Intégrée dans la chaîne décisionnelle de la Supply Chain à un niveau tactique, la Planification de Production est un process clé qui permet de répondre au mieux aux besoins selon les ressources de l’entreprise. Un des défis du domaine est la gestion des incertitudes prévisionnelles, ayant des conséquences importantes sur des indicateurs clés comme le taux de service ou les coûts. Pour y faire face, des méthodes améliorant la flexibilité des processus sont mais en place, comme le contexte de travail en Plan Glissant. Cependant, en actualisant fréquemment les données, la stabilité du système se retrouve dégradée. Ainsi, malgré les gains issus de la gestion des incertitudes, ce cadre crée une complexité dynamique à gérer. Ce travail traite de cette complexité issue de l’actualisation des prévisions pour la planification de production en plan glissant. Plus particulièrement, la question traitée ici concerne l’optimisation du plan de production, en considérant u n système mono-produit monoétage. Une modélisation mathématique générique est tout d’abord développée pour construire un modèle d’optimisation théorique du problème. Ensuite, une procédure de résolution optimale est développée en utilisant le cadre d’optimisation dynamique stochastique. Ce modèle est appliquée à des cas concrets pour lesquels l’optimalité des solutions calculées est prouvée analytiquement grâce à un raisonnement inductif basé sur des séquences de calcul d’espérances mathématiques. Des analyses numériques finalement conduites mettent en exergue les performances de la méthode développée, ses limites, et sa sensibilité vis-à-vis de l’environnement industriel. / Production Planning, as part of tactical operations integrated into the Supply Chain process, is a key procedure allowing decisioners to balance demand and production resources. One of its most challenging issues is to handle uncertainties, especially the ones coming from the Forecasted Demand. In order to manage indicators at stake, such as service level and costs, best practices increasing flexibility in the process are implemented, as Rolling-Plan Framework. However, it creates instability since the updates procedures make the data set on change constantly. Consequently, although the gain in terms of flexibility is non-negligible for the uncertainties management, it generates on the other hand dynamics complexity. We study in this work how to deal this dynamics complexity generated by updates of the Forecasted Demand made in a Rolling-Plan Framework of a Production Planning Process. In particular, the question to which it answers is how to optimize the Production Plan in such a context. This issue is tackled considering a single item single level production system. A general mathematical model in the context of our study is built to be exploitable for analytical optimization. A theoretical optimization framework is designed, and a specific solutions computation framework using stochastic dynamic programming is developed. We apply it in some precise study cases in order to compute optimal solutions and get some valuable analytical results thanks to a dynamic computation process. The optimality of the solutions is proven through an inductive reasoning based on expectations computation. Solutions are finally implemented and calculated numerically with simulations in some particular numerical examples. Analyses and sensitivity studies are performed, highlighting the performances of our optimization method.

Reservoir Computing (RC) is an emerging machine learning paradigm where a fixed kernel, built from a randomly connected "reservoir" with sufficiently rich dynamics, is capable of expanding the problem space in a non-linear fashion to a higher dimensional feature space. These features can then be interpreted by a linear readout layer that is trained by a gradient descent method. In comparison to traditional neural networks, only the output layer needs to be trained, which leads to a significant computational advantage. In addition, the short term memory of the reservoir dynamics has the ability to transform a complex temporal input state space to a simple non-temporal representation. Adaptive real-time systems are multi-stage decision problems that can be used to train an agent to achieve a preset goal by performing an optimal action at each timestep. In such problems, the agent learns through continuous interactions with its environment. Conventional techniques to solving such problems become computationally expensive or may not converge if the state-space being considered is large, partially observable, or if short term memory is required in optimal decision making. The objective of this thesis is to use reservoir computers to solve such goal-driven tasks, where no error signal can be readily calculated to apply gradient descent methodologies. To address this challenge, we propose a novel reinforcement learning approach in combination with reservoir computers built from simple Boolean components. Such reservoirs are of interest because they have the potential to be fabricated by self-assembly techniques. We evaluate the performance of our approach in both Markovian and non-Markovian environments. We compare the performance of an agent trained through traditional Q-Learning. We find that the reservoir-based agent performs successfully in these problem contexts and even performs marginally better than Q-Learning agents in certain cases. Our proposed approach allows to retain the advantage of traditional parameterized dynamic systems in successfully modeling embedded state-space representations while eliminating the complexity involved in training traditional neural networks. To the best of our knowledge, our method of training a reservoir readout layer through an on-policy boot-strapping approach is unique in the field of random Boolean network reservoirs.

Thesis (M.S.)--Massachusetts Institute of Technology, Dept. of Civil Engineering, 1981. / MICROFICHE COPY AVAILABLE IN ARCHIVES AND ENGINEERING. / Bibliography: leaves 135-137. / by Katherine Oven Thompson. / M.S.

京都大学 / 0048 / 新制・課程博士 / 博士(情報学) / 甲第18701号 / 情博第551号 / 新制||情||97(附属図書館) / 31634 / 京都大学大学院情報学研究科知能情報学専攻 / (主査)教授 阿久津 達也, 教授 山本 章博, 教授 岡部 寿男 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

No description available.

Cache replacement policies have applications that are nearly ubiquitous in technology. Among these is an interesting subset which occurs when referentially transparent functions are memoized, eg. in compilers, in dynamic programming, and other software caches. In many applications the least recently used (LRU) approach likely preserves items most needed by memoized function calls. However, despite its popularity LRU is expensive to implement, which has caused a spate of research initiatives aimed at approximating its cache miss performance in exchange for faster and more memory efficient implementations. We present a novel caching algorithm, Far From Recently Used (FFRU), which offers a simple, but highly configurable mechanism for providing lower bounds on the usage recency of items evicted from the cache. This algorithm preserves the constant time amortized cost of insertions and updates and minimizes the memory overhead needed to administer the eviction guarantees. We study the cache miss performance of several memoized optimization problems which vary in the number of subproblems generated and the access patterns exhibited by their recursive calls. We study their cache miss performance using LRU cache replacement, then show the performance of FFRU in these same problem scenarios. We show that for equivalent minimum eviction age guarantees, FFRU incurs fewer cache misses than LRU, and does so using less memory. We also present some variations of the algorithms studied (Fibonacci, KMP, LCS, and Euclidean TSP) which exploit the characteristics of the cache replacement algorithms being employed, further resulting in improved cache miss performance. We present a novel implementation of a well known approximation algorithm for the Euclidean Traveling Salesman Problem due to Sanjeev Arora. Our implementation of this algorithm outperforms the currently known implementations of the same. It has long remained an open question whether or not algorithms relying on geometric divisions of space can be implemented into practical tools, and our powerful implementation of Arora's algorithm establishes a new benchmark in that arena. / Computer and Information Science

[pt] A gestão de ativos é uma abordagem crucial para o desempenho das organizações uma vez que buscam alinhar aspectos técnicos de engenharia com conceitos financeiros para otimizar o ciclo de vida de uma máquina. O Problema de Substituição de Equipamentos é uma das questões tratadas dentro dos estudos de gestão de ativos que visa decidir a melhor opção entre manter ou substituir o equipamento em um determinado intervalo de tempo. Uma das metodologias que vêm sendo utilizadas na literatura para solucionar este problema é a Programação Dinâmica, que se baseia em encontrar soluções parciais em uma série de estágios do problema até alcançar a ótima global. Este trabalho teve como objetivo determinar uma curva de substituição para um conjunto de locomotivas de uma empresa do setor ferroviário, considerando um limite de idade para poderem circular e seus históricos de receitas e custos ao longo dos anos. Os resultados alcançados permitiram que a empresa conhecesse a melhor forma de otimizar seu capital, levando em consideração os impactos financeiros caso opte por antecipar ou postergar o momento ótimo para substituição dos ativos. / [en] Asset management is a crucial approach for the performance of organizations as they seek to align technical aspects of engineering with financial concepts to optimize the life cycle of a machine. The Equipment Replacement Problem is one of the issues addressed within asset management studies that aims to decide the best option between maintaining or replacing equipment in a given time interval. One of the methodologies that have been used in the literature to solve this problem is Dynamic Programming, which is based on finding partial solutions in a series of stages of the problem until reaching the global optimum. This work aimed to determine a substitution curve for a set of locomotives of a company in the railway sector, considering an age limit for them to circulate and their revenue and cost history over the years. The results achieved allowed the company to know the best way to optimize its capital, taking into account the financial impacts if it chooses to anticipate or postpone the optimal moment for the replacement of assets.

No description available.

No description available.

An embedded model predictive controller for velocity control of trucks is developed and tested. By using a simple model of a heavy duty vehicle and knowledge about the slope of the road ahead, the fuel consumption while traveling near a set speed is diminished by almost 1% on an example road compared to a rule based speed control system. The problem is formulated as a look-ahead optimization problem were fuel consumption and total trip time have to be minimized. To find the optimal solution dynamic programming is used, and the whole code is designed to run on a Scania gearbox ECU in parallel with all the current software. Simulations were executed in a Simulink environment, and two test rides were performed on the E4 motorway. / En algoritm för hastighetsstyrning baserad på modell-prediktiv reglering har utvecklats och testats på befintlig styrsystem i ett Scania lastbil. Genom att använda en enkel modell av fordonet och kunskap om lutningen på vägen framför den kunde man sänka bränsleförbrukningen med nästan 1% i vissa sträckor, jämfört med en regelbaserad farthållare. Problemet är formulerat som en optimerings-problem där bränsleförbrukning och total restid måste minimeras. För att hitta den optimala lösningen användes dynamisk programmering och hela koden är skriven så att den kan exekveras på en Scania styrenehet. Koden är kan köras parallellt med den mjukvara som är installerad på styrenheten. Simuleringar utfördes i en miljö utvecklad i Simulink. Två test-körningar på E4 motorvägen utfördes.