande sureté aux réseaux, est de développer et d'améliorer les outils de prévisions. Le travail de thèse consiste à améliorer les performances d'un outil de prédiction basé sur les réseaux de neurones bayesiens, permettant la prédiction de la puissance à très court terme . Le prédicteur fonctionne notamment par J'ajustement de paramètres, certain se détermine « automatiquement » via le mécanisme des réseaux de neurones bayesiens d' autres, que nous nommerons paramètres temporels, sont à l' appréciation de l'utilisateur. Le travail mené consiste à établir un protocole pour la fixation de ces paramètres tout en améliorant les performances du prédicteur . Nous avons donc décidé de conditionner leurs valeurs en fonction de la variabilité des séquences de puissance précédent l'instant de prévision. Tout d'abord nous avons classifié des séquences de puissance en fonction de leurs coefficients de variation en appliquant la méthode des C-moyennes floues. Puis, chaque classe formée a été testée sur plusieurs valeurs de paramètres, les valeurs associées aux meilleures prédictions ont été retenues. Enfin, ces résultats couplés au formalisme des Chaines de Markov, par le biais de la matrice de transition , ont perm is d'obtenir des taux d'amélioration par rapport à la persistance allant de 7,73 à 23,22 % selon l'horizon de prédiction considéré / The high variability of the wind speed has for conse quences that the energy produced by a wind farm is not constant over time. Therefore, the manager can't size the electrical network by takin g into account this type of production. One solution advocated for the development of wind energy and its integrati on with greater security at network, is to develop and improve fore casting tools. The thesi s objective is to improve the performance of a predi ction tool based on Bayesian neural networks, allowing the predi ction of wind power for short timescales. The predictor works, in part icular by the adjustment of parameters, sorne is determined "automatically" through the mechan ism of neural networks Bayesian other , which we cali temporal parameters are at the discretion of the user. The work involves establishing a protocol for the determination of these parameters and improving the performance of the predictor. So, we decided to condition their values depending on the sequence variability of wind power previous the moment of the forecast. First we classified sequences of power according to their coefficients of variation using the method of fuzzy C-means. Then, each formed class was tested for several parameters values, the values associated with the best predictions were selected. Finally , these result s coupled with the formalism of Markov chains , through the transition matrix allowed to obtain rates of improvement over the persistence ranging from 7.73 to 23.22 % depending on the prediction horizon considered

Deep Neural Networks (DNNs) have proven excellent performance and are very successful in image classification and object detection. Safety critical industries such as the automotive and aerospace industry aim to develop autonomous vehicles with the help of DNNs. In order to certify the usage of DNNs in safety critical systems, it is essential to prove the correctness of data within the system. In this thesis, the research is focused on investigating the sources of uncertainty, what effects various sources of uncertainty has on NNs, and how it is possible to reduce uncertainty within an NN. Probabilistic methods are used to implement an NN with uncertainty estimation to analyze and evaluate how the integrity of the NN is affected. By analyzing and discussing the effects of uncertainty in an NN it is possible to understand the importance of including a method of estimating uncertainty. Preventing, reducing, or removing the presence of uncertainty in such a network improves the correctness of data within the system. With the implementation of the NN, results show that estimating uncertainty makes it possible to identify and classify the presence of uncertainty in the system and reduce the uncertainty to achieve an increased level of integrity, which improves the correctness of the predictions.

Text plays a pivotal role in our daily lives, encompassing various forms such as social media posts, news articles, books, reports, and more. Consequently, Natural Language Processing (NLP) has garnered widespread attention. This technology empowers us to undertake tasks like text classification, entity recognition, and even crafting responses within a dialogue context. However, despite the expansive utility of NLP, it frequently necessitates a critical decision: whether to place trust in a model's predictions. To illustrate, consider a state-of-the-art (SOTA) model entrusted with diagnosing a disease or assessing the veracity of a rumor. An incorrect prediction in such scenarios can have dire consequences, impacting individuals' health or tarnishing their reputation. Consequently, it becomes imperative to establish a reliable method for evaluating the reliability of an NLP model's predictions, which is our focus-uncertainty estimation on NLP. Though many works have researched uncertainty estimation or NLP, the combination of these two domains is rare. This is because most NLP research emphasizes model prediction performance but tends to overlook the reliability of NLP model predictions. Additionally, current uncertainty estimation models may not be suitable for NLP due to the unique characteristics of NLP tasks, such as the need for more fine-grained information in named entity recognition. Therefore, this dissertation proposes novel uncertainty estimation methods for different NLP tasks by considering the NLP task's distinct characteristics. The NLP tasks are categorized into natural language understanding (NLU) and natural language generation (NLG, such as text summarization). Among the NLU tasks, the understanding could be on two views, global-view (e.g. text classification at document level) and local-view (e.g. natural language inference at sentence level and named entity recognition at token level). As a result, we research uncertainty estimation on three tasks: text classification, named entity recognition, and text summarization. Besides, because few-shot text classification has captured much attention recently, we also research the uncertainty estimation on few-shot text classification. For the first topic, uncertainty estimation on text classification, few uncertainty models focus on improving the performance of text classification where human resources are involved. In response to this gap, our research focuses on enhancing the accuracy of uncertainty scores by bolstering the confidence associated with winning scores. we introduce MSD, a novel model comprising three distinct components: 'mix-up,' 'self-ensembling,' and 'distinctiveness score.' The primary objective of MSD is to refine the accuracy of uncertainty scores by mitigating the issue of overconfidence in winning scores while simultaneously considering various categories of uncertainty. seamlessly integrate with different Deep Neural Networks. Extensive experiments with ablation settings are conducted on four real-world datasets, resulting in consistently competitive improvements. Our second topic focuses on uncertainty estimation on few-shot text classification (UEFTC), which has few or even only one available support sample for each class. UEFTC represents an underexplored research domain where, due to limited data samples, a UEFTC model predicts an uncertainty score to assess the likelihood of classification errors. However, traditional uncertainty estimation models in text classification are ill-suited for UEFTC since they demand extensive training data, while UEFTC operates in a few-shot scenario, typically providing just a few support samples, or even just one, per class. To tackle this challenge, we introduce Contrastive Learning from Uncertainty Relations (CLUR) as a solution tailored for UEFTC. CLUR exhibits the unique capability to be effectively trained with only one support sample per class, aided by pseudo uncertainty scores. A distinguishing feature of CLUR is its autonomous learning of these pseudo uncertainty scores, in contrast to previous approaches that relied on manual specification. Our investigation of CLUR encompasses four model structures, allowing us to evaluate the performance of three commonly employed contrastive learning components in the context of UEFTC. Our findings highlight the effectiveness of two of these components. Our third topic focuses on uncertainty estimation on sequential labeling. Sequential labeling involves the task of assigning labels to individual tokens in a sequence, exemplified by Named Entity Recognition (NER). Despite significant advancements in enhancing NER performance in prior research, the realm of uncertainty estimation for NER (UE-NER) remains relatively uncharted but is of paramount importance. This topic focuses on UE-NER, seeking to gauge uncertainty scores for NER predictions. Previous models for uncertainty estimation often overlook two distinctive attributes of NER: the interrelation among entities (where the learning of one entity's embedding depends on others) and the challenges posed by incorrect span predictions in entity extraction. To address these issues, we introduce the Sequential Labeling Posterior Network (SLPN), designed to estimate uncertainty scores for the extracted entities while considering uncertainty propagation from other tokens. Additionally, we have devised an evaluation methodology tailored to the specific nuances of wrong-span cases. Our fourth topic focuses on an overlooked question that persists regarding the evaluation reliability of uncertainty estimation in text summarization (UE-TS). Text summarization, a key task in natural language generation (NLG), holds significant importance, particularly in domains where inaccuracies can have serious consequences, such as healthcare. UE-TS has garnered attention due to the potential risks associated with erroneous summaries. However, the reliability of evaluating UE-TS methods raises concerns, stemming from the interdependence between uncertainty model metrics and the wide array of NLG metrics. To address these concerns, we introduce a comprehensive UE-TS benchmark incorporating twenty-six NLG metrics across four dimensions. This benchmark evaluates the uncertainty estimation capabilities of two large language models and one pre-trained language model across two datasets. Additionally, it assesses the effectiveness of fourteen common uncertainty estimation methods. Our study underscores the necessity of utilizing diverse, uncorrelated NLG metrics and uncertainty estimation techniques for a robust evaluation of UE-TS methods. / Doctor of Philosophy / Text is integral to our daily activities, appearing in various forms such as social media posts, news articles, books, and reports. We rely on text for communication, information dissemination, and decision-making. Given its ubiquity, the ability to process and understand text through Natural Language Processing (NLP) has become increasingly important. NLP technology enables us to perform tasks like text classification, which involves categorizing text into predefined labels, and named entity recognition (NER), which identifies specific entities such as names, dates, and locations within text. Additionally, NLP facilitates generating coherent and contextually appropriate responses in conversational agents, enhancing human-computer interaction. However, the reliability of NLP models is crucial, especially in sensitive applications like medical diagnoses, where errors can have severe consequences. This dissertation focuses on uncertainty estimation in NLP, a less explored but essential area. Uncertainty estimation helps evaluate the confidence of NLP model predictions. We propose new methods tailored to various NLP tasks, acknowledging their unique needs. NLP tasks are divided into natural language understanding (NLU) and natural language generation (NLG). Within NLU, we look at tasks from two perspectives: a global view (e.g., document-level text classification) and a local view (e.g., sentence-level inference and token-level entity recognition). Our research spans text classification, named entity recognition (NER), and text summarization, with a special focus on few-shot text classification due to its recent prominence. For text classification, we introduce the MSD model, which includes three components to enhance uncertainty score accuracy and address overconfidence issues. This model integrates seamlessly with different neural networks and shows consistent improvements in experiments. For few-shot text classification, we develop Contrastive Learning from Uncertainty Relations (CLUR), designed to work effectively with minimal support samples per class. CLUR autonomously learns pseudo uncertainty scores, demonstrating effectiveness with various contrastive learning components. In NER, we address the unique challenges of entity interrelation and span prediction errors. We propose the Sequential Labeling Posterior Network (SLPN) to estimate uncertainty scores while considering uncertainty propagation from other tokens. For text summarization, we create a benchmark with tens of metrics to evaluate uncertainty estimation methods across two datasets. This benchmark helps assess the reliability of these methods, highlighting the need for diverse, uncorrelated metrics. Overall, our work advances the understanding and implementation of uncertainty estimation in NLP, providing more reliable and accurate predictions across different tasks.

L'intégration de la production des énergies renouvelables intermittentes dans le mix énergétique est aujourd'hui limitée à un seuil de 30 % de la puissance totale produite. Cette mesure vise à assurer la sécurité de l'alimentation électrique des réseaux insulaires en France. La levée de ce verrou technique ne pourra se faire qu'en apportant des solutions au caractère intermittent des sources d'énergies éolienne et photovoltaïque. Les difficultés énergétiques auxquelles sont confrontés aujourd'hui les milieux insulaires préfigurent celles que rencontreront la planète à plus ou moins long terme. Ces territoires sont des laboratoires uniques pour éprouver les nouvelles technologies de stockage, de gestion et de prévision de l'énergie. La contribution de ce travail de thèse se focalise sur la prévision du rayonnement solaire global à différents horizons de temps car la puissance photovoltaïque produite découle directement de l'intensité du rayonnement solaire global. Dans un premier temps, l'étude bibliographique a permis de classer les modèles de prévision numériques et les modèles de prévision statistiques en fonction de la résolution spatiale et temporelle. Par ailleurs, elle montre que les meilleurs performances sont obtenues avec les modèles hybrides. Dans un deuxième temps, un modèle de prévisions à court terme (J+1) est proposé avec le modèle Weather Research and Forecasting (WRF) et un réseau de neurone bayésien. L'hybridation de ces deux méthodes améliore les performances de prévisions à J+1. Dans un troisième temps, un modèle de prévision à très court terme (t+h) est proposé avec le modèle hybride de Kalman. Cette méthode produit d'une part une prévision énergétique et d'autre part une prévision multi-horizon. La comparaison de la performance de ces modèles avec la méthode de référence dite de persistance montre une amélioration de la qualité de la prévision. Enfin, la combinaison du filtre de Kalman avec le modèle numérique WRF permet une mise en œuvre opérationnelle de la prévision. / The integration of intermittent renewable energy in the energy mix is currently limited to a threshold of 30% of the total power being produced. This restriction aims at ensuring the safety of the power input. The elimination of this technical obstacle will be possible with solutions to energy intermittence of wind and solar energy. The energy issues which islands are facing today prefigure global problems in a more or less long term. These territories constitute unique laboratories for testing new technologies of storage, management and forecasting of energy. The contribution of this thesis focuses on the forecasting of global horizontal irradiance at different time horizons. Indeed, the generated PV power stems directly from the intensity of the global horizontal irradiance. First, the review of solar irradiance forecasting methods allows to classify numerical weather models and statistical forecasting methods depending on spatial and temporal resolution. Moreover, it shows that best performance is obtained with hybrid models. Second, a short-term forecast model (day ahead forecast) is developed with the Weather Research and Forecasting model (WRF) and a Bayesian neural network. The hybridization of these methods improves the day ahead forecast performance. Third, a model for forecasting the very short term is developped with the Kalman hybrid model. This method offers on the one hand an energy forecasting and on the other hand a multi-horizon forecast. Comparing the performance of the aforesaid with the reference method, namely the persistence method, shows an improvement of the quality of the forecasts. Combining the Kalman filter with the WRF numerical model allows an operational implementation of the forecast.

Remote Electrical Tilt (RET) is a method for configuring antenna downtilt in base stations to optimize mobile network performance. Reinforcement Learning (RL) is an approach to automating the process by letting an agent learn an optimal control strategy and adapt to the dynamic environment. Applying RL in real world comes with challenges, for the RET problem there are performance requirements and partial observability of the system through exogenous factors inducing noise in observations. This thesis proposes a solution method through modeling the problem by a Partially Observable Markov Decision Process (POMDP). The set of hidden states are modeled as a high- level representation of situations requiring one of the possible actions uptilt, downtilt, no change. From this model, a Bayesian Neural Network (BNN) is trained to predict an observation model, relating observed Key Performance Indicators (KPIs) to the hidden states. The observation model is used for estimating belief state probabilities of each hidden state, from which decision of control action is made through a restrictive threshold policy. Experiments comparing the method to a baseline Deep Q- network (DQN) agent shows the method able to reach the same average performance increase as the baseline while outperforming the baseline in two metrics important for robust and safe control behaviour, the worst- case minimum reward increase and the average reward increase per number of tilt actions. / Fjärrstyrning av Elektrisk Lutning (FEL) är en metod för att reglera lutningen av antenner i basstationer för att optimera presentandan i ett mobilnätverk. Förstärkande Inlärning (FI) används som metod för att automatisera processen genom att låta en agent lära sig en optimal strategi för reglering och anpassa sig till den dynamiska miljön. Att tillämpa FI i ett verkligt scenario innebär utmaningar, för FEL specifikt finns det krav på en viss nivå av prestanda samt endast en delvis observerbarhet av systemet på grund av externa faktorer som orsakar brus i observationerna. I detta arbete föreslås en metod för att hantera detta genom att modellera problemet som en Delvis Observerbar Markovprocess (DOM). De dolda tillstånden modelleras för att representera situationer där var och en av de möjliga aktionerna behövs, det vill säga att luta antennen upp, ner eller inte ändra på lutningen. Utifrån denna modellering så tränas ett Bayesiskt Neuralt Nätverk (BNN) för att estimera en observationsmodel som kopplar observerade nyckeltal till de dolda tillstånden. Denna observationsmodel används för att estimera sannolikheten att vardera dolt tillstånd är det rätta. Utifrån dessa sannolikheter så görs valet av aktion genom ett tröskelvärde på sannolikheterna. Genom experiment som jämför metoden med en standardimplementering av en agent baserad på ett Djupt Qnätverk (DQN) visas att metoden har samma prestation när det kommer till en medelnivå på prestandaökning i nätverket. Metoden överträffar dock standardmetoden i två andra mätvärden som är viktiga ur aspekten säker och robust reglering, minimumvärdet på prestandaökningen samt medelökningen av prestandan per antal up- och nerlutningar som används.