Streamed data classification problems (SDCPs) require classifiers with the ability to learn and to adjust to the underlying relationships in data streams, in real-time. This requirement poses a challenge to classifiers, because the learning task is no longer just to find the optimal decision boundaries, but also to track changes in the decision boundaries as new training data is received. The challenge is due to concept drift, i.e. the changing of decision boundaries over time. Changes include disappearing, appearing, or shifting decision boundaries. This thesis proposes an online learning approach for feed forward neural networks (FFNNs) that meets the requirements of SDCPs. The approach uses regularisation to optimise the architecture via the weights, and quantum particle swarm optimisation (QPSO) to dynamically adjust the weights. The learning approach is applied to a FFNN, which uses rectified linear activation functions, to form a novel SDCP classifier. The classifier is empirically investigated on several SDCPs. Both weight decay (WD) and weight elimination (WE) are investigated as regularisers. Empirical results show that using QPSO with no regularisation, causes the classifier to completely saturate. However, using QPSO with regularisation enables the classifier to dynamically adapt both its implicit architecture and weights as decision boundaries change. Furthermore, the results favour WE over WD as a regulariser for QPSO. / Dissertation (MSc)--University of Pretoria, 2020. / National Research Foundation (NRF) / Computer Science / MSc / Unrestricted

In the process industries, cooling capacity is an important enabler for the facility to manufacture on specification product. The cooling water network is an important part of the over-all cooling system of the facility. In this research a cooling water circuit consisting of 3 cooling towers in parallel, 2 cooling water pumps in parallel, and 11 heat exchangers in parallel, is modelled. The model developed is based on first principles and captures the dynamic, non-linear, interactive nature of the plant. The modelled plant is further complicated by continuous, as well as discrete process variables, giving the model a hybrid nature. Energy consumption is included in the model as it is a very important parameter for plant operation. The model is fitted to real industry data by using a particle swarm optimisation approach. The model is suitable to be used for optimisation and control purposes. Cooling water networks are often not instrumented and actuated, nor controlled or optimised. Significant process benefits can be achieved by better process end-user temperature control, and direct monetary benefits can be obtained from electric power minimisation. A Hybrid Non-Linear Model Predictive Control strategy is developed for these control objectives, and simulated on the developed first principles dynamic model. Continuous and hybrid control cases are developed, and tested on process scenarios that reflect conditions seen in a real plant. Various alternative techniques are evaluated in order to solve the Hybrid Non-Linear Control problem. Gradient descent with momentum is chosen and configured to be used to solve the continuous control problem. For the discrete control problem a graph traversal algorithm is developed and joined to the continuous control algorithm to form a Hybrid Non-Linear Model Predictive controller. The potential monetary benefits that can be obtained by the plant owner through implementing the designed control strategy, are estimated. A powerful computation platform is designed for the plant model and controller simulations. / Thesis (PhD)--University of Pretoria, 2019. / Electrical, Electronic and Computer Engineering / PhD / Unrestricted

In this thesis, exergy based SI engine model optimisation (EBSIEMO) is studied and evaluated. A four-stroke bi-fuel spark ignition (SI) engine is modelled for optimisation of engine performance based upon exergy analysis. An artificial neural network (ANN) is used as an emulator to speed up the optimisation processes. Constrained particle swarm optimisation (CPSO) is employed to identify parameters such as equivalence ratio and ignition time for optimising of the engine performance, based upon maximising ¿total availability¿. In the optimisation process, the engine exhaust gases standard emission were applied including brake specific CO (BSCO) and brake specific NOx (BSNOx) as the constraints. The engine model is developed in a two-zone model, while considering the chemical synthesis of fuel, including 10 chemical species. A computer code is developed in MATLAB software to solve the equations for the prediction of temperature and pressure of the mixture in each stage (compression stroke, combustion process and expansion stroke). In addition, Intake and exhaust processes are calculated using an approximation method. This model has the ability to simulate turbulent combustion and compared to computational fluid dynamic (CFD) models it is computationally faster and efficient. The selective outputs are cylinder temperature and pressure, heat transfer, brake work, brake thermal and volumetric efficiency, brake torque, brake power (BP), brake specific fuel consumption (BSFC), brake mean effective pressure (BMEP), concentration of CO2, brake specific CO (BSCO) and brake specific NOx (BSNOx). In this model, the effect of engine speed, equivalence ratio and ignition time on performance parameters using gasoline and CNG fuels are analysed. In addition, the model is validated by experimental data using the results obtained from bi-fuel engine tests. Therefore, this engine model was capable to predict, analyse and useful for optimisation of the engine performance parameters. The exergy based four-stroke bi-fuel (CNG and gasoline) spark ignition (SI) engine model (EBSIEM) here is used for analysis of bi-fuel SI engines. Since, the first law of thermodynamic (the FLT), alone is not able to afford an appropriate comprehension into engine operations. Therefore, this thesis concentrates on the SI engine operation investigation using the developed engine model by the second law of thermodynamic (the SLT) or exergy analysis outlook (exergy based SI engine model (EBSIEM)) In this thesis, an efficient approach is presented for the prediction of total availability, brake specific CO (BSCO), brake specific NOx (BSNOx) and brake torque for bi-fuel engine (CNG and gasoline) using an artificial neural network (ANN) model based on exergy based SI engine (EBSIEM) (ANN-EBSIEM) as an emulator to speed up the optimisation processes. In the other words, the use of a well trained an ANN is ordinarily much faster than mathematical models or conventional simulation programs for prediction. The constrained particle swarm optimisation (CPSO)-EBSIEM (EBSIEMO) was capable of optimising the model parameters for the engine performance. The optimisation results based upon availability analysis (the SLT) due to analysing availability terms, specifically availability destruction (that measured engine irreversibilties) are more regarded with higher priority compared to the FLT analysis. In this thesis, exergy based SI engine model optimisation (EBSIEMO) is studied and evaluated. A four-stroke bi-fuel spark ignition (SI) engine is modelled for optimisation of engine performance based upon exergy analysis. An artificial neural network (ANN) is used as an emulator to speed up the optimisation processes. Constrained particle swarm optimisation (CPSO) is employed to identify parameters such as equivalence ratio and ignition time for optimising of the engine performance, based upon maximising ¿total availability¿. In the optimisation process, the engine exhaust gases standard emission were applied including brake specific CO (BSCO) and brake specific NOx (BSNOx) as the constraints. The engine model is developed in a two-zone model, while considering the chemical synthesis of fuel, including 10 chemical species. A computer code is developed in MATLAB software to solve the equations for the prediction of temperature and pressure of the mixture in each stage (compression stroke, combustion process and expansion stroke). In addition, Intake and exhaust processes are calculated using an approximation method. This model has the ability to simulate turbulent combustion and compared to computational fluid dynamic (CFD) models it is computationally faster and efficient. The selective outputs are cylinder temperature and pressure, heat transfer, brake work, brake thermal and volumetric efficiency, brake torque, brake power (BP), brake specific fuel consumption (BSFC), brake mean effective pressure (BMEP), concentration of CO2, brake specific CO (BSCO) and brake specific NOx (BSNOx). In this model, the effect of engine speed, equivalence ratio and ignition time on performance parameters using gasoline and CNG fuels are analysed. In addition, the model is validated by experimental data using the results obtained from bi-fuel engine tests. Therefore, this engine model was capable to predict, analyse and useful for optimisation of the engine performance parameters. The exergy based four-stroke bi-fuel (CNG and gasoline) spark ignition (SI) engine model (EBSIEM) here is used for analysis of bi-fuel SI engines. Since, the first law of thermodynamic (the FLT), alone is not able to afford an appropriate comprehension into engine operations. Therefore, this thesis concentrates on the SI engine operation investigation using the developed engine model by the second law of thermodynamic (the SLT) or exergy analysis outlook (exergy based SI engine model (EBSIEM)) In this thesis, an efficient approach is presented for the prediction of total availability, brake specific CO (BSCO), brake specific NOx (BSNOx) and brake torque for bi-fuel engine (CNG and gasoline) using an artificial neural network (ANN) model based on exergy based SI engine (EBSIEM) (ANN-EBSIEM) as an emulator to speed up the optimisation processes. In the other words, the use of a well trained an ANN is ordinarily much faster than mathematical models or conventional simulation programs for prediction. The constrained particle swarm optimisation (CPSO)-EBSIEM (EBSIEMO) was capable of optimising the model parameters for the engine performance. The optimisation results based upon availability analysis (the SLT) due to analysing availability terms, specifically availability destruction (that measured engine irreversibilties) are more regarded with higher priority compared to the FLT analysis.

The objectives of this work were to investigate, design and implement beam steering antenna arrays for mobile and wireless applications using the genetic algorithm (GA) and particle swarm optimisation (PSO) techniques as optimisation design tools. Several antenna designs were implemented and tested: initially, a printed dipole antenna integrated with a duplex RF switch used for mobile base station antenna beam steering was investigated. A coplanar waveguide (CPW) to coplanar strip (CPS) transition was adopted to feed the printed dipole. A novel RF switch circuit, used to control the RF signal fed to the dipole antenna and placed directly before it, was proposed. The measured performance of the RF switch was tested and the results confirmed its viability. Then two hybrid coupled PIN diode phase shifters, using Branchline and Rat-Race ring coupler structures, were designed and tested. The generation of four distinct phase shifts was implemented and studied. The variations of the scattering parameters were found to be realistic, with an acceptable ±2 phase shift tolerance. Next, antenna beam steering was achieved by implementing RF switches with ON or OFF mode functions to excite the radiating elements of the antenna array. The switching control process was implemented using a genetic algorithm (GA) method, subject to scalar and binary genes. Anti-phase feeding of radiating elements was also investigated. A ring antenna array with reflectors was modelled and analysed. An antenna of this type for mobile base stations was designed and simulation results are presented. Following this, a novel concept for simple beam steering using a uniform antenna array operated at 2.4 GHz was designed using GA. The antenna is fed by a single RF input source and the steering elements are reactively tuned by varactor diodes in series with small inductors. The beam-control procedure was derived through the use of a genetic algorithm based on adjusting the required reactance values to obtain the optimum solution as indicated by the cost function. The GA was also initially used as an optimisation tool to derive the antenna design from its specification. Finally, reactive loading and time modulated switching techniques are applied to steer the beam of a circular uniformly spaced antenna array having a source element at its centre. Genetic algorithm (GA) and particle swarm optimisation (PSO) processes calculate the optimal values of reactances loading the parasitic elements, for which the gain can be optimised in a desired direction. For time modulated switching, GA and PSO also determine the optimal on and off times of the parasitic elements for which the difference in currents induced optimises the gain and steering of the beam in a desired direction. These methods were demonstrated by investigating a vertically polarised antenna configuration. A prototype antenna was constructed and experimental results compared with the simulations. Results showed that near optimal solutions for gain optimisation, sidelobe level reduction and beam steering are achievable by utilising these methods. In addition, a simple switching process is employed to steer the beam of a horizontally polarised circular antenna array. A time modulated switching process is applied through Genetic Algorithm optimisation. Several model examples illustrate the radiation beams and the switching time process of each element in the array.

The developments of advanced non-linear control strategies have attracted a considerable research interests over the past decades especially in process control. Rather than an absolute reliance on mathematical models of process plants which often brings discrepancies especially owing to design errors and equipment degradation, non-linear models are however required because they provide improved prediction capabilities but they are very difficult to derive. In addition, the derivation of the global optimal solution gets more difficult especially when multivariable and non-linear systems are involved. Hence, this research investigates soft computing techniques for the implementation of a novel real time constrained non-linear model predictive controller (NMPC). The time-frequency localisation characteristics of wavelet neural network (WNN) were utilised for the non-linear models design using system identification approach from experimental data and improve upon the conventional artificial neural network (ANN) which is prone to low convergence rate and the difficulties in locating the global minimum point during training process. Salient features of particle swarm optimisation and a genetic algorithm (GA) were combined to optimise the network weights. Real time optimisation occurring at every sampling instant is achieved using a GA to deliver results both in simulations and real time implementation on coupled tank systems with further extension to a complex quadruple tank process in simulations. The results show the superiority of the novel WNN-NMPC approach in terms of the average controller energy and mean squared error over the conventional ANN-NMPC strategies and PID control strategy for both SISO and MIMO systems.

Ces travaux de recherche abordent un problème de défense anti-aérien, usuellement appelé problème d'allocation d'armes à des cibles dans la littérature. L'allocation d'armes à des cibles est un problème bien connu de la recherche opérationnelle militaire qui a rencontré un franc succès parmi la communauté des chercheurs, et qui aujourd'hui encore suscite un large engouement puisque sa propriété démontrée NP-difficile en fait un problème qui reste irrésolu. Que ce soit par des méthodes analytiques ou meta-heuristiques, le problème d'allocation d'armes à des cibles a fait l'objet de nombreuses propositions de résolution. Cependant, il est assez surprenant de voir que la modélisation proposée pour ce problème n'a guère évolué depuis qu'il est apparu pour la première fois dans la littérature en 1950. Cette modélisation peut être considérée comme obsolète aujourd'hui et ne répond plus aux exigences qui accompagnent les technologies modernes. En effet, en 60 ans le champ de bataille a complètement changé, et dans la littérature seulement un nombre limité d'études proposent de prendre en compte ces évolutions. L'étude menée dans cette thèse propose de s'intéresser aux systèmes de Commandes et Contrôles (C2) pour des applications anti-aériennes. Habituellement un système C2 est composé de senseurs, d'un centre d'opérations tactiques et d'un ou plusieurs lanceurs. Les senseurs alimentent le centre d'opérations tactiques à partir des informations qu'ils recueillent, puis, une fois ces informations reçues, le centre d'opérations tactiques va interpréter ces données afin de calculer l'atteignabilité des menaces. Enfin, un plan d'engagement qui comprend l'allocation des munitions disponibles aux cibles et une date de tir sont proposés à un opérateur humain qui aura pour mission de valider cette proposition en totalité ou partiellement, puis va procéder à l'engagement des menaces. Pour remplir cet objectif, une approche innovante et faisant l'objet d'un dépôt de brevet a été développée afin de répondre aux difficultés relatives aux problèmes d'optimisation multi-objectifs. Ensuite, un algorithme d'optimisation continue basé sur la combinaison de l'optimisation par essaim particulaires avec la théorie des jeux évolutionnaires est proposé pour optimiser les dates de tirs. L'allocation optimale, elle, est obtenue en adaptant cette méthode continue au cas discret. La preuve que l'algorithme développé est localement convergent est donnée dans cette thèse. D'autre part, l'aspect temps-réel a également fait l'objet d'une recherche attentive et l'algorithme précédemment cité a été hybridé avec les réseaux de neurones afin d'accélérer le temps de calcul des composants identifiés comme "lourds" en termes de charge de calcul. Enfin, cette étude ne se limite pas à une application de recherche opérationnelle militaire, mais inclut quelques concepts élémentaires de guidage et de navigation pour le calcul d'atteignabilité des menaces. Finalement, cette thèse permet d'identifier que les points suivants doivent faire l'objet d'une attention très particulière afin de développer un outil d'aide à la décision efficace. D'abord, la métrique d'évaluation d'un bon plan d'engagement doit être clairement analysée. Ensuite, le plan d'engagement proposé doit être stable et ne pas proposer de changements soudains qui pourraient perturber l'opérateur. Le troisième point concerne la robustesse de la solution proposée et sa capacité à faire face aux situations les plus compliquées. Quatrièmement, le temps et la charge de calcul sont des contraintes techniques qui ne peuvent pas être outrepassées. Finalement, les exigences posées lors de la préparation de mission et qui dépendent du contexte doivent faire l'objet d'une attention particulière. C'est pourquoi, l'outil d'aide à la décision proposé doit permettre un allègement significatif de la charge de travail de l'opérateur ainsi que la réduction considérable du stress lié à ce contexte / This research investigates a practical air defence problem, usually named Weapon Target Assignment (WTA) in the literature. The WTA problem is a well-known problem of military operation research that encountered a wide success in the research community, but still nowadays since it remains an unsolved problem because of its NP-hardness property. From analytical to heuristic methods, the WTA was deeply investigated and many attempts to solve this problem have been proposed. However, the proposed modelling of this problem is consistent with the 1950's technologies. Thus, the proposed modelling found in the literature can be considered as obsolete and cannot fit the requirement of the current technology advances. Indeed, the battle field dramatically changes over 60 years, and the recent literature proposes only few studies taking into account these amendments. The herein study proposes to investigate a Command & Control system (C2) in air defence applications. Usually a C2 system includes sensors, a Tactical Operation Centre (TOC) and one or more launchers. The sensors provide information about aerial tactical situation to the TOC. This TOC is in charge of evaluating the received information in order to compute the attainability of the targets, then an engagement plan that includes the assignment of the available weapons to the incoming targets and a date to fire for each assignment. This engagement plan is then proposed to one human operator in charge of accepting whole or part of this engagement plan and engage the targets following the received instructions. To achieve this goal, an innovative and patented approach to mitigate the issues related to multi-objective optimisation is proposed. Then, a continuous optimisation algorithm based on the combination of the Particle Swarm Optimisation and the Evolutionary Game Theory was proposed to determine the best dates to fire. The optimal assignment was obtained by adapting the aforementioned algorithm to the discrete case. This thesis also gives the proof that the designed algorithms are locally convergent and intensive benchmarking confirms the developed theory. In order to respect the real-time requirement, it was also devised to use the Neural Networks to lighten the identified burdensome parts of the algorithm and decrease computational time. Not limited to the military operation research field, the herein study reuse some basic concepts of missile guidance and navigation to compute the attainability of the targets. From this thesis, it can be identified that following aspects need to be carefully considered to provide an efficient decision making support to a human operator: First, clearly define what a good engagement plan is. Second, the engagement plan must be steady to avoid high rate changing in the assignments that could significantly disturb the operator. Third, the proposed engagement also must be reliable and robust to face any possible situations. Fourth, the computation time and computation load are technical constraints that cannot be overstepped. Finally, the operational constraints related to the mission context defined during a pre-mission stage must also be taken into account. Therefore, the proposed decision making support must help and significantly reduce the operator's work load in this situation of high stress and sensitive context

The Bayesian network (BN) formalism is a powerful representation for encoding domains characterised by uncertainty. However, before it can be used it must first be constructed, which is a major challenge for any real-life problem. There are two broad approaches, namely the hand-crafted approach, which relies on a human expert, and the data-driven approach, which relies on data. The former approach is useful, however issues such as human bias can introduce errors into the model. We have conducted a literature review of the expert-driven approach, and we have cherry-picked a number of common methods, and engineered a framework to assist non-BN experts with expert-driven construction of BNs. The latter construction approach uses algorithms to construct the model from a data set. However, construction from data is provably NP-hard. To solve this problem, approximate, heuristic algorithms have been proposed; in particular, algorithms that assume an order between the nodes, therefore reducing the search space. However, traditionally, this approach relies on an expert providing the order among the variables --- an expert may not always be available, or may be unable to provide the order. Nevertheless, if a good order is available, these order-based algorithms have demonstrated good performance. More recent approaches attempt to ''learn'' a good order then use the order-based algorithm to discover the structure. To eliminate the need for order information during construction, we propose a search in the entire space of Bayesian network structures --- we present a novel approach for carrying out this task, and we demonstrate its performance against existing algorithms that search in the entire space and the space of orders. Finally, we employ the hand-crafting framework to construct models for the task of diagnosis in a ''real-life'' medical domain, dementia diagnosis. We collect real dementia data from clinical practice, and we apply the data-driven algorithms developed to assess the concordance between the reference models developed by hand and the models derived from real clinical data.

Cette thèse a pour objectif principal le développement de nouveaux algorithmes globaux pour la résolution de problèmes d’optimisation mono et multi-objectif, en se basant sur des formules de représentation ayant la tâche principale de générer des points initiaux appartenant à une zone proche du minimum globale. Dans ce contexte, une nouvelle approche appelée RFNM est proposée et testée sur plusieurs fonctions non linéaires, non différentiables et multimodales. D’autre part, une extension à la dimension infinie a été établie en proposant une démarche pour la recherche du minimum global. Par ailleurs, plusieurs problèmes de conception mécanique, à caractère aléatoire, ont été considérés et résolus en utilisant cette approche, avec amélioration de la méthode multi-objectif NNC. Enfin, une contribution à l'optimisation multi-objectif par une nouvelle approche a été proposée. Elle permet de générer un nombre suffisant de points pour représenter la solution optimale de Pareto. / The main objective of this work is to develop new global algorithms to solve single and multi-objective optimization problems, based on the representation formulas with the main task to generate initial points belonging to an area close to the global minimum. In this context, a new approach called RFNM is proposed and tested on several nonlinear, non-differentiable and multimodal finctions. On the other hand, an extension to the infinite dimension was established by proposing an approach for finding the global minimum. Moreover,several random mechanical design problems were considered and resolved using this approach, and improving the NNC multi-objective method. Finally, a new multi-objective optimization method called RSMO is presented. It solves the multi-objective optimization problems by generating a sufficient number o fpoints in the Pareto front.