There is NO WAY they should be able to screw up PID like this!
Notice that the MPC control on page 5549 starts moving BEFORE the control signal increase. That isn't right. Often in the FL examples they FL champions do a horrible job of tuning the PID to make the FL look better. I think the same happened in this case. The plant model they used is too simple to screw up if they knew what they are doing. The sad truth is that they don't know how to tune a PID. There are two reasons why the PID response would look like that. Poor tuning and poor feedback resolution. Since they are simulating the resolution should be that of floating point numbers.
I don't see where enough information is provided make sure this is a fair test.
The article is written by students and not checked well by their instructor.
I have read many articles from China and India. You must be careful. I have seen plenty of these articles before. The motivation is to write a paper and not anything really useful or even accurate.
I would have failed these students for butchering the PID.
I have a very good example of how to control the level in the second tank of a cascaded tank system where a pump pumps water into a higher first tank. The water in the first tank flows out a fixed orifice in the bottom into the second tank. The water in the second lower tank flows out a fixed orifice. This is a much more complicated example of level control and it is done with PI or PID control. I would have like to see the students that attempted simple MPC tank level control tackle this problem.
This is the final version. I don't assume the tanks have vertical sides so the gain changes due to the changing surface areas. I don't assume the flow through the orifices is constant as a function of height. The flow will change with the square root of the tank level. I even add disturbances. The tank level control is perfect in that there is no over shoot because I can calculate the instantaneous plant gains and time "constants" and then calculate the resulting controller gains. Yes, time "constants" are not always constant.
http://deltamotion.com/peter/Mathcad/TwoTanks/Mathcad - t0p2 p pi Alin's two tanks Cascade.pdf
I have provided the model and the level set points. I challenge the FL, NN and MPC advocates to do better and I am only using a P and PI controller.