colinl
commented
3 years ago Why do you use 0.5 as an offset? I seems like a virtual baseline or virtual "zero"?
Because the maths is symetric about zero the power it calculates has a centre point of zero, so if we wanted the scale the output to have a span of 1 then the output would be in the range -0.5 to +0.5. As a mathematician I would be perfectly happy with that, but for practical usage, such as heating, the output is more convenient in the range 0 to 1, hence the addition of 0.5.
In other words. In a heating environment, I always need a negative integral to keep the PV at setpoint, but I never need a positiv integral to keep the PV. Or am I wrong?
Your logic may well make some sense, however what you have to remember is that PID is a very simple way of attempting to take the non-linearities out of a complex real world process to make it more easy to control. As soon as you start adding complexity to the algorithm, such as clamping the integral, if that is what you are suggesting, then you are adding non-linearities back in. Very often tweaks like that have unforeseen consequences. By all means adjust the algorithm and see if it helps for your process, it may well improve the control marginally, or it may not. You might even find that the system goes unstable in unexpected ways. It is virtually impossible to know what effect it might have in your process. Another problem is that if you consider a cooling process such as a refrigeration system then the opposite is true of the integral, so the clamp would have to work the other way. Even worse would be an exothermic chemical process. They commonly that requires heating to get up to temperature and then cooling once the reaction starts, so the integral may be anywhere in the range.
if PV approaches setpoint from below, t_derivative should be larger, than if PV approaches setpoint from above.
Again, something which appears to be entirely logical can have often have unforeseen side effects as it is introducing non-linearities into the system. By all means try it, but I shall be surprised if it makes much difference, and could easily make the control worse. In this particular case consider what will happen as the pv crosses the setpoint. At that point the derivative would jump from one value to another which would cause the power output to step from one value to another, which would not be good. What would you do about that? Possibly you could adjust the I term as it crossed in order to take out the step. But then what would happen if the Integral happened to be against the clamp that you suggested earlier? It wouldn't be able to adjust it. It all just gets so complex that it becomes impossible to predict the consequences under all circumstances. Believe me, I have tried such tweaks on control systems over a period of 50 years in process control and they are rarely helpful.
The most important point is to make sure that you have tuned the loop optimally before embarking on any such exploits. If you wanted me comment on whether I think the tuning could be improved then post some plots showing process value, setpoint and power output and I will happily take a look. Tell me a bit about the process too.
An unrelated question, what resolution have the sensors that you are measuring temperature with got? If you are using, for example, DS18B20 they have a resolution of 0.0625C, if my memory is correct. That sounds better than is needed for room temperature control, but it can play havoc with derivative. The problem is that you get long periods with no change in temperature at all and then it steps by 0.0625 over one sample. This causes great difficulty getting a good estimate of the rate of change of pv and very much limits the amount of derivative you can use, so you may well find that you can't use as much as you would like. If you are using a much better sensor then that becomes less of a problem.
Mahai17
juangburgos
Hi, I use this node for room heating (P and I only, planning for D). I am happy so far, but I have some questions. I looked into your code to better understand the algorithm. 1: the 0.5 offset power = -1.0/node.prop_band * (proportional + node.integral + node.derivative) + 0.5 Why do you use 0.5 as an offset? I seems like a virtual baseline or virtual "zero"?
2: positiv integral Maybe I am wrong, but I think for a heating application, the integral should not get positiv. Because, a heating can't actively accelerate the decrease of PV. Example: PV=21°, setpoint=21°, integral=-0.2, system (room temperature) is stable. Now I reduce setpoint to 20°. Power=0, and integral increases over time. Integral becomes even positiv. Some times later, PV reaches setpoint and still decreasing. Now comes the strange thing: Even PV falls below setpoint, the Power output remains 0, because of a positiv integral. In other words. In a heating environment, I always need a negative integral to keep the PV at setpoint, but I never need a positiv integral to keep the PV. Or am I wrong?
3: Feature request: Different settings for derivative time In a heating application, it would be nice to have two different derivative time settings. I mean, if PV approaches setpoint from below, t_derivative should be larger, than if PV approaches setpoint from above. In other words, opening the valve in a radiator immediately adding a lot of energy to the "system", t_derivative should be small. But after closing the valve, there is still energy emitted, t_derivative should be large.