O método do Gradient Sampling com comprimento de passo limitado aplicado ao problema de determinação das condições de equilíbrio operacionais de sistemas dinâmicos

Abstract

In this Master Dissertation, it is proposed to restrict (limit) the search direction of the Gradient Sampling method, which means to add a box constraint to each quadratic optimization sub-problem. It is called Gradient Sampling with Limited Step Length the algorithm proposed with this change and mathematically demonstrated that this modification doesn’t change the convergence analysis to an e-stationary, or, stationary, point of the optimization’s problem. Finally, this proposed method, Gradient Sampling with Limited Step Length, is tested for a problem of finding the operational equilibrium condition of low complexity dynamic systems, with up to 40 states’ variables. As an example of dynamic systems, therm-pneumatic systems have been chosen for this task.