Nonlinear Control Design Blockset

Control and Identification

Problem: Which variables should I choose to tune/identify? Is there a limit to how many I can optimize?

Recommendation: Because the time necessary for optimization is proportional to the number of tunable variables, it is best to use minimal parameterizations (i.e., tune the fewest number of variables possible for a given control structure). For SISO state-space controllers, minimal parameterizations are given by the various canonical forms. This generalizes to MIMO state-space systems although MIMO canonical forms are less familiar to the average control engineer.

Problem: How should I choose initial conditions for my tunable variables?

Recommendation: No theory exists which can guarantee an initial stabilizing controller of arbitrary structure. In fact, determining whether a stabilizing controller of arbitrary structure even exists is an unsolved problem. Typical methods for attempting to generate an initial stabilizing controller include

• Generating a controller for a linearized system
• Trying a controller that worked for a similar system

Actually, the Nonlinear Control Design Blockset does not require an initial stabilizing controller in order to begin its optimization. As long as the constrained signals remain bounded during the time horizon of the optimization, you can use the Nonlinear Control Design Blockset. Of course, we cannot guarantee that the Nonlinear Control Design Blockset converges to a stabilizing solution.

Troubleshooting

Optimization