Undesirable Parameter Values
What to do if the optimization drives the tuned compensator elements and parameters to undesirable values?
When a tuned compensator element or parameter is positive, or when its value is physically constrained to a given range, enter the lower and upper bounds (Minimum and Maximum) in one of the following:
Dialog box to select design variables (in Response Optimizer)
Compensators pane (in Control System Designer)
This information helps guide the optimization method towards a reasonable solution.
Specify initial guesses that are within the range of desirable values.
In the Compensators pane in Control System Designer, verify that no integrators/differentiators are selected for optimization. Optimizing the pole/zero location of integrators/differentiators can result in drastic changes in the system gain and lead to undesirable values.
What to do if the optimization violates bounds on parameter values?
Gradient descent optimization method
fmincon violates the parameter bounds when it cannot simultaneously
satisfy the signal constraints and the bounds. When this happens, try one of the
Specify a different value for the parameter bound and restart the optimization. A guideline is to adjust the bound by 1% of the typical value.
For example, for a parameter with a typical value of
1and lower bound of
0, change the lower bound to
Relax the signal constraints and restart the optimization. This approach results in a different solution path for the
Restart the optimization immediately after it terminates by clicking Optimize in the Response Optimizer. This approach uses the previous optimization results as the starting point for the next optimization cycle to refine the results.
Use the following two-step approach to perform the optimization:
Run an initial optimization to satisfy the signal constraints.
For example, run the optimization using the
Simplex searchmethod. This method satisfies the signal constraints but does not support the bounds on parameter values. The results obtained using this method provide the starting point for the optimization performed in the next step. To learn more about this method, see the
fminsearchfunction reference page in the Optimization Toolbox™ documentation.
Reconfigure the optimization by selecting a different optimization method to satisfy both the signal constraints and the parameter bounds.
For example, change the optimization method to
Gradient descentand run the optimization again.
If Global Optimization Toolbox software is installed, you can select the
Pattern searchoptimization method to optimize the model response.