Diseño, simulación y comparación de controladores diseñados por métodos tradicionales frente a controladores PID adaptativos basados en redes neuronales con propagación inversa aplicados a un tanque reactor de agitado continuo (CSTR).

Abstract

The present work seeks to compare the performance of controllers designed by traditional methods, such as Ziegler Nichols, Dahlin, Critical Gain, Direct Synthesis versus a predictive controller based on Neural Networks, which consists of two stages. The first one is a neural network in charge of predicting the dynamic behavior of the plant, and a second part consisting of an optimizer and a tuner, which use the prediction made to calculate the optimal control action. Performance tests of the controllers applied to the plant are carried out for both: reference changes and disturbances (increment and decrement). The analysis is performed by means of performance indexes such as the integral of the squared error, ISE, and the integral of the control action, ISCO, which allow verifying by what magnitude the error has been reduced and how aggressive the control action is, respectively. As well as the parameters that characterize the system response (overshoot and settling times). In addition, a user-friendly interface is designed, developed in MATLAB App Designer, which allows visualizing the data and graphs obtained from the tests performed on all the controllers.