1-Conventional controllers are considered as analog hardware, although the modern industrial applications depend mainly on digital control platforms like DSP/FPGA.

2-Modern digital controllers have higher computational power and lower cost compared to analog controllers (with analog circuit elements) for long runs, avoiding the aging, and parameter changing values.

3- Programming flexibility of modern controllers (software and hardware solutions) compared to the analog ones (hardware only).

4- Digital control is suitable and efficient for industrial regulations’ and codes’ requirements. These requirements can’t be achieved using hardware only, but also it needs software solution.

5-MPC has faster and smoother response than conventional systems and control complexity is less.

6- Unlike MPC, traditional control approach requires modeling of the physical system and needs its transfer function to design and tune the control parameters.

7- MPC has a sound theoretical basis and its stability, optimality, and robustness properties are well understood.

8- Despite being very simple to design and implement, MPC algorithms can control large scale systems with many control variables, and, most importantly, MPC provides a systematic technique of dealing with constraints on inputs and states.

9- Nonlinear plant dynamics can be incorporated in MPC control.

10- MPC has lower rising time, settling time and overshoots compared to PID controllers. Moreover, digital controllers eliminate dangerous oscillations and provide smooth operation in transient period.

in addition ,* Multiple controlled variables within a single control law can be achieved 

*Inclusion of constraints for controlled variables

Dear Dr. Emad;

Of course I do agree with all the answers added above. 

The Main problem associated with the PI controller is that it is a linear controller,While all power electronics systems are nonlinear system. This may raise alot of control difficulties.

Predictive controllers are nonlinear controllers and can easily deal with with nonlinear systems.

In addition, MPC control could be regarded like an optimal control in real time; so the advantage is evidently ( with condition that we have enough power of CPU...)

I recommend this link

https://www.sheffield.ac.uk/polopoly_fs/1.60743!/file/report20.pdf

This raises a question about the accuracy of modeling, especially in nonlinear cases.

Most widely used control algorithm in material and chemical processing

industries.

• Increased consistency of discharge quality. Reduced off-specs products

during grade changeover. Increased throughput. Minimizing the operating

cost while meeting constraints (optimization, economic).

• Superior for processes with large number of manipulated and controlled

variables (Multivariable, strong coupling)

• Allows constraints to be imposed on both MV and CV. The ability to operate

closer to constraints. (constraints)

• Allow time delays, inverse response, inherent nonlinearities (difficult

dynamics), changing control objectives and sensor failure. (predictive)

I have a somewhat different view on the question. First, it containes evidently a student task. Secondly, the task consists in comparing two hardly comparable or too different entities. Each item has its shortcomings and advantages in control efficiency sense. And by these reasons I dare to gracefully disagree with parts of colleague's replies above.

In the preview, the control theory makes no difference between "PI controller" and "PI control algorithm". Furthermore, there is evidently no sense to speak about digital or analog PI control laws at all. Functionally both are equivalent, not concerning the technical disadvantages of the analog oldtimers and the versatility (provisional) and fastness (unconditional) of the digital versions.

Another point treats the power and price of both algorithms. Its evident, that MPC is too powerful and rather universal, though very sophisticated, too expensive for implementation and ... inherently clumsy weapon! An important its feature approves the lack of theoretical issues about the fundamental control properties in the general case: no strong results about stabilty exist till now, vague inclines about pre-specification and pre-design of system performance measures, thumb rules for selecting a great lot of important algorithm parameters. The most dangerous properties, in my opinion, are hidden in the full leaning on a particular process model. It is well known, that persistent or abrupt plant model changes seriously aggravate the performance of model-based control algorithms at all. And as a consequence, precise model tackling in real time should be supplied. Another important characteristic of the MPC concerns its long calculation time and as a consequence - the controller time interval. For this reason, MPC is widely applicabe predominantly for slower and sluggish processes (thermal power stations, technology industries, etc.), no matter the computation power involved. Finally, the static control error is only algorithmically guaranteed implicitly, not in general and theoretically postulated.

One, in my opinion valuable advantage of MPC, is the evident algorithmic simplicity and historical succession: Every novelty is a well forgotten oldie... But, if on-line optimization takes place - it should be paid for! And nobody knows what happens in a non-linear control problem with on-line non-linear programming task on each control step.

The PI algorithm suffers some well-known shortcomings: Except its strong simplicity and bounding linearity (possible, however to be evaded), the lack of D-component makes it to ride the process too gently and too late with high overshoots and long control times. Generally, the system stability margins are clear and theoretically aggravated, but the steady state behaviour is theoretically guaranteed without any error! And last, but not the least, the controller discretization time can be put at arbitrary small value!   As for the tuning procedure, in contrast to the claim of a colleague above, it doesn't need in general any plant model (e.g. Ziegler-Nichols tuning) and could attain any predefined performance, including overdamping. The tuning itself could be performed experimentally in real time even by a dumb, i.e. non-qualified technician or low-level worker. As for the control constraints, they could be set additionally by the aids of the industrial controllers via lot of supporting options to the main control algorithms: integral windup, initializing, PV tracking, etc.

So, in the light of the Project  "Efficiency ImprovEment of the Electric Drives" the simple old PI algorithm is, in my modest opinion, much more suitable to practical applications than the science-heavy MPC, even implemented by advanced digital means.

Dear Emad,

I suggest to you to consider the following book, section 1.4.3;

https://books.google.dz/books?id=cSKsBwAAQBAJ&printsec=frontcover&dq=isbn:1118921569&hl=fr&sa=X&ved=0ahUKEwiPp47Sjp_TAhWJVhoKHWOiD7cQ6AEIGjAA

Good luck.