Can smart meters give us energy saving advice?

Would you like a smart meter that told you how much you can save by turning down your heating?  Current smart meters can't do this but researchers at the University of Bath reckon that it should be possible, with the aid of some extra sensors for temperatures and CO2 levels. They devised a system which could give you prompts such as 'We have noticed your thermostat is set to 23C. If you reduced the thermostat to 21°C you would save 11 kWh; this is equivalent to £1.43.'  (I think this must be per day and using electricity) [1]. They found that residents in their experiment were very likely to follow this advice. But all these extra sensors will be expensive - can we manage with less?



There are two key features about this advice. One was that it was tailored to the residents' behaviour, so advice was only given when it was deemed relevant (see table below). The second point was that it gave a numeric value to the savings. Some of the residents were given savings in terms of money, others in terms of carbon savings using friendly units such as trees. Either option seemed to be effective.

This is streets away from the typical sort of advice you get. For example the Energy Savings Trust says that turning down your thermostat 1 degree typically saves £80/year. This is obviously an average and says very little about your particular case. It depends on the size of your house and how well insulated it is. Sometimes you see savings as a percentage rather than a value - in my book I said that turning down the thermostat 1 degree would save 8% off your heating. However that depends on how warm you were to start with, and it does not mean 8% off your gas bill, as it only applies to the space heating part, not hot water or cooking (but cooking is usually minor).  Hot water is on average 15% of your gas bill but this is highly variable. Also turning down your thermostat one degree does not necessarily reduce your average temperature by one degree - that depends on your heating schedule and how long it takes for your house to cool down.

So, how did the Bath researchers do it and is the advice accurate?

The system relies heavily on the extra sensors to decide what advice to give. CO2 sensors were used in two ways - to check ventilation levels and to indicate if people were present in the house*.

Recommendation	Sensors	When advice was given.
Turn down your heating to 21°C.	Temperature	If the temperature sensors showed your room was heated to more than 21°C.
Change your timer settings (with specific details about what should be changed, like when it should turn off).	Temperature and CO2 (for presence)	If the temperature sensors showed your room was often being heated when there was no one at home.
Keep windows closed when the heating is on.	CO2	If the CO2 levels were always low (below 800ppm).
Turn off appliances when you are out.	CO2 (for presence) and electricity consumption	If the electricity consumption was high when you were out.

In practice, some of the recommendations were more effective than others. People were quite good at turning down the thermostat but seemed to ignore information about windows and appliances. Probably they weren't sure which appliances needed switching off.

The clever bit is to calculate how much energy you will save. Leaving aside the last one (which is no easier but is an entirely different game) they are all about space heating which means you have two possible approaches. The first is to survey the building as for an Energy Performance Certificate. You measure all the walls and windows and guess what they are made of and build a model of the whole house. The second is to work backwards - by measuring the heat inputs to the house and the temperatures achieved you can infer parameters for a simple model. In either case you can then use the model to simulate the effect of savings.

In practice neither approach is accurate. There are all sorts of difficulties. For example with the survey approach your construction guesses may be not quite right, there may be defects you can't see and you can have little idea of how effective your draught proofing is without actually measuring air tightness. In the second approach, you need a lot of sensors to get an accurate model and you need to measure under a range of conditions. If residents are in the habit of opening and closing windows that adds another level of complexity! The same researchers claim to have shown the backwards approach is possible but their analysis was mostly based on models, with only six real homes included [2]. I have attempted a somewhat similar analysis in a different project (unpublished) and I found that for some homes you get quite consistent results - and others not.

But how accurate does the simulation need to be, to be useful?

A smart meter measures how much energy is used and what this costs.

If you add temperature data from the area of the main thermostat, this is probably a good guide to temperatures in the house and, in particular the hours of heating. So you could fairly easily estimate what the change in average temperature would be for example from reducing the level of the thermostat. In this chart I show what I mean: using actual temperature averaged over December 2015 and January 2016 for a particular house. (The data comes from another project.) The modelled temperature is an approximation to this based on analysis of just these temperatures and external temperatures, inferring when the heating is on from the changes. Then the adjusted temperature shows what happens using the same model assuming the thermostat is set to 21°C. The average temperature drops from 19.8°C down to 18.7°C, so 1.1°C.

Mean daily temperature cycle for Dec/Jan for a house, showing modelled temperature with and without thermostat adjustments. 

Now, to find out what this means in terms of gas savings, we can use the same temperature data plus external temperature from a nearby weather stations and gas use from a smart meter. In this plot each dot is one day, and the X axis shows difference in temperature between inside and outside. There is a lot of scatter because this home does not have consistent habits but there is a significant trend: gas use decreases by 13.6 kWh/day for each degree drop in temperature difference. We can use this to estimate that a drop of 1.1°C would have meant savings of approximately 15 kWh/day, which is about 5% of the total gas use  during the period measured. However, since there is a lot of scatter there is a good deal of uncertainty - we can be 80% certain savings would be between 8 and 19 kWh/day.

Gas use for this house correlating with the difference in temperature between inside and outside. Each dot represents one day in Dec/Jan

Using this approach we don't need to worry about how much gas use is for hot water, unless this a large proportion and it is also strongly affected by the weather.  During Dec/Jan hot water is likely to be a small proportion of overall use anyway. Day to day variation in numbers of baths and lengths of showers, generates some of the scatter in the above chart.

Incidentally, I have not used my own case as the example in this blog post because we are unusually consistent, probably partly because of the house - since we have mechanical ventilation with heat recovery we don't need to open windows at all in winter. Also we don't have a smart meter and I don't have daily readings. However  I can generate monthly data and over that timescale we are very regular in our hot water use. This graph from another post is based on several years worth of monthly data.

Monthly gas use by external temperature for my house from How to make use of your meter readings

In summary then, the Bath researchers have shown that people are much more likely to respond to energy saving advice if it is tailored to them and specific about their savings. It is extremely hard to generate advice with accurate estimates of savings, without lots of sensors in the house on top of a smart meter. However maybe you can get useful results from just one temperature sensor inside (ideally from your actual thermostat) and some readings for external temperature too (which could be a nearby weather station).  This would allow you to say things like 'Your thermostat seems to be set rather high. If you changed the setting to 21°C you would have saved around 15 kWh/day, or £45 over the last 2 months.'

Is this useful? Is it practical? Would it be better to give the uncertainty range or is that unnecessarily confusing? Do tell me what you think.


* CO2 comes from people breathing and also from burning gas for cooking etc. I have found fairly good agreement with PIR sensors using the principle that if no-one is home then levels are either low or decreasing.

[1] How smart do smart meters need to be? (Nataliya Mogles, Ian Walker, Alfonso P. Ramallo-Gonza lez, JeeHang Lee, Sukumar Natarajan, Julian Padget, Elizabeth Gabe-Thomas, Tom Lovett, Gang Ren , Sylwia Hyniewska, Eamonn O'Neill, Rachid Hourizi, David Coley, University of Bath) in Buildings and Environment Volume 125 November 2017

[2] The reliability of inverse modelling for the wide scale characterization of the thermal properties of building (Alfonso P. Ramallo-González, Matthew Brown, Elizabeth Gabe-Thomas, Tom Lovett & David A. Coley) in Journal of Building Performance Simulation DOI: 10.1080/19401493.2016.1273390