Adaptive neural-network-based approach for the control of continuously stirred tank reactor

An online adaptive neural-network-based controller (OANNC) is developed in the paper. The detailed design procedures of the OANNC are given, along with an illustrative example of controlling a continuously stirred tank reactor. Simulation results show that the OANNC is successful in controlling nonlinear time-varying systems with slow dynamics. Compared with conventional neural-network controllers. The OANNC has the following advantages. First, it is capable of controlling nonlinear systems with time-varying parameters, which is not usually the case for a nonadaptive neural-network controller. Secondly, the selection of the initial training data set is trivial due to the online adaptive training ability of the neural network. Normally, for a conventional neural-network structure, the selection of an initial training data set is crucial, with the requirement that the training data set should be persistently exciting, which is quite difficult in many practical situations