Inplementation of a Model Ref-erence Adaptive Control System Using Neural Network to Control a Fast Breeder Reactor Steam Evaporator

Ugolini, D.; Yoshikawa, Shinji ; Endo, Akira

This paper discusses the development of an indirect model reference adaptive control (MRAC) system, using the artificial neural network (ANN) technique, and its implementation to control the outlet steam temperature of a sodium to water helical-coil once-through evaporator. The ANN technique is applied in the identification process and in the control process of the indirect MRAC system. The evaporator is simulated with a nonlinear dynamic modular model representing a superheated cycle with three regions, subcooled, saturated, and superheated, and moving boundaries. The emphasis is placed on demonstrating the efficacy of the indirect MRAC system in the control of the outlet steam temperature of the evaporator model, and on showing the important function covered by the ANN technique, whose adaptation and learning capabilities, contribute to improve the performance of the control action of the indirect MRAC system. The implementation of the ANN technique in the indirect MRAC system generates a strong control system. An important characteristic of this control system is that it relays only on some selected input variables and on the output variables of the evaporator model. These are the variables that can be actually measured or calculated in a real environment. Therefore, the internal variables, which are needed to develop the model, but that can be hardly measured or calculated in a real environment, are not utilized during the control action performed by the indirect MRAC system. The results obtained applying the indirect MRAC system to control the evaporator model are quite remarkable. The outlet temperature of the steam is almost perfectly kept close to its desired set point, when the evaporator model is forced to depart from steady state conditions, either due to the variation of some input variables or due to the alteration of some of its internal parameters. The results also show the importance of the role played by the ANN technique in the overall ...