I just have a basic doubt regarding how the limits of various PI controllers used for the design of current control technique of the grid connected VSI is being selected.
I feel that applying limits to the PI controllers is a critical aspect, but I am not sure how the values is to be kept for simulation (I have attached the paper titled "Single- and Two-Stage Inverter-Based Grid-Connected Photo voltaic Power Plants With Ride-Through Capability Under Grid Faults" kindly refer to Fig. 4 of the paper and the line highlighted in green.)
Also find attached the snapshot of the Simulink model made by me in the page-2 of the second attachment titled "pid_simulink_limit"
You can tune your PI controller using Ziegler-Nichols method ( Refer to Modern Control Engineering by K. Ogata). Based on these values, you can set minimum and maximum values for KP and KI.
I disagree for some things..
In fact Ziegler-Nichols is completely rubbish for today people that use MATLAB.
Z-N determined the parameters on a system with a global gain, a PI pole and PD pole. whereas MATLAB takes three different gains in their model, P, I, D, so it needs totally different parameters. I looked at the original Ziegler Nichols paper, they omitted to even describe the transfer function they were using. It results in a lot of errors fro people that try to apply it. For integrating systems with some dead time, if you do not want much overshoot, simply take 40° phase margin in Nyquist. In resonant systems such as filters, a D-action is required, and even measures to damp the Q-factor. Limitations: one has to avoid wiind-up of integrators, so the best is to limit the value of the integrator as much as possible, as it can cumulate errors when the control saturates, e.g; by a limited voltage or duty ratio.
I find myself agreeing with Mr. Ajanovic since, it looks like the problem is not the tuning of the PI controller. It's rather setting the saturation limits, which also in my opinion depends on the conditions of the circuit that the controller has to work in. Please do correct me if I'm wrong.
Hi Amit,
I must say that Alex has introduced some useful points over and above the conventional wisdom of ZN-rules.
However if I have interpreted your question correctly, you primarily wish to discuss the choice of limits with PI, and that too for a specific application. This is generally based on two considerations (for any application):
For current control of grid connected VSI, I would suggest the following course:
- Simulate the system and controllers adequately on a suitable platform, such as the PSCAD/EMTDC. Include all limits due to hardware, since those are any way unavoidable.
- Work with the dynamics of the system exhaustively, including the nonlinear dynamics that are eventually to be avoided. Set the limits (in stages), until all such dynamics are eliminated.
Though the second step seems to be a trial-and-error approach, you will be surprised by the speed with which you converge to your best choice of control limits. Just be systematic about the procedure, and you won't go wrong.
-Sanjay
Sanjoy Roy
What you propose is what is typically done nowadays, and for simple systems it works, skilled people do it even without simulation.
But if 5 parameters, P, I, D, low pass in D and saturation have to be chosen, there is a big chance that trial and error does not converge to the right solution. If there are resonances in filters or a DC-link with small film capacitors that add to the problem, the chances to find a solution by trial and error get very low. In practice also noise, disturbances and non-linearities can be present.
A typical non-linearity is the saturation of a controller (wind-up) but also that a control jumps from 95% duty to 100% duty ratio.
The Nyquist (or Bode) is an independent way to know if one is far away or not from what could be obtained.
My experience is that even a lot of systems get unstable at half of the switching frequency rather than the resonance frequencies.
A typical problem is that Matlab implements PWM without delay, which is true for a pure analog PWM. Most of people use that model for a digital system where there is definitely a delay between sampling and PWM compare levels.
In drives, the wind-up is eliminated by resetting the speed integrator when the torque is limiting during acceleration. If this is not done a lot of error is cumulated and will result in a big overshoot in speed.
Wind-up is one aspect, but there are many reasons why simulations do not correspond to the reality.
In theory the practice is equal to the theory, however, in practice it is not.
Alex
- Badges
- Science topic