Mimicking natural evolution to solve hard optimization problems has played an important role in the artificial intelligence arena. Such techniques are broadly classified as Evolutionary Algorithms (EAs) and have been investigated for around four decades during which important contributions and advances have been made. One main evolutionary technique which has been widely investigated is the Genetic Algorithm (GA). GAs are stochastic search techniques that follow the Darwinian principle of evolution. Their application in the solution of hard optimization problems has been very successful. Indeed multi-dimensional problems presenting difficult search spaces with characteristics such as multi-modality, epistasis, non regularity, deceptiveness, etc., have all been effectively tackled by GAs. In this research, a competitive form of GAs known as fine or cellular GAs (cGAs) are investigated, because of their suitability for System on Chip (SoC) implementation when tackling real-time problems. Cellular GAs have also attracted the attention of researchers due to their high performance, ease of implementation and massive parallelism. In addition, cGAs inherently possess a number of structural configuration parameters which make them capable of sustaining diversity during evolution and therefore of promoting an adequate balance between exploitative and explorative stages of the search. The fast technological development of Integrated Circuits (ICs) has allowed a considerable increase in compactness and therefore in density. As a result, it is nowadays possible to have millions of gates and transistor based circuits in very small silicon areas. Operational complexity has also significantly increased and consequently other setbacks have emerged, such as the presence of faults that commonly appear in the form of single or multiple bit flips. Tough environmental or time dependent operating conditions can trigger faults in registers and memory allocations due to induced radiation, electron migration and dielectric breakdown. These kinds of faults are known as Single Event Effects (SEEs). Research has shown that an effective way of dealing with SEEs consists of a combination of hardware and software mitigation techniques to overcome faulty scenarios. Permanent faults known as Single Hard Errors (SHEs) and temporary faults known as Single Event Upsets (SEUs) are common SEEs. This thesis aims to investigate the inherent abilities of cellular GAs to deal with SHEs and SEUs at algorithmic level. A hard real-time application is targeted: calculating the attitude parameters for navigation in vehicles using Global Positioning System (GPS) technology. Faulty critical data, which can cause a system’s functionality to fail, are evaluated. The proposed mitigation techniques show cGAs ability to deal with up to 40% stuck at zero and 30% stuck at one faults in chromosomes bits and fitness score cells. Due to the non-deterministic nature of GAs, dynamic on-the-fly algorithmic and parametric configuration has also attracted the attention of researchers. In this respect, the structural properties of cellular GAs provide a valuable attribute to influence their selection pressure. This helps to maintain an adequate exploitation-exploration tradeoff, either from a pure topological perspective or through genetic operations that also make use of structural characteristics in cGAs. These properties, unique to cGAs, are further investigated in this thesis through a set of middle to high difficulty benchmark problems. Experimental results show that the proposed dynamic techniques enhance the overall performance of cGAs in most benchmark problems. Finally, being structurally attached, the dimensionality of cellular GAs is another line of investigation. 1D and 2D structures have normally been used to test cGAs at algorithm and implementation levels. Although 3D-cGAs are an immediate extension, not enough attention has been paid to them, and so a comparative study on the dimensionality of cGAs is carried out. Having shorter radii, 3D-cGAs present a faster dissemination of solutions and have denser neighbourhoods. Empirical results reported in this thesis show that 3D-cGAs achieve better efficiency when solving multi-modal and epistatic problems. In future, the performance improvements of 3D-cGAs will merge with the latest benefits that 3D integration technology has demonstrated, such as reductions in routing length, in interconnection delays and in power consumption.

The objective of this study is to develop a calibration and verification tool for the QUAL2E Model by using Genetic Algorithms. In the developed optimization model, an objective function that is formulated on the basis of the sum-of-least squares approach aiming at minimizing the difference between the observed and simulated quantities was used. In order to perform simultaneous calibration and verification, verification of the calibrated results was treated as a constraint and inserted into the objective function as a penalty function. The performance of the optimization model was tested for different observation data qualities represented by the synthetic perfect and biased data sets. Although it was not possible to obtain the exact values of the kinetic coefficients for any of the tests performed, the coefficient estimates were successful in reflecting the water quality variable profiles in the river. The results of the tests showed that the performance of the optimization model is generally sensitive to the error in the observed data sets, to the number and location of sampling points, and to the objective function formulation. For the problems that involve multiple water quality variables, a weighting approach used in the objective function formulation resulted in better performances. The optimization model was also applied for a case study. For the same input data, calibration obtained with the genetic algorithm optimization &ndash / simulation was better compared to the trial-and-error approach.

Dans les systèmes mécaniques ou mécatroniques, les actionneurs sont les composants utilisés pour convertir l’énergie d’entrée, généralement l’énergie électrique, en tâche mécanique telles que le mouvement, la force ou une combinaison des deux. Actionneur analogique et actionneur numérique sont les deux types d’actionneurs les plus communs. Les actionneurs numériques possèdent les avantages du contrôle en boucle ouverte, faible consommation d’énergie par rapport aux actionneurs analogiques. Cependant, les actionneurs numériques présentent deux inconvénients majeurs. Les erreurs de fabrication de ces actionneurs doivent être contrôlées précisément parce que, contrairement à des actionneurs analogiques, une erreur de fabrication ne peut pas être compensée par la loi de commande. Un autre inconvénient est leur capacité à réaliser les tâches continues en raison de leur corse discrète. Un assemblage de plusieurs actionneurs numériques peut néanmoins réaliser des tâches multiples discrètes. Cette thèse porte sur la caractérisation et l’optimisation d’une conception expérimentale actionneurs tableau numériques pour l’application planaire de transport. Le premier objectif principal de la présente thèse est axé sur la caractérisation de l’ensemble des actionneurs existants et aussi une application planaire de transport sur la base du tableau des actionneurs. A cette fin, une modélisation de la matrice des actionneurs essais expérimentaux ont été effectués afin de déterminer l’influence de certains paramètres sur le comportement des actionneurs de tableau. Le deuxième objectif est de concevoir une nouvelle version du tableau actionneurs sur la base de l’expérience du premier prototype. Une optimisation de la conception a ensuite été réalisée en utilisant des techniques d’algorithmes génétiques tout en tenant compte de plusieurs critères. / In mechanical or mechatronical systems, actuators are the components used to convert input energy, generally electrical energy, into mechanical tasks such as motion, force or a combination of both. Analogical actuator and digital actuator are two common types of actuators. Digital actuators have the advantages of open-loop control, low energy consumption and etc compared to analogical actuators. However, digital actuators present two main drawbacks. The manufacturing errors of these actuators have to be precisely controlled because, unlike to analogical actuators, a manufacturing error cannot be compensated using the control law. Another drawback is their inability to realize continuous tasks because of their discrete stroke. An assembly of several digital actuators can nevertheless realize multi-discrete tasks. This thesis focuses on the experimental characterization and optimization design of a digital actuators array for planar conveyance application. The firs main objective of the present thesis is focused on the characterization of the existing actuators array and also a planar conveyance application based on the actuators array. For that purpose, a modeling of the actuators array and experimental test has been carried out in order to determine the influence of some parameters on the actuators array behavior. The second objective is to design a new version of the actuators array based on the experience of the first prototype. An optimization of the design has then been realized using genetic algorithm techniques while considering several criteria.

This thesis is focused on the design and optimization of automated trading system, which will be traded in FOREX. The aim is to create a business strategy that is relatively safe, stable and profitable. Optimization and testing on historical data are a prerequisite for the deployment into real trading.