Desarrollo de modelo diferencial neurodifuso de hidrogenerador de la Central Agoyán para la sintonización de controlador PID lineal invariante en el tiempo y PID neurodifuso.

Abstract

In this investigation, the differential linear model and the invariant of time (LTI) was developed and the neuro-fuzzy of the hydro-generation plant of the Central Agoyan, located in Banos de Agua Santa, for the subsequent tuning of the PID LTI controllers and the neuro-fuzzy; for this reason, the LTI models of the turbine, the generator and the servo-motor were divided and studied, making use of the optimization method of the decreasing gradient, accomplishing the adjustment of registration of electric power from the generator in response to the variation of the control signal. After obtaining the model LTI, the neuro-fuzzy model was put together, and with the decreasing gradient method, the single-tones were adjusted, or the weights of the model that allowed reproduction of the already obtained linear behavior, allowing for equality of performance at the point of comparison of records between both models, LTI and neuro-fuzzy. Finally, a cost function was planted that minimized the mechanical stress calculated by the level of servomotors of the model, allowing the adjustment of the PID controller, both the neuro-fuzzy and the LTI. As a result, the hypothesis was verified, demonstrating that a PID neuro-fuzzy controller, when presenting more parameters for its tuning, based on the initial response, equal to the controller PID LTI, was capable of presenting better performance when contrasting its response in a comparative way with a sign of performance pattern, and employing the origin of the average quadratic error and the correlation factor Pearson quadratic multivariable as criteria for valuation, compared to a cost function, as criteria for valuation of performance during the optimization employing the decreasing gradient method, which considers the quadratic error and the square of the velocity of the global servo-motor.