Evohome 'Optimum Start' Switching on too Early

[Zildjian]

We have a honeywell evohome system and the master bedroom is set to heat up by 10:30pm each evening. I have recently switched on optimum start after not using it for some time and it's consistently starting to heat the room a full hour early, despite it only taking 20-30 minutes to reach the required temperature. It doesn't overshoot but sits at temperature for 30+ minutes before it's needed.

From my understanding, optimum start is supposed to be a learning system, yet I've waited a week and it's failed to try starting later than a full 1 hour sooner at all so far. It should only need several days to learn the parameters, or even just a single day if it also monitors how fast the temperature drops and knows the outside temp. Should I just be more patient, or do I perhaps need to reset the system, and if so how?

Thanks

[g6ejd]

I have not seen how long the Evohome system takes to learn room characteristics and let's assume it is a room and not all zones/the whole house.

Do you have the means of measuring temperature in the room, when you say it has achieved the required temperature, do you mean at the HR92 or at some other place?

I used to have the Nest thermostat and that would start heating the property some 2-3 hours earlier than the scheduled time, it varied daily depending on the outside temperature, but on that basis, the Evohome is much better!

A full reset will clear the values learnt by the Optimum Start, but so may disabling and re-enabling it again, but how do we know? AFAIK there is nothing in the manual to tell us. You could try disabling it, then try the next night and see if it does the same.

Is it always 1-hour or does it vary with temperature?

[paulockenden]

Optimum start does vary, but it's always optimistic. It always makes sure the room WILL be at the right temp at the right time, even if there happens to be a really cold snap. I'm not overly fussed by it coming on a little bit early. At least the comfort levels are maintained. And any extra gas is compensated by optimum stop.

[G4RHL]

Reading this, this morning has alerted me to the fact that optimal start in my case does seem to have learned but it had not dawned on me. I was a constant grumbler in these pages that optimal start seemed a waste. In my case also it was always coming on 60 minutes before needed and the room could be up to temperature in 10 or 15 minutes. I have though left it alone, which is the best thing to do with Evohome once set up, let it do its stuff, stop playing and it works really well. For the past few months optimal start has reduced its switch on time to around 30 minutes or slightly less. It may of course be because it knows it has been summer, but it has certainly learned, albeit it seems to have taken several months to do that. As the days get colder no doubt I will see whether what I am saying is true!

[PaulB]

Over the past 12 months of having mine installed, initially it was starting to heat room way before it needed to, but over time seems to have settled down & now comes on just at the right time to hit the setpoint. Same with constant temperature control, used to suffer with some serious overshoot problems, but again over the past year, this has pretty much eliminated itself & no longer an issue.

[G4RHL]

Over the past 12 months of having mine installed, initially it was starting to heat room way before it needed to, but over time seems to have settled down & now comes on just at the right time to hit the setpoint. Same with constant temperature control, used to suffer with some serious overshoot problems, but again over the past year, this has pretty much eliminated itself & no longer an issue.

I forgot, yes, "constant temperature control". This is fine now, virtually no overshooting like it used to do. However, looking back, it has taken some months to learn and adjust. The more you fiddle with it, the more it has to start learning again!

[DBMandrake]

We have a honeywell evohome system and the master bedroom is set to heat up by 10:30pm each evening. I have recently switched on optimum start after not using it for some time and it's consistently starting to heat the room a full hour early, despite it only taking 20-30 minutes to reach the required temperature. It doesn't overshoot but sits at temperature for 30+ minutes before it's needed.

From my understanding, optimum start is supposed to be a learning system, yet I've waited a week and it's failed to try starting later than a full 1 hour sooner at all so far. It should only need several days to learn the parameters, or even just a single day if it also monitors how fast the temperature drops and knows the outside temp. Should I just be more patient, or do I perhaps need to reset the system, and if so how?

Well, it doesn't actually know the outside temperature. The Wi-Fi controller doesn't support an outside temperature sensor at all, and Honeywell staff have advised on here previously that even the older Evohome colour (pre-Wifi) that did support the outdoor sensor used it for "informational purposes" only - eg it displayed it on the status bar but did not use it in any way to help with optimal start or anything else.

So the only feedback optimal start has to learn from is the temperature measurement of the room and timing how long it takes to heat up.

As far as we know it uses two factors - initial temperature, and the rate at which the room warms up to decide how soon to turn on.

In my experience the adjustment of startup time for differences in initial temperature has always worked well - the colder the room is to begin with the sooner it will come on.

However adaption to the rate that your room heats up seems a lot slower and a bit more spotty. Many people including myself had problems early on where some rooms would come on far too soon, and day after day the room might be up to temperature as much as an hour early with no apparent attempt to remedy this.

So I feel your frustration. Mine no longer does this and I find that optimal start generally hits the target about 15 minutes early and does seem to adapt over time, and the frustrating thing is I have no idea why it works now and didn't work very well for me at first. I have tinkered so much with my system that I don't know what might or might not have helped.

One thing I would suggest though, is make sure that the room that is having the issue is actually able to comfortably meet a target of a couple of degrees higher than your set point without the radiator struggling to get there. (which could be a result of too small a radiator, too low a flow temperature, a draughty door being left open etc)

For example if you have a set point of 20 degrees and it turns out that your radiator is only big enough to just barely get the room to 20 then optimal start will tend to massively over-estimate the required lead time as it might be struggling to reach the last fraction of a degree. Because it rounds the temperature and also biases it towards the set point you may not realise this. For example it might say it has reached 20 degrees but in reality the temperature is only 19.4 with the radiator running flat out and it is struggling to get any higher. This will cause optimal start to keep coming on early in an attempt to hit the target.

It may be helpful to set up graphing of your room temperatures to see if this is happening - if the radiator is struggling you'll see a wide arching curve to the temperature increase that doesn't quite or barely reaches the target, compared to a straighter rapid rise with a small knee at the top which is what you want to see if your radiator has enough oomph. Graphing temperature and set point also makes it very easy to see exactly when optimal start is bringing each zone on.

You could try dropping the target set point by 1 degree or even 2 and see how early it comes on then. If it is the radiator struggling to meet your original target this will make a big difference to how early optimal start starts heating the zone.

[Zildjian]

Thanks for the replies and suggestions. Seems others have experienced the same and I should just be more patient indeed.

Do you have the means of measuring temperature in the room, when you say it has achieved the required temperature, do you mean at the HR92 or at some other place?
...
Is it always 1-hour or does it vary with temperature?

Temp is based on the HR92, and it's consistently 1 hour earlier so far (could be a few minutes less haven't been watching at 9:30 precisely)

Well, it doesn't actually know the outside temperature.

Good point, I was confusing it with my Loop energy monitor system which doesn't have a sensor but uses online weather data for the postcode.

One thing I would suggest though, is make sure that the room that is having the issue is actually able to comfortably meet a target of a couple of degrees higher than your set point without the radiator struggling to get there.

I could see that being an issue generally, though in my case the room is well insulated and heats very fast, it's also the only room with heating on in the late evening apart from the loungeroom.


It's reassuring to see that most peoples systems adapt after some time, still I think the system should be able to learn quicker. A typical machine learning system would start with big jumps, overshoot (i.e. start heating too late the next day), and then change direction moving in smaller steps until it zeroed in, which is why I'm puzzled that it's still starting the full allowed 1-hour early. They could also make the system more analytical by bringing in various other measurements into play, for example to estimate external temperature in relative terms by looking at the rate of temperature drop in other rooms in between heating cycles, and in the room itself during overnight heat loss. Perhaps these are not factored in, or maybe the system is deliberately designed to learn very slowly so as not to jump around too much which might put a lot of people off.

[DBMandrake]

It's reassuring to see that most peoples systems adapt after some time, still I think the system should be able to learn quicker. A typical machine learning system would start with big jumps, overshoot (i.e. start heating too late the next day), and then change direction moving in smaller steps until it zeroed in, which is why I'm puzzled that it's still starting the full allowed 1-hour early. They could also make the system more analytical by bringing in various other measurements into play, for example to estimate external temperature in relative terms by looking at the rate of temperature drop in other rooms in between heating cycles, and in the room itself during overnight heat loss. Perhaps these are not factored in, or maybe the system is deliberately designed to learn very slowly so as not to jump around too much which might put a lot of people off.

I don't think anyone (other than Honeywell) really knows exactly how optimal start adaption works and what factors it is taking into account, and Honeywell aren't telling.

People like me have experimented with it, set up scenarios to try to fool it, monitored it closely (graphing set points and measured temperature of zones to 5 minute intervals and then analysing the graph data) and speculated a lot but at the end of the day we don't really know how the adaption to rate of temperature rise is handled and whether there is an initial learning period or whether it is constantly adapting, and whether it is a gradual process or whether it has to be consistently early or late for some time before it will make an adjustment.

My impression is that it starts with an initial estimate of about 3 degrees per hour rise in temperature with the zone heating "flat out", as in a room I had that heated at about 6 degrees per hour it initially came on way too early, while another room that only heated at about 2 degrees per hour initially came on too late and kept missing its time target. Both these rooms got to their targets early and late respectively for quite some time. But eventually did sort themselves out.

Perhaps it deliberately avoids day by day changes (for rate of temperature rise adaption) and does weekly or even monthly trending before taking action. I don't know. I'm pretty sure there is at least one bug in the adaption to rate of temperature rise that causes it to sometimes not adapt but Honeywell have said they're not aware of any bugs in Optimal Start. (but would you expect them to admit to any bugs ? )

One thing it does do immediately though is the adaption to initial starting temperature is constantly recomputed, so as the temperature falls in the night if the temperature falls further than the previous night it will start heating earlier that very same morning. It's only the adaption to change in rate of temperature rise during heating (for example adaption to manual flow temperature adjustment) that seems slow and inscrutable.

If you figure it out or your system comes right let us know!

[bruce_miranda]

And here is where I miss the per room optimisation that was on the previous non Wifi model. There are some rooms where optimisation just isn't useful or helpful. So I wish Honeywell bring this feature back and allow the user to choose which ones should have it.

[DBMandrake]

And here is where I miss the per room optimisation that was on the previous non Wifi model. There are some rooms where optimisation just isn't useful or helpful. So I wish Honeywell bring this feature back and allow the user to choose which ones should have it.

Agreed with per room being desirable.

I toyed with Optimal Stop being enabled again recently but had to disable it again because it causes problems in the bathroom. The bathroom is scheduled for 22 degrees during shower time in the morning and for about half an hour afterwards to dry towels and moisture out of the room with the window left open slightly. However the room warming up above 22 due to the shower running causes Optimal Stop to shut the radiator off an hour earlier than I have set, thus the towels and room don't get dried out afterwards.

With Optimal Stop turned off the radiator does throttle back while the shower is in progress to maintain 22 then afterwards when the window is open and the room cools it comes back on to maintain 22 and does dry out the towels and the room. I'd like to turn off Optimal stop in one or two rooms like the bathroom that it conflicts with but I'm forced to turn it off everywhere. It works just fine in other locations like the hallway and living room.

[DBMandrake]

So I thought I'd have a quick look at my graphs to see how optimal start is doing for me lately. I did increase my flow temperature from 65 to 70 about a week ago - which usually takes a few days for it to adapt to, and the last couple of days have been very cold at night. (sub zero) Here's the results:

Bathroom: Scheduled to 5 at night, 22 at 6am. It turned on at 4:40am at 14.5 degrees and reached 22 degrees bang on 6am.

Hallway: Scheduled to 5 at night, 20 at 5:30am. It turned on at 3:50am at 17.5 degrees and reached 20 at 5:40am (slightly late) although it was within a small fraction of a degree at 5:30am.

Kitchen: Scheduled to 5 at night, 20 at 5:30am. It turned on at 4:50am at 17 degrees and reached 20 at 5:25am.

Living Room: Scheduled to 5 at night, 20 at 5:30am. It turned on at 3:40am at 17.5 degrees and reached 20 at 5:05am - 25 minutes early.

Main Bedroom: Scheduled to 14 at night, 18 at 6:30am. It turned on at 5:30am at 15 degrees and reached 18 bang on 6:30am. (Rather clever given that the window was also open a bit)

Guest Bedroom: Scheduled to 14 at night, 18 at 7:00am. It turned on at 6:05am at 14 degrees and reached 18 at 6:45am - 15 minutes early.

Of the zones, the Living room and Main Bedroom have remote wall mounted DTS92's, the Hallway uses the Evotouch sensor on the wall mount and all other rooms use the HR92 built in sensors.

So that's some real world data of the system having settled down, especially when I did have problems early on with some rooms like the living room coming on more than an hour early and seemingly refusing to adapt to this. As I said earlier, I don't know what I might have done to solve that or whether it did just take a long time to learn but I'm very happy with the results above - this is the first time I've checked up on it in detail in quite a while. I now just trust Optimal Start.

[bruce_miranda]

How are you tracking that it has started? Is it just by the sharp increase in temperature?

[DBMandrake]

How are you tracking that it has started? Is it just by the sharp increase in temperature?

I graph the set points on the same graph. So I see the set point jump vertically up from 5 to 20. (Or whatever)

[DBMandrake]

Well, a bit of an optimal start mystery here.

Last night I reprogrammed our weekly schedules - for the last 6 months someone has been home during the day so most downstairs rooms have been on during a week day including the living room. Now that has changed and the house is empty again during the day Monday to Friday so all rooms are scheduled off between about 7am and 6:30pm.

However additionally, the living room is now not scheduled to come on at all in the mornings (we don't need the living room hot in the morning before work!) and Optimal Start has gone completely whacko in the living room today...

The original living room week day schedule was:

05:30am - 20 degrees
05:30pm - 21 degrees
11:00pm - 5 degrees

Optimal start is set to a maximum of 3 hours but this is never approached - as per a previous post the living room was hitting its target a little bit early but not much:

Living Room: Scheduled to 5 at night, 20 at 5:30am. It turned on at 3:40am at 17.5 degrees and reached 20 at 5:05am - 25 minutes early.

So in this case it came on 1 hour 50 minutes before the set time which still got the room up to temperature 25 minutes early. (Acceptable I think)

Now the schedule looks like:

06:50am - 5 degrees (to turn it back off in case it is manually overridden in the morning)
06:30pm - 21 degrees
11:00pm - 5 degrees

Nice and simple.

Well despite the room only starting at 16 degrees - so not too much colder than the example above, it has turned on exactly 3 hours before the set point - eg the maximum configured amount allowed for optimal start.

No other room has turned on yet despite the fact that all rooms have had their schedules reprogrammed to go off during the middle of the day. The only difference between the living room and the others is that the other zones come on in the morning as well and the living room does not. Also I did actually override the zone to 20 this morning rather than letting it stay off, but tomorrow I will not do that.

I know for a fact that that room is going to be up to temperature at least an hour and a half before the set point is due, the questions are WHY has it come on so early and WILL it learn not to after a few days ?

Clearly reprogramming the schedule has lost what has been learnt for that zone, or, optimal start is learnt individually for every set point, so when a new set point at 6:30pm was created, optimal start has to start from scratch with that set point ?

Watch this space...