<p> Hierarchical Interface-based Supervisory Control (HISC) decomposes a discrete-event system (DES) into a high-level subsystem which communicates with n ≥ 1 low-level subsystems, through separate interfaces which restrict the interaction of the subsystems. It provides a set of local conditions that can be used to verify global conditions such as nonblocking and controllability. As each clause of the definition can be verified using a single subsystem, the complete system model never needs to be stored in memory, offering potentially significant savings in computational resources.</p> <p> Currently, a designer must create the supervisors for a HISC system himself, and then verify that they satisfy the HISC conditions. In this thesis, we develop a synthesis method that respects the HISC hierarchical structure. We replace the supervisor for each level by a corresponding specification DES. We then do a per level synthesis to construct for each level a maximally permissive supervisor that satisfies the corresponding HISC conditions.</p> <p> We define a set of language based fixpoint operators and show that they compute the required level-wise supremal languages. We then present algorithms that implement the fixpoint operators. We present a complexity analysis for the algorithms and show that they potentially offer significant improvement over the monolithic approach.</p> <p> A large manufacturing system example (estimated worst case state space on the order of 10^22) extended from the AIP example is discussed. A software tool for synthesis and verification of HISC systems using our approach was also developed.</p> / Thesis / Master of Applied Science (MASc)

The optimal estimation theory of the Wiener-Kalman filter is extended to cover the situation in which the number of memory elements in the estimator is restricted. A method, based on the simultaneous diagonalisation of two symmetric positive definite matrices, is given which allows the weighted least square estimation error to be minimised. A control system design method is developed utilising this estimator, and this allows the dynamic controller in the feedback path to have a low order. A 12-order once-through boiler model is constructed and the performance of controllers of various orders generated by the design method is investigated. Little cost penalty is found even for the one-order controller when compared with the optimal Kalman filter system. Whereas in the Kalman filter all information from past observations is stored, the given method results in an estimate of the state variables which is a weighted sum of the selected information held in the storage elements. For the once-through boiler these weighting coefficients are found to be smooth functions of position, their form illustrating the implicit model reduction properties of the design method. Minimal-order estimators of the Luenberger type also generate low order controllers and the relation between the two design methods is examined. It is concluded that the design method developed in this thesis gives better plant estimates than the Luenberger system and, more fundamentally, allows a lower order control system to be constructed. Finally some possible extensions of the theory are indicated. An immediate application is to multivariable control systems, while the existence of a plant state estimate even in control systems of very low order allows a certain adaptive structure to be considered for systems with time-varying parameters.

Since the field of computer networks has rapidly grown in the last two decades, congestion control of traffic loads within networks has become a high priority. Congestion occurs in network routers when the number of incoming packets exceeds the available network resources, such as buffer space and bandwidth allocation. This may result in a poor network performance with reference to average packet queueing delay, packet loss rate and throughput. To enhance the performance when the network becomes congested, several different active queue management (AQM) methods have been proposed and some of these are discussed in this thesis. Specifically, these AQM methods are surveyed in detail and their strengths and limitations are highlighted. A comparison is conducted between five known AQM methods, Random Early Detection (RED), Gentle Random Early Detection (GRED), Adaptive Random Early Detection (ARED), Dynamic Random Early Drop (DRED) and BLUE, based on several performance measures, including mean queue length, throughput, average queueing delay, overflow packet loss probability, packet dropping probability and the total of overflow loss and dropping probabilities for packets, with the aim of identifying which AQM method gives the most satisfactory results of the performance measures. This thesis presents a new AQM approach based on the RED algorithm that determines and controls the congested router buffers in an early stage. This approach is called Dynamic RED (REDD), which stabilises the average queue length between minimum and maximum threshold positions at a certain level called the target level to prevent building up the queues in the router buffers. A comparison is made between the proposed REDD, RED and ARED approaches regarding the above performance measures. Moreover, three methods based on RED and fuzzy logic are proposed to control the congested router buffers incipiently. These methods are named REDD1, REDD2, and REDD3 and their performances are also compared with RED using the above performance measures to identify which method achieves the most satisfactory results. Furthermore, a set of discrete-time queue analytical models are developed based on the following approaches: RED, GRED, DRED and BLUE, to detect the congestion at router buffers in an early stage. The proposed analytical models use the instantaneous queue length as a congestion measure to capture short term changes in the input and prevent packet loss due to overflow. The proposed analytical models are experimentally compared with their corresponding AQM simulations with reference to the above performance measures to identify which approach gives the most satisfactory results. The simulations for RED, GRED, ARED, DRED, BLUE, REDD, REDD1, REDD2 and REDD3 are run ten times, each time with a change of seed and the results of each run are used to obtain mean values, variance, standard deviation and 95% confidence intervals. The performance measures are calculated based on data collected only after the system has reached a steady state. After extensive experimentation, the results show that the proposed REDD, REDD1, REDD2 and REDD3 algorithms and some of the proposed analytical models such as DRED-Alpha, RED and GRED models offer somewhat better results of mean queue length and average queueing delay than these achieved by RED and its variants when the values of packet arrival probability are greater than the value of packet departure probability, i.e. in a congestion situation. This suggests that when traffic is largely of a non bursty nature, instantaneous queue length might be a better congestion measure to use rather than the average queue length as in the more traditional models.

Submitted by Maria Aparecida (cidazen@gmail.com) on 2017-08-30T15:33:12Z No. of bitstreams: 1 Patricia Helena.pdf: 11110405 bytes, checksum: ca1f067231658f897d84b86181dbf1b9 (MD5) / Made available in DSpace on 2017-08-30T15:33:12Z (GMT). No. of bitstreams: 1 Patricia Helena.pdf: 11110405 bytes, checksum: ca1f067231658f897d84b86181dbf1b9 (MD5) Previous issue date: 2014-07-24 / In this thesis a proposal of an uni ed approach of dynamic programming, reinforcement learning and function approximation theories aiming at the development of methods and algorithms for design of optimal control systems is presented. This approach is presented in the approximate dynamic programming context that allows approximating the optimal feedback solution as to reduce the computational complexity associated to the conventional dynamic programming methods for optimal control of multivariable systems. Speci cally, in the state and action dependent heuristic dynamic programming framework, this proposal is oriented for the development of online approximated solutions, numerically stable, of the Riccati-type Hamilton-Jacobi-Bellman equation associated to the discrete linear quadratic regulator problem which is based on a formulation that combines value function estimates by means of a RLS (Recursive Least-Squares) structure, temporal di erences and policy improvements. The development of the proposed methodologies, in this work, is focused mainly on the UDU T factorization that is inserted in this framework to improve the RLS estimation process of optimal decision policies of the discrete linear quadratic regulator, by circumventing convergence and numerical stability problems related to the covariance matrix ill-conditioning of the RLS approach. / Apresenta-se nesta tese uma proposta de uma abordagem uni cada de teorias de programação dinâmica, aprendizagem por reforço e aproximação de função que tem por objetivo o desenvolvimento de métodos e algoritmos para projeto online de sistemas de controle ótimo. Esta abordagem é apresentada no contexto de programação dinâmica aproximada que permite aproximar a solução de realimentação ótima de modo a reduzir a complexidade computacional associada com métodos convencionais de programação dinâmica para controle ótimo de sistemas multivariáveis. Especi camente, no quadro de programação dinâmica heurística e programação dinâmica heurística dependente de ação, esta proposta é orientada para o desenvolvimento de soluções aproximadas online, numericamente estáveis, da equação de Hamilton-Jacobi-Bellman do tipo Riccati associada ao problema do regulador linear quadrático discreto que tem por base uma formulação que combina estimativas da função valor por meio de uma estrutura RLS (do inglês Recursive Least-Squares), diferenças temporais e melhorias de política. O desenvolvimento das metodologias propostas, neste trabalho, tem seu foco principal voltado para a fatoração UDU T que é inserida neste quadro para melhorar o processo de estimação RLS de políticas de decisão ótimas do regulador linear quadrá- tico discreto, contornando-se problemas de convergência e estabilidade numérica relacionados com o mal condicionamento da matriz de covariância da abordagem RLS.