Control por Anulación Neuro-Difusa aplicado al Sistema de Oleoductos de Ecuador

Override control systems based on an Adaptive Neuro-Fuzzy Inference System allow to optimize the work of a process by a series of predictions of its possible inputs and outputs. It is fully compatible with linear and non-linear systems, but at the requirement of a controller with the enough processing capabilities to ensure that all the involved operations are done in real time. For extraction of the dynamic properties and identification of a plant model this research work uses Nonlinear Autoregressive Exogenous neural networks. With this model the authors develop the simulation of the Neuro-Fuzzy Controllers trained by Adaptive Neuro-Fuzzy Inference System, which are part of an override control structure for an oil pipeline process. The results demonstrate that the proposed neuro-fuzzy controller completely eliminates the 10% overshoot present in conventional Proportional-Integral controllers, maintaining the rise time of 200 seconds and reducing the steady-state error to zero. The developed plant model achieved mean squared errors of 9.5972x10-7 and 9.9098x10-7 for suction and discharge pressures, respectively. Finally, the behavior of neuro-fuzzy controllers is analyzed by effects of disturbances and human-errors because of the non-correlation in the setpoints of the main variables, demonstrating robustness in operating ranges between 1285 and 1300 pounds per square inch gauge.