The IoT topic is 100% worth exploring more because it’s difficult to see why ASHPs wouldn’t connect to everything else around them to support them in their heating quest. Homely, as a product, is trying to do that. The biggest issue is probably going to be underlying OSs for the various manufacturers that control units and inter-communication issues that may cause. 

I've been looking at this subject for the past 4 years from the 'other side of the fence' - the regional grids that supply the electricity for our heat-pumps. There are significant benefits to be gained by integrating HP controls with forecast data on local renewable-energy generation. 

I guess a lot of this comes down to legalities ie manufacturing warranties and insurance.

Posted by: @knukes

↑

@cathoderay You’d be surprised at how often machine learning is now being used in real life. Sectors including manufacturing, healthcare, financial services and others are using it to find patterns in the data and develop new solutions based on that. 

I’ve been involved in quite a few projects myself implementing it. 

While I’m not an expert on algorithms, I can’t see why AI wouldn’t be able to see patterns in the data in your house and outside (assuming you have the right sensors), and make adjustments based on that to give you consistent temperature and efficiency. 

Is it that the machines are actually 'learning' and re-writing some of their software, and making intuitive decisions, or are the machines merely making decisions based upon pre-programmed criteria?

Can a supermarket AI computer system, when seeing that there has been a sudden surge in the sale of 'wiggits', so would normally be programmed to double the order for 'wiggits' from their supplier, but instead decide not to do so because the surge in 'wiggit' sales has been due to two weeks of exceptional weather, which may not occur again for several years, hence leaving the supermarket warehouse full of unsold 'wiggits'?

In their most basic form controllers are really quite simple, they receive two inputs, say A and B, and produce one output, say C. The controller then performs a fairly simple mathematical calculation, if A is greater than B, it may be configured to increase C, and vice versa. The degree by which C is increased or reduced will depend upon the difference in magnitude between A and B, the error signal, the objective being to reduce the error to zero. Obviously in the real World there can be many other factors to be taken into account, such as how quickly the system responds, and is the response linear.

Controllers have had the ability to self tune for many years now, and even the cheaper ones can have this ability, so self tuning algorithms are readily available, but may or may not have been incorporated within heat pump controllers. Having the ability to self tune does not mean that a controller will perform any better, if the equipment that it is controlling is not up to scratch, the heating system is not correctly designed and installed, and the allowed adjustments within the controller are adequate for the particular system. To maintain a constant indoor temperature in a home with a large thermal mass would require a controller whose response is much slower than that of a controller used in a home with a low thermal mass. This is only one of the problems that heat pump designers need to consider.

Posted by: @cathoderay

↑

even the most sophisticated heating control system is all said and done a feedback system, possibly even an adaptive feedback system. Such systems can be extremely sophisticated [...] but they don't involve intelligence, artificial or otherwise,

I disagree. A feedback system is, by definition, operating in the present time.

The AI which would be most welcome for controlling HPs (and Storage Batteries) is that which receives forecast data of events that are in the future. The embedded AI then makes control decisions based on both that forecast and 'experience' of similar previous occurrences.

Consider the two following illustrations:

1: You have a ToU tariff, such as Octopus Agile. The time is 3pm - about 90-mins before the national peak-demand period. The next lower-cost time-slot isn't until 10pm.

If the Smart HP Controller knows the forecast external temperature for the evening, it can calculate the anticipated heat-demand which the HP must supply for the next 7 hours.

The options are:

a. Raise the room temperatures (and that of the in-house heat-store) a couple of degrees higher than the current set point whilst electricity is cheaper before 4:30pm. At 4:30, reduce the flow temperature to prevent the HP coming on whilst the price is high.

b. Use the immersion heater in the thermal-store to raise its temperature higher. This is fast, but operates at a COP of 1 (!). Even so, this may be cheaper than using more electricity before 10pm.

c. Lower the set-point of the in-home room thermostats by a degree or two. This will allow the room temperatures to drop further before the heat-pump starts up. Restore the thermostat settings at 10pm.

And any combination of the three of course.

2: It is 7am and the Heat-pump starts raising room temperatures towards 20degC.

Weather Forecast data shows that cloud cover will disperse, with bright sunshine arriving at 9am.

Historic data within the Smart Controller indicates that the heat-pump doen't draw power from the grid in full sunshine. Passive solar gain, plus electricity from the roof-top solar panels is more than sufficient to keep the house at the set-temp, given the external air temperature forecast for the morning.

The options are:

a. Lower the set-point of the in-home room thermostats by a degree or two. The house will remain slightly colder than the user has set, but it avoids already having a warm house when the sun comes out at 9am. The energy deficit in the home can then be restored.

b. Divert HP output away from DHW. Send an alert to the home owner to avoid using the shower until after 9.15am. Provide an on-screen accept/reject for that option in case they want to use the shower before then anyway.

I call this strategy Hold-Off. It uses AI to hold off from implementing the preset configuration until conditions are more favourable, and then reverts to using those settings.

I must point out that the two scenarios above are theoretical. I have had no communication with @heacol and therefore no knowledge of what features may or may not be in consideration for his forthcoming controller!

@transparent - fascinating, but I am not sure I want a smart phone telling when to have a shower! Next thing will be AI pong sensors, and then the AI will forecast when I will get a bit smelly, by combining historical data for when I got a bit smelly in the past with current forecast for temps later today, and then tell me I need to have a shower at 1130 hours.

Seriously, though, the AI you describe isn't using actual future data (an impossibility, at least for now), is it using modelled forecasts of the future. Furthermore, those forecasts are in fact based on historical data. This I suggest means that it is in fact operating in present time. Very simply, for the second example, it says: at the present time, I predict a rise in solar power/gain in 2 hours time, so I am going to drop the room stat set point now. It's doing a present time response to a current prediction of a future event. 

In my view, AI should work for you to make your life easier, more comfortable and cheaper. As soon as it starts dictating to me, it goes in the bin.

But it is still AI based (machine learning) @cathoderay . It's not merely using a feedback loop.

I should also point out that the two illustrations I've given are based on current  tariff structures and features of our Smart Meters.

Once we have nodal (locational) pricing and our Smart Meters are used to the full extent of their available functionality, then some even more interesting energy/cost-saving measures become possible.

I find it deeply annoying that our Energy Suppliers seem only interested in using Smart Meters to read consumption data. The specification (2013) describes far more useful features, which would actually benefit the consumer rather than the Supplier.

They won't progress further until Ofgem instructs them to do so.

Artificial intelligence (AI) and machine learning (ML) can be used to calculate weather compensation for heat pumps by using data from weather forecasting systems to optimise the performance of the heat pump.

The efficiency of a heat pump is directly related to the temperature difference between the heat source and the heat sink. In cold weather, the heat source is the outdoor air and the heat sink is the indoor space being heated. In warm weather, the heat source is the indoor space being cooled and the heat sink is the outdoor air.

Weather compensation systems use sensors to measure the temperature of the heat source and the heat sink and use this information to adjust the performance of the heat pump. AI and ML can be used to analyse this data and make more accurate predictions about the efficiency of the heat pump based on the current weather conditions. This can help to optimise the performance of the heat pump and reduce energy consumption.

There are a number of different algorithms and machine learning models that can be used to calculate weather compensation for heat pumps, including linear regression, decision trees and neural networks. These models can be trained on historical data from weather forecasting systems and used to make predictions about the performance of the heat pump based on the current weather conditions.

The true benefit will be for the system to know (and learn) how quickly our house loses heat so that it can prepare (and preheat) the house using electricity when it's cheaper. That would be ideal, but I suspect we're still quite a way from that being implemented and being a reality.

Posted by: @editor

↑

Weather compensation systems use sensors to measure the temperature of the heat source and the heat sink

From what I've read in the Midea manual (see posts passim), in heating mode weather comp only measures the temp of the heat source, ie the ambient, there is no way of adding feedback about the temp of the heat sink ie the house/room temps. The system can be run using the room temp as the feedback signal, if a suitable sensor is attached to the system, but doing so turns off weather comp, and runs the heat pump on a simple on/off fossil fuel type basis.

The inability to fine tune the weather comp curve depending on the house temp is a major headache, as I am currently being reminded: the warmer outside ambients mean the heat pump is only putting out the heat to match losses at these higher ambients, but the house is still cold from the cold spell, ie it is as if it was sitting in a lower ambient, and so is taking forever to warm up, about 1 degree per day at the moment. If the control system knew the house was actually colder than it should be at these ambients, it could boost the weather comp curve briefly to make up the shortfall. Only one extra sensor, and I should imagine a simple job to do in control software/circuits, but it's just not there.

I wrote a rather long essay about AI and machine learning, but I am not sure I am going to publish it just yet, if ever!

Edit: thinking about the above, and this has come up before, the heat pump does perhaps have some feedback about the house temp, in the form of the RWT. But it doesn't appear to have any way of using that info to say give the weather comp curve a boost because it is colder than it should at this point in the cycle.    

@cathoderay, interesting. Most installers have also said it’s the outdoor sensor that’s only needed for weather compensation to work. Makes sense. But what doesn’t make sense is that if you have a super insulated house and it’s cold outside, the weather compensation instructs the heat pump to increase the flow temperature and just run indefinitely continuously heating the inside space - what happens when your target temperature is exceeded? Surely the heat pump should then stop, so the internal sensor should play a role too, otherwise the temperature will just keep rising.

But there’s nothing simple about the simplicity of weather compensation.

Posted by: @cathoderay

↑

Posted by: @editor

↑

Weather compensation systems use sensors to measure the temperature of the heat source and the heat sink

From what I've read in the Midea manual (see posts passim), in heating mode weather comp only measures the temp of the heat source, ie the ambient, there is no way of adding feedback about the temp of the heat sink ie the house/room temps. The system can be run using the room temp as the feedback signal, if a suitable sensor is attached to the system, but doing so turns off weather comp, and runs the heat pump on a simple on/off fossil fuel type basis.

The inability to fine tune the weather comp curve depending on the house temp is a major headache, as I am currently being reminded: the warmer outside ambients mean the heat pump is only putting out the heat to match losses at these higher ambients, but the house is still cold from the cold spell, ie it is as if it was sitting in a lower ambient, and so is taking forever to warm up, about 1 degree per day at the moment. If the control system knew the house was actually colder than it should be at these ambients, it could boost the weather comp curve briefly to make up the shortfall. Only one extra sensor, and I should imagine a simple job to do in control software/circuits, but it's just not there.

I wrote a rather long essay about AI and machine learning, but I am not sure I am going to publish it just yet, if ever!

Edit: thinking about the above, and this has come up before, the heat pump does perhaps have some feedback about the house temp, in the form of the RWT. But it doesn't appear to have any way of using that info to say give the weather comp curve a boost because it is colder than it should at this point in the cycle.    

I suspect that your cold house is not actually the fault of your controller. If your system is working correctly, and your weather compensation curve is correctly adjusted, then your home should maintain the desired temperature even as the outside air temperature varies.

Does the Midea not have a temperature offset to vary the LWT when in WC mode?