Neural networks based control of pH neutralization plant

This investigation identifies the relative properties of eight control structures of the pH neutralization process by extensive simulations. Eight different feedforward neural networks (NN) are used to cover a variety of functions in control systems: first principle model emulation in order to overcome computational restrictions, non measurable disturbances estimation, gain scheduling tuning of standard PID controller parameters, feedforward control action formation, time delay compensation by NN-based prediction. Combining prior knowledge of the pH neutralization process with NN learning a Hammerstein plant model is accepted. Special attention is paid to study the impact of NN accuracy on the generalization ability and the performance of the control system in the presence of various NNs and closed loops.