This is my first post, so I will introduce myself. I am an electrical engineer (MSc, TU Eindhoven) from the Netherlands. I did my master thesis on self-learning, model-based heat pump controls and that’s what started my heat pump obsession. So my background is more engineering/theoretical and I know less about the practical side, which I am learning more about right now and this forum is really helpful. If anyone is interested in my thesis I can share more about it.
Beginning of next year I will start at Intergas, as a research engineer working on heat pump R&D. Specifically I will further develop the work I did during my thesis, and to implement it on the heat pump control unit. To prepare for that a bit I was researching currently available control options.
Feature request: For homeowners that don’t want or need to understand all this stuff, the room influence needs to be more dynamic. More set and forget. A homeowner Auto mode!!
I am curious to hear about any other ‘advanced’ heat pump controller options currently available in the market, good or bad.
Good luck with your initiative, @langestefan.
Just for information, I have to say that I’ve rarely seen the words “Samsung” and “advanced” in the same sentence. I love mine to bits, but you might be well advised to concentrate on other brands…
Don’t forget this vaillant sensoconfort can also control gas and oil boiler… It is a single controller with several parameters and functions to be configured on a case-by-case basis depending on the components of the installation, not only for heat pump.
Mitsubishi heatpumps have “auto adapt” mode, which many users have said works very well. This replaces the need to set the weather compensation curve, and reduces cycling. Anecdotally, heatpumps running with this mode get better performance, particularly if oversized.
Yes good point. I worked with hybrid heat pumps (intergas xtend). It’s nice to have a backup which can correct essentially every situation, but it does make the modelling / control more complicated. Particularly, you need to weigh gas/CO2 cost against heating with a low COP. That decision depends on a lot of variables, including energy and gas cost, flow temps etc.
Our advanced Auto Adaptation Function measures the room temperature and outdoor temperature, calculating the required heating capacity for the room. The flow temperature is automatically controlled according to the required heating capacity, while optimal room temperature is maintained at all times; ensuring the appropriate heating capacity and preventing energy wastage.
Hi Stefan and welcome to the forum! Is your thesis published somewhere?
In my opinion the main challenge is to properly control temperature throughout the entire house and not just based on a single sensor location. I’m on pure weather control (Vaillant) as my controller is in the basement, but I would not know where to put the controller instead. My living room/kitchen gets a lot of solar gains and heat from cooking (open floor plan). If the sensor was there, the rest of the house would be colder than desired. Putting the sensor in other rooms will not capture the heat gains in the living room. The next level in control algorithms for the heat pump would in my opinion be a holistic approach taking individual rooms into account while at the same time ensuring the required minimum flow rate of the heat pump is always respected. Simple TRVs might all close at the same time, so some sort of coordination is needed, also the flow rate of an individual emitter should be limited to not drop below a certain threshold. I’m currently experimenting with variable speed fans under my radiators that allow me to steer room temperature by 1-2°C based on my current needs without having to adjust flow temperature or rate. Smart approaches using the weather forecast for optimized timing of DHW runs or preemptive adjustment of the flow temperature would also be nice. You might even throw presence detection and immediate reaction to in-room electrical use into the mix to directly adjust heating power without having to wait for the temperature to change.
That said, I’m still impressed how well the pure weather compensation works. Having just a bit more granularity and flexibility in there would be great already. A slightly more complex weather curve taking into account different emitter exponents instead of a fixed value woulk probably remove my need to sometimes switch between a curve slope of 0.25 and 0.2.
Thanks! Not yet, but if you give me your email via DM I can send it.
The solar gains you can deal with by taking into account solar radiation at your specific location, by using for example a (near real time) weather forecast. Thanks to AI, super computers and just better science this kind of approach is now feasible. My method already accounts for solar gains, although it is a bit rudimentary with only a single gain parameter which is learned from observed data. You can go even further by considering solar azimuth, zenith, expected cloud cover etc.
A controller which has learned an accurate model is also able to estimate the magnitudes of transient disturbances and account for them. For example rain clouds that pass over and cause more cooling than expected.
I think having the ability to look ahead at the weather 1, or even multiple days could really help in this respect. Weather compensation is great, but it is still only considering the temperature ‘right now’.
It’s my first winter with a heatpump, I choose the inactive mode, I tried the active mode in order to avoid overheating when we turn on the wood fire in the evening, unfortunately the correction in this case is too important and the rooms not taking advantage of the fire cool down too much.
But here in the case of the sun, the whole house suffers it. On December 28th we had sun, I did a manual test and switched it to active mode to measure how house is reacting. I prefer it to expanded mode in order to avoid over-cooling the thermal mass of the building and thus limit the demand during the restart. My goal is to remain as constant as possible in order to avoid spikes in the demands bad for the COP.
In HA indeed, the Forecast entity giving me the Sunny info, I’ve create an automation that switches the mode to active when Sunny is announced, as well as another that switches it back to inactive when Sunny is no longer there.
There’s no real need for fancy prediction algorithms here - I have a solar irradiation sensor and local weather stations also publish current measurement values on a 1-hour grid. If you have a PV install on your roof, the current power can be used as a proxy. Measurements always beat predictions in my experience . All that’s needed is this measurement and a spatial weighting function that encapsulates the house’s response to a specific solar irradiation coming in from a certain azimuth and elevation. Getting this function is going to be the tricky thing though. And not everyone has such a sensor of course, but I would think a simple brightness sensor together with some calibration data would do. With the high-quality 3D GIS data available in large parts of the world, this might be entirely sufficient to build such a model and not depend on any fancy black-box AI but have a model based on first principles that is loaded onto the local controller.
I personally would at no point buy a device where this functionality depends on the manufacturer’s cloud. Heat pumps are very long-lived devices and there are plenty of examples where companies went bancrupt or simply disabled their cloud infrastructure. I’m not paying thousands of Euros for a device that can be partially bricked if a future product manager decides that it’s too expensive to support the 10-year old device.
The problem is not so much live irradiance measurements, the forecasts are more important. For example imagine it’s cold and cloudy in the morning, and the heat pump is running continuously. In the afternoon the weather clears up, but since you have no forecast available you are not aware of this. You end up overshooting the temperature set point, while you could have lowered the flow temperature earlier in the day.
Yes I agree, it’s risky to run everything in the cloud so that is not the plan at all. I am also a big proponent of keeping everything local (I run home assistant myself for example), and that is also the intention for this feature. However, to make this work we have to get a weather forecast into the device one way or another. Bricking the device is not possible, in absence of a weather forecast you would simply fall back on simpler algorithms.
I used to run my own heat pump control algorithm, which would factor in the weather forecast as well as current conditions inside and out, plus solar generation. So, if warmer weather was coming, the system would hold back a little. If it was likely to be getting colder, it would put more heat into the home. If there was sun shining on the panels, it would try to use the free energy, rather than assume any solar gain.
Trouble is the setup was janky, and susceptible to failure if any of the internal or external data sources went offline. For example, temperature sensors in every room were requiring frequent battery changes.
I’ve since switched to Mitusubishi’s own auto adaption algorithm, and enjoyed much improved comfort and performance (SCOP of 3.5 raised to 3.8 over a year). More importantly, the heating still works when the internet is down, and can still be controlled locally.
I wonder how much this can really be leveraged. Current indoor temperature and required heating power only depend on the past and current conditions, so preemptively lowering the flow temperature will lead to a room temperature decrease below the set point; how strong this is depends on the emitter types and thermal inertia of the house. Solar effects are immediate, so if I model my solar input as a step function where I suddenly gain 1kW of solar heat power into my building, I need to decrease heat output of the heat pump by this amount to stay in the steady state. Solar heating of my external house envelope that propagates into the building with a delay can be considered for future flow temperatures, but again, this is the future depending on the past. At least for my house I see very sudden heating by the sun in early spring, but I personally would not like to preempt a 1°C unwanted rise by a corresponding drop beforehand, but that’s just personal preference. The question is of course what the goal is - lowered electricity use, constant temperature or something entirely different?
Solar gains are indeed difficult to model. I have also considered leaving them out entirely, but that would leave performance on the table for houses which are strongly affected by it (large windows on south).
I think for most people it would be comfortable to not have a drop at all. By utilizing the inertia of the building you can avoid the drop, and minimize the overshoot.
That depends on what is more important for you. You can capture this in a multi-objective optimization program, which you will find in the thesis I sent you. I think what would really help is to make this visible in an app. For example you can show heating programs / plans, expected cost, and (perceived) comfort level to help you. I think most people would perfer an ‘auto’ mode that ‘just’ works . I am an engineer however and not a usability / UX designer so I’ll leave that part up to the experts…
I’ve recently installed a Homely Controller, which I’m sure you will have heard of. It sounds very similar to what you are trying to accomplish. Some more competition and knowledge in this space will be great to drive and improve the standards of controls.
Importantly, the Homely also uses time-of-use tariff data to try to minimise operating costs and I think it’s quite good at this, but theres certainly room for improvement.
. All that’s needed is this measurement and a spatial weighting function that encapsulates the house’s response to a specific solar irradiation coming in from a certain azimuth and elevation. Getting this function is going to be the tricky thing though. And not everyone has such a sensor of course, but I would think a simple brightness sensor together with some calibration data would do. With the high-quality 3D GIS data available in large parts of the world, this might be entirely sufficient to build such a model and not depend on any fancy black-box AI but have a model based on first principles that is loaded onto the local controller.
. I am an engineer however and not a usability / UX designer so I’ll leave that part up to the experts…