Introduction and Project Objective
As the UK attempts to decarbonise heating, load will shift from the gas grid to the electricity grid. Since the efficiency of heat pumps is dependent on many variables, it is important to quantify how installing less efficient heat pumps might affect the grid – what is the consequence of 10 million inefficient heat pumps vs 10 million efficient ones?
The aim of this project was to quantify the potential impact of widespread heat pump adoption on the UK electricity grid. Specifically, the effect of low-quality installations on electricity demand.
Data Source and Initial Considerations
Three datasets feed into this model.
- The heat pump electricity consumption and heat generated for January 2024 were provided in a cleaned dataset by Trystan Lee and are available at https://emoncms.org/heatpumpmonitororg/feed/view
- The output from my previous heat pump electricity usage prediction model. This provides a 10-year dataset of predicted electricity consumption. This dataset was based on average energy consumption from heat pumps monitored on https://heatpumpmonitor.org/
- Weather data provided by https://open-meteo.com/
Data Preprocessing and Analysis
Categorisation and COP Calculation
The January heat pump data was split into performance groups: top 20, top 50, all units, and derived groups (all minus top 20, all minus top 50). These groups represent how well different groups of systems performed over the month of January.
To calculate COP, the heat output from each group was divided by their respective electricity input, for each hour period. The COP data is an hourly time-series dataset.
Temperature Correlation and Curve Fitting
Note: In reality, heat pump COP doesn’t work like this. It is not dependent on just the outside temperature, but other factors like the heat pump flow temperature, the flow rate, etc. Some of those features are encapsulated in the heat pump electricity consumption dataset, but this method is suitable for experimentation..
The hourly COP measurements were matched to their corresponding hourly outside ambient temperature. COP vs Temperature was then plotted and a Generalized logistic function was fitted.
The function is defined as:
Where:
- A: lower asymptote
- K: upper asymptote
- B: growth rate
- v: affects near which asymptote maximum growth occurs
- M: time of maximum growth
A simple best-fit line was chosen over a more complex analysis as the data is somewhat limited. The reason for using the Richards fit was that the tail ends tend toward a zero gradient. This means COP stabilises rather than continues to decay or increase. The consequence of this is that the function won’t drastically over-predict or under-predict electricity load in the next step.
| Category | Average COP |
|---|---|
| top20 | 4.3523 |
| top50 | 4.0429 |
| all | 3.5893 |
| all_minus_top20 | 3.4805 |
| all_minus_top50 | 3.3055 |
These values demonstrate that the Openenergymonitor dataset is skewed towards well-performing heat pumps. This isn’t surprising as people who are monitoring their heat pumps with external monitoring systems are likely to care about improving performance and optimising their installation.
Normalisation and Scaling
The average UK household uses 12,000 kWh of gas a year. That gas used in an 85% efficient boiler would equate to 10,200 kWh (12,000 kWh * 0.85) of heat being delivered to the house. Some gas combi boilers can reach over 90% efficiency; however, they must be in the condensing temperature range to do so and delivering hot water & heating at below 50 degrees C.
Grid Load Estimate
The following process to predict electricity demand is repeated for multiple scenarios, each representing different groups of heat pumps.
- The hourly Coefficient of Performance (COP) for each scenario, for the entire data range, is calculated by applying the Generalised Logistic Function (Richards function) to the ambient outside temperature dataset. This accounts for the varying efficiency of heat pumps at different temperatures.
- The COP for the “all” group of heat pumps is multiplied by the ‘predicted electricity consumption’ that was calculated in my previous work, to calculate a baseline for how much heat should be delivered each hour.
- The heat delivered per hour is normalised across the time frame so that the average heat delivered, yearly, across the decade period is 10,200 kWh.
- The normalised hourly heat output is then divided by the hourly COP for each heat pump group to provide hourly electricity consumption.
The output of that process is shown below.
This chart below shows the same data as the one above, but on a more granular scale.
The all_minus_top50_minus_1 is the all_minus_top50 data, but with all COP values reduced by 1. This serves to show how low COP values dramatically increase electricity usage.
The table below shows the variation in electricity usage for each scenario.
| Scenario | Total Electricity Input (kWh) | Avg Yearly Electricity Input (kWh) | Average COP |
|---|---|---|---|
| top20 | 24,781.52 | 2,477.81 | 4.66 |
| top50 | 26,776.52 | 2,677.29 | 4.37 |
| all | 30,169.55 | 3,016.54 | 3.87 |
| all_minus_top20 | 31,089.37 | 3,108.51 | 3.75 |
| all_minus_top50 | 32,746.76 | 3,274.23 | 3.56 |
| all_minus_top50_minus_1 | 48,228.12 | 4,822.15 | 2.56 |
Results and Analysis
Even within the range of well-performing heat pumps on openenergymonitor, we see that if the installations tend towards lower SCOP vs higher SCOP, very cold days might see a grid load increase of over 4GW for 10 million heat pump installations. While a fairly typical winter’s day might see a 3GW load difference.
Max Load (GW)
| Year | top20 Max (GW) | top50 Max (GW) | all Max (GW) | all_minus_top20 Max (GW) | all_minus_top50 Max (GW) | all_minus_top50_minus_1 Max (GW) |
|---|---|---|---|---|---|---|
| 2014 | 10.61 | 11.33 | 12.80 | 13.07 | 13.99 | 24.28 |
| 2015 | 11.17 | 11.67 | 13.21 | 13.44 | 14.50 | 25.53 |
| 2016 | 10.79 | 11.52 | 13.01 | 13.28 | 14.22 | 24.71 |
| 2017 | 12.03 | 12.42 | 14.06 | 14.29 | 15.48 | 27.37 |
| 2018 | 14.53 | 14.79 | 16.74 | 17.00 | 18.46 | 32.76 |
| 2019 | 12.16 | 12.68 | 14.35 | 14.60 | 15.76 | 27.77 |
| 2020 | 10.60 | 11.27 | 12.74 | 13.00 | 13.94 | 24.29 |
| 2021 | 13.56 | 13.98 | 15.82 | 16.08 | 17.43 | 30.84 |
| 2022 | 13.28 | 13.66 | 15.46 | 15.71 | 17.03 | 30.15 |
| 2023 | 12.19 | 12.71 | 14.38 | 14.63 | 15.80 | 27.84 |
| 2024 | 13.24 | 13.65 | 15.44 | 15.69 | 17.00 | 30.09 |
Cumulative energy requirements showed marked differences across scenarios, with implications for overall generating capacity needs. Peak loads typically occurred during prolonged cold spells, highlighting the need for robust peak management strategies.
Cumulative Load (GWh)
| Year | top20 Cum. (GWh) | top50 Cum. (GWh) | all Cum. (GWh) | all_minus_top20 Cum. (GWh) | all_minus_top50 Cum. (GWh) | all_minus_top50_minus_1 Cum. (GWh) |
|---|---|---|---|---|---|---|
| 2014 | 9365.50 | 10101.09 | 11384.07 | 11732.69 | 12360.80 | 18192.32 |
| 2015 | 24714.51 | 26740.36 | 30125.84 | 31038.27 | 32691.85 | 48894.13 |
| 2016 | 25626.36 | 27845.45 | 31361.11 | 32318.29 | 33998.36 | 51348.53 |
| 2017 | 24765.84 | 26795.14 | 30191.38 | 31093.74 | 32769.18 | 49452.45 |
| 2018 | 26360.84 | 28504.36 | 32118.96 | 33062.90 | 34869.47 | 53269.46 |
| 2019 | 25044.09 | 27173.16 | 30603.47 | 31541.30 | 33184.37 | 49747.56 |
| 2020 | 24317.43 | 26378.20 | 29707.19 | 30626.59 | 32210.26 | 47903.65 |
| 2021 | 26178.96 | 28323.56 | 31914.69 | 32856.16 | 34644.23 | 52779.17 |
| 2022 | 23691.23 | 25610.28 | 28857.15 | 29723.34 | 31324.49 | 47064.02 |
| 2023 | 23464.13 | 25389.69 | 28602.19 | 29468.64 | 31037.11 | 46451.78 |
| 2024 | 13802.83 | 14941.04 | 16829.45 | 17336.87 | 18261.00 | 27434.42 |
Model Limitations
The model has some limitations, including:
- Uncertainty in extreme temperature scenarios due to limited data in these ranges. This is shown in the plot below where the COP values predicted by the model may be less accurate.
- Potential inaccuracies during transitional seasons.
- Assumptions about uniform distribution of heat pump efficiencies across the UK.
- Hot water & space heating are combined.
Future Improvements
Potential enhancements to the model include:
- Incorporating geographical variation in heat pump adoption and performance.
- Refining the model with more data from unusually cold periods & hot periods.
- It might be better to work this whole process in reverse, so heating load is taken from OpenEnergyMonitor and then the electricity load is calculated using COP values.
Conclusion
This project has provided insights into the potential impact of heat pump adoption on the UK electricity grid. It highlights the importance of installation quality and the need for robust grid management strategies.
All the code and outputs are available at https://github.com/Miteas/Grid_Load_from_COP. It should be considered a work in progress for personal experimentation rather than a definitive heat pump grid model.
Acknowledgements
I extend my gratitude to OpenEnergyMonitor for the initial dataset that formed the basis of this work, and to Open-Meteo for their weather API used in the underlying electricity usage prediction model.