Control of real-time processes using back-propagation neural networks

The authors discuss the use of appropriately trained back-propagation neural networks as physical controllers similar to conventional feedforward controllers in real-time control systems. Experiments were concluded on two process models; one was a single-input single-output water bath process, and the other a multi-input multi-output nonlinear furnace. By obtaining a set of a plant's input-output patterns, the neural networks were trained to learn their inverse dynamics and then were configured as feedforward controllers to the plants. The results show that the neural network controllers perform well. The applicability of other types of neural network control schemes is discussed.<
>