The step response of an MPC control system for a nonlinear plant gives a response with vertical line before it starts tracking. I have been trying to figure out what could be responsible for the vertical line and how to eliminate it. suggestions on how to eliminate the vertical or what could be repsonsible is welcome. I have attached the response below
Kamal Abdulraheem
Thank you for sharing the step response plot. The vertical line artifact you see before the controller response kicks in is likely caused by the predictive model and receding horizon control mismatch:
- At the step change, the model predictive controller is basing its control actions on predicting the future plant output using its internal model.
- However, the real plant output does not start responding until the actual control input from MPC starts actuating it. This causes the temporary mismatch.
- Once the plant starts receiving the optimal control inputs and its output aligns with the predictions, the vertical artifact disappears.
Some ways to potentially minimize this:
- Tune the prediction horizon Hp to be as small as possible while still maintaining control performance. Shorter horizons reduce the artifact duration.
- Set the control horizon Hc shorter than Hp so control inputs react faster to changes. But stability has to be ensured.
- Improve dynamic accuracy of the internal process model to better match real plant lag. Model mismatch exacerbates the issue.
Please let me know if this explanation helps point you in the right direction.
@Qamar UI Islam Thank you so much for the insight. I have tuned it with several values of prediction horizon and kept it at higher value than control horizon. Also, I have applied small values of prediction horizon as low as 1 and 2. But the vertical line is still there no matter the value!
- Badges
- Science topic