I'm thinking about which one is the correct nomenclature: Non-model based control or model free control?
I'm trying to establish these concepts and challenge them, because I'm researching new control systems techniques that are "totally" model independent (or more independent than the actual ones). In other words, a control technique that rely only on process historical data.
During my bibliography review, I've faced these two nomenclatures, but in my point of view, with the same meaning. The community always say: "PID controllers are Non-model based...."
"...PID are model free because you don't need to model the process to apply them..."
"...fuzzy controllers are non-model based..."
"..model free controllers can be categorized as the controllers which have online training process (talking about neural network an machine learning techniques applied to control)..."
However I'm not satisfied with those concepts, because the core of all those techniques needs to assume some premise that ends up in a model (with PID we have to assume some "first order behavior"...NN is a kind of weighted ensemble of models empirical...and so on...)
(I stated to think that Pseudo-Non-Model Control is better to call them)
All that said, I expose these thoughts to looking for some help to think outside of the box....it is possible to apply control actions only based on the historical data structure?
note: please, correct me if I'm wrong!
📷
---------------------------------DIRECTION TO Solve--------------------------
Dear Giovani Gerevini ,
Now come to your Question:
- PID, fuzzy, NN etc., all are algorithms (or control tools, for example Control System Toolbox in MATLAB software) and nothing, and these algorithms may be in the form of time dynamics (that means what to do at what time).some algorithm require more information and some are less.
- To identify the behavior of process data, it is required a algorithm to handle these process data. Ultimately we need an algorithm. And some algorithms require process data only or some algorithms require model/dynamics or some algorithms require both the process data + Model/dynamics.
- Think just one second, for example if we take AUTONOMOUS VEHICLE as a system and Line Following as target of the system then how can this possible to achieve this target by one output history. So, in this process, we have to require various types of hard sensors (such as LIDAR, Camera sensor, Accelerometer, and so many) and soft sensors to achieve the target.
- Also, see this following page that is written by me.
https://www.researchgate.net/post/Do_we_have_misunderstanding_in_control_theory
- SO, THERE MAY HAVE ANY NAME FOR ANY CONTROL METHODS (but, according to the functions of particular algorithms).
-------------------------------------THANKS--------------------------------------
A clarification: PID (or other control schemes or structures) can be both model-based and model-free. It really depends on the way you determine the control parameters. If you obtain those parameters via frequency-response-based loop shaping or heuristic tunning (e.g., Ziegler–Nichols), it can be considered model-free. However, if you compute control parameters based on some physical modelings (e.g. transfer-function-based pole placement, H-inf synthesis, etc.), then PID should be classified as model-based.
Overall, model-free in practice means control laws are determined with non-physical-based models (e.g. data-driven models or learning-based models).
Dear Giovani Gerevini ,
I think the answers of Aman Parkash and Xingyu Zhou are pretty comprehensive and would give a good overview of the topic. To the previous answers, I would like to mention the recent work:
Article Deterministic continuous-time Virtual Reference Feedback Tun...
It relies on the data-driven paradigm, which addresses the problem of computing a PID controller based on the historical data of the process.
Moving a little bit out of the PID context, quite recently, some interesting research has been done to synthesize different controllers based only on the historical data of the process using the so-called Willems' Lemma. Some interesting references would be:
Article Formulas for Data-Driven Control: Stabilization, Optimality,...
Article Data-driven model predictive control: closed-loop guarantees...
https://www.researchgate.net/project/Data-driven-model-predictive-control-with-guarantees
I hope this information would be useful for your research.
Best,
Jorge
- Badges
- Science topic
- Similar topics
- Cognitive Science
- Thinking