Trajectory-Scheduled Controller Networks and Application in Nonlinear Process Control

This paper develops a trajectory controller network (TCN), formed with a set of simple linear blocks, for nonlinear control. Utilising linear controller synthesis techniques, the network and its nodes are straightforward to interpret and apply, unlike a neural network or fuzzy logic system, which need excessive network elements and weighting with physical meanings indirect to interpret. The TCN can be structured easily and designed directly from plant step-response data along the operating trajectory, effectively covering the entire operating envelope. The paper also offers an evolutionary computation automated approach to the design of the TCN with multiple design objectives satisfied. Application to a polyethylene chemical process is illustrated, which confirms super performance in terms of minimum error, minimum control effort and maximum load disturbance rejection, by using just three simple PID controller nodes.