in @8:08, why CV target and CV actual don't get to the steadystate, can this introduce some identification errors !!!! or the dynamic model is already identified?
More information on model identification is here: apmonitor.com/do/index.php/Main/AdvancedTemperatureControl (see Labs A-E). In this example, it doesn't matter that we haven't reached steady state because we aren't identifying a model. However, even if it were model identification, steady state isn't required although it is nice to see that the model performs well in reaching a target value.
How do I simulate mpc for a power system network in Matlab? I tried loading the case file and then calling the mpc function but it said the mpc command must be an LTI object or structure of models and offset
This package doesn't use the MPC toolbox provided by Mathworks. Here is the mpc toolbox with APMonitor: apmonitor.com/wiki/index.php/Main/MATLAB You may also want to consider using Python instead with the Gekko package. See apmonitor.com/do/index.php/Main/EnergyBenchmarks and apmonitor.com/do/index.php/Main/TCLabF
1/ How do you compute the reference trajectory (target) in the first place? 2/ How to use MPC for nonlinear control systems for which the optimization becomes nonconvex?
The reference trajectory is defined here: apmonitor.com/do/index.php/Main/ControllerObjective If you have a nonconvex system then you may need to use multi-start methods or a global optimizer. This may make real-time calculation impossible if the computation time is longer than the required sampling and cycle time.
Emmanuel Da Yes, predictive control has a fixed model that doesn't change as new data arrives. Most industrial controllers are not adaptive. With adaptive controls the model is changing as new data arrives and an improved representation of the process is discovered. Sometimes bad data or the wrong type of data (e.g. steady state) can degrade the model predictive capabilities. Adaptive control is typically only used on processes that change over time where the model becomes inaccurate.
Thank you very much for the excellent explanations... save my lot of time. Just curious to know, Is this control concept can be applied to highly nonlinear system ? Lets say for an inverted triple pendulum? If not do you have any performance comparison with some nonlinear control theme like sliding mode (except PID) or anything with MPC. Keep up the good work..!
Nahian Rahman We've completed applications with pendulums and inverted pendulums. One of the drawbacks of MPC is that the computational time can be much more than something like PID or LQR controllers because you need to solve an implicit and iterative optimization problem each time step. As long as the MPC application can calculate fast enough, there shouldn't be any problem with controlling a highly nonlinear system. Check out the dynamic optimization course at apmonitor.com/do for additional examples and videos.
Here is the PDF: apmonitor.com/che436/uploads/Main/Lecture32b_notes.pdf or other course material: apmonitor.com/che436/index.php/Main/LectureNotes33 I don't distribute the PPT files.
janaki sreeram I don't have that specific model but here is a reservoir model: apmonitor.com/do/index.php/Main/ModelFormulation The hydro power valley model could be derived from this by relating flow and reservoir height to power generation.
I can't understand the target thing, it's like we know the solution before we turn on the system,the robot or anything... what if the environment is unknow
+Karam AbuGhalieh, sometimes the target changes as we get new information. The target is usually a reference trajectory that also tells the application the route to follow. You may need a supervisory application that gives target or trajectory information to the Model Predictive Controller.
APMonitor.com thanks for reply... so we are predicting the future input value to match the output we desire. sorry if my question it self is wrong. but I am still confused about the whole concept.
Yes, we are calculating a sequence of process inputs to drive the system to a desired state or along a desired trajectory. More information on model predictive control is in the dynamic optimization course at apmonitor.com/do Check out the links for MPC on the right.
That was an excellent lecture. I was having a hard time getting through textbooks on MPC before i watched your lecture
I'm glad you liked it - I personally enjoy lectures that include simple examples that demonstrate the concepts.
great !!! I have seen several of your tutorial video. really appreciate it
in @8:08, why CV target and CV actual don't get to the steadystate, can this introduce some identification errors !!!! or the dynamic model is already identified?
More information on model identification is here: apmonitor.com/do/index.php/Main/AdvancedTemperatureControl (see Labs A-E). In this example, it doesn't matter that we haven't reached steady state because we aren't identifying a model. However, even if it were model identification, steady state isn't required although it is nice to see that the model performs well in reaching a target value.
How do I simulate mpc for a power system network in Matlab? I tried loading the case file and then calling the mpc function but it said the mpc command must be an LTI object or structure of models and offset
This package doesn't use the MPC toolbox provided by Mathworks. Here is the mpc toolbox with APMonitor: apmonitor.com/wiki/index.php/Main/MATLAB You may also want to consider using Python instead with the Gekko package. See apmonitor.com/do/index.php/Main/EnergyBenchmarks and apmonitor.com/do/index.php/Main/TCLabF
@@apm Thank you very much
1/ How do you compute the reference trajectory (target) in the first place?
2/ How to use MPC for nonlinear control systems for which the optimization becomes nonconvex?
The reference trajectory is defined here: apmonitor.com/do/index.php/Main/ControllerObjective
If you have a nonconvex system then you may need to use multi-start methods or a global optimizer. This may make real-time calculation impossible if the computation time is longer than the required sampling and cycle time.
Great tutorial !! Thank you.
thank you for your information...excellent lecture
Is there a difference between Predictive control and Predictive adaptive control? Please assist with quick response.
Emmanuel Da Yes, predictive control has a fixed model that doesn't change as new data arrives. Most industrial controllers are not adaptive. With adaptive controls the model is changing as new data arrives and an improved representation of the process is discovered. Sometimes bad data or the wrong type of data (e.g. steady state) can degrade the model predictive capabilities. Adaptive control is typically only used on processes that change over time where the model becomes inaccurate.
Thank you very much for the excellent explanations... save my lot of time. Just curious to know, Is this control concept can be applied to highly nonlinear system ? Lets say for an inverted triple pendulum? If not do you have any performance comparison with some nonlinear control theme like sliding mode (except PID) or anything with MPC. Keep up the good work..!
Nahian Rahman We've completed applications with pendulums and inverted pendulums. One of the drawbacks of MPC is that the computational time can be much more than something like PID or LQR controllers because you need to solve an implicit and iterative optimization problem each time step. As long as the MPC application can calculate fast enough, there shouldn't be any problem with controlling a highly nonlinear system. Check out the dynamic optimization course at apmonitor.com/do for additional examples and videos.
Thank you so much for this explanation.
Noree Stirak You are welcome - thanks for your comment.
Sir, can I get your ppt?
Here is the PDF: apmonitor.com/che436/uploads/Main/Lecture32b_notes.pdf or other course material: apmonitor.com/che436/index.php/Main/LectureNotes33 I don't distribute the PPT files.
Sir can you plz give me hydro power valley model with model predictive control.
janaki sreeram I don't have that specific model but here is a reservoir model: apmonitor.com/do/index.php/Main/ModelFormulation The hydro power valley model could be derived from this by relating flow and reservoir height to power generation.
Sir can you guide me for Solid oxide fuel cell model MPC ?
I can't understand the target thing, it's like we know the solution before we turn on the system,the robot or anything... what if the environment is unknow
+Karam AbuGhalieh, sometimes the target changes as we get new information. The target is usually a reference trajectory that also tells the application the route to follow. You may need a supervisory application that gives target or trajectory information to the Model Predictive Controller.
APMonitor.com thanks for reply... so we are predicting the future input value to match the output we desire.
sorry if my question it self is wrong. but I am still confused about the whole concept.
Yes, we are calculating a sequence of process inputs to drive the system to a desired state or along a desired trajectory. More information on model predictive control is in the dynamic optimization course at apmonitor.com/do Check out the links for MPC on the right.
This link? apmonitor.com/do/index.php/Main/ControlTypes