Posts Tagged ‘Battery’

Assessing Powerwall battery degradation

December 12, 2022

Click on image for a larger version. Three screenshots from my phone showing the performance of the battery on 9th and 17th January 2022, and 11th December 2022. The key data concerns the total amount of energy discharged from the Powerwall. See text for details.

Friends, the Tesla Powerwall2 battery that we installed in March 2021 has transformed the way we use electricity and allowed us to go off-grid for prolonged periods each year. I have no regrets.

But lurking at the back of my mind, is the question of battery degradation.

This phenomena arises due to parasitic chemical reactions that occur as the battery approaches either full charge or full discharge. These reactions ‘capture’ some lithium and remove its ability to be used to store charge. Hence one expects the capacity of a battery to decline with extended use, particular near the extremes of battery capacity.

This particularly affects batteries used for domestic applications as they are often charged fully and then discharged fully – particularly in the winter.

The extent of the degradation depends on the specific chemistry of the battery. More modern battery chemistries labelled as ‘LiFePO4: Lithium Iron Phosphate” perform better than the previous best in class so-called “NMC: Nickel Manganese Cobalt”. Unfortunately, the Powerwall2 uses NMC batteries. This article has a comparison of the properties of different lithium-ion battery chemistries.

Battery degradation is a real phenomenon, but unsurprisingly, battery manufacturers do not make it straightforward to spot. I first looked at this about a year ago, but I don’t think my analysis was very sensible.

I now think I have a better method to spot degradation, and 20 months after installation, initial degradation is apparent.


The new method looks at data from winter days during which the battery is discharged from full to empty, with little or no solar ‘top up’.

In winter our strategy is to charge up the battery with cheap electricity (currently 7.5p/kWh) between 00:30 and 04:30 and to run the house from this until the battery is empty. When it’s cold and the heat pump is working hard we can use up to 30 kWh/day and so the nominal 13.5 kWh of stored electricity is not enough to run through the day. So we run out of battery typically in the early evening and then run off full-price electricity until we can top up again.

The run-time can be extended by a top-up from the solar PV system, which can be anything from 0 kWh in overcast conditions, up to around 7 kWh in full December sun.

My idea is to measure the Powerwall’s total discharge and to compensate for any solar top up. By restricting measurements to days when the battery goes from full to empty, I don’t have to rely on estimates of battery remaining capacity. These days mainly occur in December and January.

For example, today (12 December 2022), the battery was charged to 100% at 04:30 and discharged 12.8 kWh to give 0% just after midday. I the estimate battery capacity as 12.8 kWh.

But on 7 December 2022, the battery was charged to 100% at 04:30 and discharged 15.5 kWh to give 0% just 22:00. This was a sunny day and the battery was topped up by 3.0 kWh of solar. I thus estimate battery capacity as 15.5-3.0 = 12.5 kWh.

In this latter case the way to compensate for the 3.0 kWh of charging is not clear. Why? Because the 3 kW of solar is used to charge the battery and so this may be done with say 95% efficiency (say) in which case only 2.85 kWh of solar energy would be stored. So there is some ambiguity in data which is solar compensated, but for this analysis I am ignoring this difficulty.

The data are shown below:

Click on image for a larger version. Graph showing total Powerwall discharge after compensating for any solar top-up. See text for details.


The nominal capacity of the Powerwall2 is 13.5 kWh. This is – presumably – the stored electrical energy of the batteries when they are fully charged. To be useful, this energy must be discharged and converted to AC power, and this cannot be done with 100% efficiency.

Considering the data from the winter of 2021/22, the average Full-to-Empty discharge was 13.1 kWh, and so it looks like the discharge losses were around 3%. I think this is probably a fair estimate for the performance of a new battery.

The data show a considerable amount of scatter: the standard deviation is around 0.2 kWh. I am not sure why this is. Last winter, the battery would sometimes only charge to 99% rather than 100% and I corrected for this. That is why the capacity data do not lie entirely on exact tenths of a kWh.

Considering the data from the winter of 2022/23, the average Full-to-Empty discharge is currently 12.8 kWh. This represents a reduction in capacity of 2.3% (0.3 kWh) compared with last winter. However, there is a whole winter ahead with another 50 or so full discharges before spring and that average could well fall.

If the trend continued then battery capacity would fall to 10 kWh in around 2030. That would still be a useful size battery, and by that time hopefully a newer (and cheaper!) model will be available.

I‘ll be keeping an eye on this and will write an update at the end of the winter season. But I thought it was worth publishing this now in case fellow battery owners wanted to monitor their own batteries in a similar way.


Domestic Batteries: Purchase decisions and realistic models

February 1, 2021

Friends, earlier this week I ordered a Tesla PowerWall 2 from the charming people at The Little Green Energy Company (TLGEC). They have given me a nominal installation date in late March 2021 and I will be sure to keep you updated.

So in my excitement I wrote another article about using batteries – and you can read it at length below. But AFTER I had spent hours calculating and graphing , I realised something very obvious but very profound.

  • The triple-glazing and external wall insulation have been ‘green’ investments. They avoid the need to burn fossil fuels.
  • The solar panels have been a ‘green’ investment. They produce low-carbon electricity.
  • The heat pump (when I install it) will be a ‘green’ investment. It will avoid the need to burn gas to heat the house.
  • But the battery is a financial investment. It will actually use extra electricity! However, it will lower the cost to me personally of making the ‘green’ investments.

My aim is to transition away from burning gas by using a heat pump. This switch requires me to use more electricity each year and without the financial savings that a battery yields this would be punitive.

More battery modelling: but using a climate re-analysis database!

I chose TLGEC over other installers because of their willingness – and ability – to answer tricky questions. And in one of their answers they gave me a jewel of link to this EU funded site with useful information about solar PV.

The site can be used like others to estimate the monthly generation from a solar PV installation. But unlike other sites the predictions are based on actual solar data over the period 2005-2016.

And uniquely – by using climate re-analysis –  it is possible to download this data for any location on Earth (!) to simulate hour-by-hour how a particular installation of panels would respond at any time during that period.

Click for a larger image. This web portal is available here.

This has enabled me to create models simulating the interaction of solar panels with a domestic battery similar to those I made previously. But instead of:

  • a minute-by minute model of a single day using simulated solar data,

I can now make…

  • an hour-by-hour model of an entire year using actual solar data.

Crucially this incorporates real-world (hour-to-hour and day-to day) variability which is one of the difficulties in trying to optimise the use of a battery.

The Model 

The Excel™ model (Solar Time Series Analysis 2005 – 2016 for Blog) is based (unsurprisingly) on a Tesla Powerwall 2 with 13.5 kWh of storage, but that can be changed in the file. Please note – this is not a simple model and is set up just for my panels in Teddington! If you want to use it for your site you will need to download data from the web portal above and place it in the spreadsheet.

The model has the following ‘features’ (default values shown in brackets)

  1. The electrical demand can have separate daily peak (1 kW) and off-peak (0.5 kW) values.
  2. The overnight charging rate can be changed (3 kW)
  3. The fractional filling of the battery in the morning can be changed seasonally between a summer value (100%) and a winter value (100%).
  4. The range of the ‘summer’ and ‘winter’ seasons can be defined (summer runs from day 60 to day 300)

The model evaluates:

  • The state of the charge of the battery hour-by-hour through the year,
  • The amount of peak and off-peak electricity which must be purchased to meet the required demand.
  • The amount of solar generation and the amount used on site, or exported.
  • The costs of different strategies.

One shortcoming of the model is that the 1-hour step is too long and so in some situations the model appears to overfill or underfill the battery. However I think the uncertainty this adds is relatively small.

The Parameters

I set the model to run with data both from individual years and from the average behaviour of all 12 years of data.

The demand I modelled was 0.5 kW overnight and 1 kW during the day. This is more than our house uses at present but is in line with the demand I expect when I install a heat pump to replace the gas boiler.

The model calculates the amount of electricity bought from the grid in both peak and off-peak periods and evaluates the fraction of demand met by solar electricity, and the cost.

I then investigated how different settings for the morning filling of the battery affected:

  • the amount of electricity bought from the grid (peak and off-peak) over the year,
  • the fraction of demand met by solar electricity,
  • the cost.

Typical Runs

The graph below shows the simulated State of Charge (SoC) of the battery during days 1 to 30 of the year 2016 i.e. January 2016.

Click for a larger view.

The graph shows daily overnight charging of the battery to 100% in the morning. The 1-hour time resolution of the simulation makes it appear the battery does not quite completely fill up, but it gets close.

The 1 kW daytime load then drains the battery completely on most days – the SoC reaches zero – and so some full price electricity must be bought.

However, there are a few days (e.g. days 7 & 8 and days 13 to 16) even in January in which strong sunlight fills the battery sufficiently that it lasts to the end of the day. These would typically be cold, crisp, clear winter days.

To indicate the variability, the equivalent graph for the year 2011 is shown below.

Click for a larger view.

But if we plot the average data from 2005 to 2016 we see it has a different character from that for individual years. Instead of the 3 or 4 bright sunny days, we have – on average – a little bit of sunshine on many more days.

Click for a larger view.

This difference between individual years and their average is important in this case, because it the intermittency of solar generation that makes a battery useful, and it is the irregularity of solar generation in any one year that makes it hard to optimise the use of a battery.

A whole year of averaged data is shown in the graph below. I have used average data to illustrate the general characteristics of the behaviour of the battery.

Click for a larger view.

In this graph the battery is charged each night to 100% SoC. In the winter it discharges through the day and the SoC reaches zero before the end of the day, requiring full price grid electricity to tide the household over to the end of the day and the start of cheap electricity.

But between days 60 and 300 there is enough solar generation – on average – such that the battery does not ever fully discharge at the end of each day. Thus in this period is not really necessary to fully charge the battery overnight.

The graph below shows the effect of only charging the battery to 70% in the mornings over this ‘summer’ period.

Click for a larger view.

The result of this is that less night-time electricity is used, and less electricity is exported. Consequently, the ‘self-use’ of solar electricity increases. However, there are now a few more occasions during the ‘summer’ when the  SoC reaches zero before the end of the day i.e. where full price electricity must be bought.

The graph below shows the same partial-charging strategy (only 70% between days 60 and 300) but using data for the year 2011: notice that the irregularity is much greater than when looking at the averaged data.

Click for a larger view.

So how does one make sense of all this? I do not want to spend my entire life optimising battery charging!

Basic Results

There are too many variables to succinctly summarise the modelling results, so here I will just summarise one investigation relevant to my own situation.

Imagining that I am running a heat pump to replace the gas boiler, I have assumed overnight use at 0.5 kW and daytime use at 1.0 kW. This amounts to 21 kWh/day or 7665 kWh/year. Due to the limited time step, the model calculates annual use as 7661 kWh – which is an error of 0.05%.

Using the solar data for each individual year – and for the average of all the years – I calculated how self-use of solar power varied as I changed the state of charge (SoC) of the battery in the morning from 0% to 100%.

By ‘self-use’ I mean that the solar electricity was either used immediately at the house or stored in the battery for later use. Nominally either of these uses is ‘free’, but in reality the storage and retrieval is only around 90% efficient.


First of all looking at solar data from each year 2005 to 2016 I calculated that on average the panels would generate 3847 kWh/year with a standard deviation of about 5%. The average value is same as is calculated from just using the average 2006-2016 datset

Click for a larger view.

The solar generation is only around half of the anticipated demand (see below). And without a battery, most of that is exported at a relatively low price (1.8 p/kWh from EDF). This benefits the planet and EDF, but means I still have to pay EDF 23.7 p/kWh for peak time electricity to operate the heat pump.

Click for a larger view.

Next – using the solar data for each individual year – and for the average of all the years – I calculated how self-use of solar electricity varied as I changed the state of charge (SoC) of the battery in the morning from 0% to 100%.

Click for a larger view. The graph shows the number of units of solar electricity (kWh) that would have been used on site.

If we pick one year (say 2014) as an example, we that in this sunnier-than-average year, charging the battery to about 30% SoC in the morning leaves plenty of capacity to store solar electricity during the day.

In a more typical year (say 2016) the optimum morning SoC is between 40% and 50%.

  • Higher morning SoC results in solar generation being ‘lost’ to export.
  • Lower morning SoC will give rise to earlier discharge of the battery and the use of more mains electricity.

Curiously, the optimum morning SoC for any individual year (30% to 60%) is quite different from that calculated from the average of all 12 years. This is because of reduced irregularity in the averaged data.

The difference between self-use calculated from data for individual years and the self-use calculated from the average data is even more striking if we show each year’s result as a fraction of that year’s total generation.

Click for a larger view. The graph shows the fraction of total solar generation (%) that would have been used on site for each year.

We see that we might hope to get around 90% of self-use in any individual year with a morning SoC of around 40%. This is much lower than the 98% which appears possible using averaged data.

Results: Economics 101

As I whiled away happy hours with Excel I became fascinated by different possible strategies. And I filled my head with clever calculations that I might attempt.

But then I realised that none of these strategies affects the carbon reduction I achieve by installing solar panels. This happens with or without a battery and is independent of the charging strategy I adopt!

  • What these charging strategies affect is who gets the benefit!

If I export electricity at low cost (1.8 p/kWh in the case of EDF) and am then forced to buy electricity later in the day for 23.7 p/kWh (EDF) then it is EDF who gets the benefit of my investment.

Financially, the optimum strategy arises from the differences between night-time and day-time electricity, and the price paid for exports. I have illustrated this for two ‘tariffs’ below – those from EDF and those from Tesla – who have a deal with Octopus.

Click for a larger view.

If I simply bought the electricity from EDF without solar panels, then the annual cost would be just over £1600.

The solar panels should reduce this cost substantially. The investment of £4200 in the solar panels should generate a saving of around £500/year, a 12% return on investment.

The battery should lower the annual cost much further. The savings generated by this £10,000 investment should be more than £800/year.

  • Using the EDF tariff, the big difference between the price of day-time and night-time electricity makes it always preferable to have a morning SoC as high as possible, thus minimising the possibility of ever having to use full-price electricity.
  • Using the Tesla tariff – the morning SoC doesn’t matter because there is no time-of-day price difference, and no difference in price between imports and exports.

But using either tariff, I calculate the savings to be massive. So large in fact that I just can’t believe them! The battery should be installed in March and I will let you know how it goes!

Of course I could also lower the cost by switching from EDF. I checked with Octopus energy (link) and it listed 80 different tariffs. Eighty! Enough for 10 octopuses to each have a tariff for each leg.  I absolutely detest this confusopoly. In any case the cheapest night time price was around 11p. Hopefully with the battery I will be able to subsist mainly on EDF’s night-time tariff.


So after all that work, I realised something very obvious but very profound. As I said at the top the article:

  • The triple-glazing and external wall insulation have been ‘green’ investments. They avoid the need to burn fossil fuels.
  • The solar panels have been a ‘green’ investment. They produce low-carbon electricity.
  • The heat pump (when I install it) will be a ‘green’ investment. It will avoid the need to burn gas to heat the house.
  • But the battery is a financial investment. It will actually use extra electricity! However, it will lower the cost to me personally of making the ‘green’ investments.

Is a UK grid-scale battery feasible?

April 26, 2019

This is quite a technical article, so here is the TL/DR: It would make excellent sense for the UK to build a distributed battery facility to enable renewable power to be used more effectively.


Energy generated from renewable sources – primarily solar and wind – varies from moment-to-moment and day-to-day.

The charts below are compiled from data available at Templar Gridwatch. It shows the hourly, daily and seasonal fluctuations in solar and wind generation plotted every 5 minutes for (a) 30 days and (b) for a whole year from April 21st 2018. Yes, that is more than 100,000 data points!

Wind (Green), Solar (Yellow) and Total (Red) renewable energy generation for the days since April 21st 2018

Wind (Green), Solar (Yellow) and Total (Red) renewable energy generation for 30 days following April 21st 2018. The annual average (~6 GW) is shown as black dotted line.


Wind (Green), Solar (Yellow) and Total (Red) renewable energy generation for the 365 days since April 21st 2018. The annual average (~6 GW) is shown as black dotted line.

An average of 6 GW is a lot of power. But suppose we could store some of this energy and use it when we wanted to rather than when nature supplied it. In other words:

Why don’t we just build a big battery?

It turns out we need quite a big battery!

How big a battery would be need?

The graphs below shows a nominal ‘demand’ for electrical energy (blue) and the electrical energy made available by the vagaries of nature (red) over periods of 30 days and 100 days respectively. I didn’t draw the whole year graph because one cannot see anything clearly on it!

The demand curve is a continuous demand for 3 GW of electrical power with a daily peak demand of 9 GW. This choice of demand curve is arbitrary, but it represents the kind of contribution we would like to be able to get from any energy source – its availability would ideally follow typical demand.



We can see that the renewable supply already has daily peaks in spring and summer due to the solar energy contribution.

The role of a big battery would be cope to with the difference between demand and supply. The figures below show the difference between my putative demand curve and supply, over periods of 30 days and a whole year.



I have drawn black dotted lines showing when the difference between demand and supply exceeds 5 GW one way or another. In spring and summer this catches most of the variations. So let’s imagine a battery that could store or release energy at a rate of 5 GW.

What storage capacity would the battery need to have? As a guess, I have done calculations for a battery that could store or release 5 GW of generated power for 5 hours i.e. a battery with a capacity of 5 GW x 5 hours = 25 GWh. We’ll look later to see if this is too much or too little.

How would such a battery perform?

So, how would such a battery affect the ability of wind and solar to deliver a specified demand?

To assess this I used the nominal ‘demand‘ I sketched at the top of this article – a demand for  3 GW continuously, but with a daily peak in demand to 9 GW – quite a severe challenge.

The two graphs below show the energy that would be stored in the battery for 30 days after 21 April 2018, and then for the whole following year.

  • When the battery is full then supply is exceeding demand and the excess is available for immediate use.
  • When the battery is empty then supply is simply whatever the elements have given us.
  • When the battery is in-between fully-charged and empty, then it is actively storing or supplying energy.


Over 30 days (above) the battery spends most of its time empty, but over a full year (below), the battery is put to extensive use.


How to measure performance?

To assess the performance of the battery I looked at how the renewable energy available last year would meet a levels of constant demand from 1 GW up to 10 GW with different sizes of battery. I consider battery sizes from zero (no storage) in 5 GWh steps up to our 25 GWh battery. The results are shown below:

Slide15It is clear that the first 5 GWh of storage makes the biggest difference.

Then I tried modelling several levels of variable demand: a combination of 3 GW of continuous demand with an increasingly large daily variation – up to a peak of 9 GW. This is a much more realistic demand curve.Slide17

Once again the first 5 GWh of storage makes a big difference for all the demand curves and the incremental benefit of bigger batteries is progressively smaller.

So based on the above analysis, I am going to consider a battery with 5 GWh of storage – but able to charge or discharge at a rate of 5 GW. But here is the big question:

Is such a battery even feasible?

Hornsdale Power Reserve

The Hornsdale Power Reserve Facility occupies an area bout the size of a football pitch. Picture from the ABC site

The Hornsdale Power Reserve Facility occupies an area about the size of a football pitch. Picture from the ABC site

The biggest battery grid storage facility on Earth was built a couple of years ago in Hornsdale, Australia (Wiki Link, Company Site). It seems to have been a success (link).

Here are its key parameters:

  • It can store or supply power at a rate of 100 MW or 0.1 GW
    • This is 50 times smaller than our planned battery
  • It can store 129 MWh of energy.
    • This is just under 40 times smaller than our planned battery
  • Tesla were reportedly paid 50 million US dollars
  • It was supplied in 100 days.
  • It occupies the size of a football pitch.

So why don’t we just build lots of similar things in the UK?

UK Requirements

So building 50 Hornsdale-size facilities, the cost would be roughly 2.5 billion dollars: i.e. about £2 billion.

If we could build 5 a year our 5 GWh battery would be built in 10 years at a cost of around £200 million per year. This is a lot of money. But it is not a ridiculous amount of money when considering the National Grid Infrastructure.

Why this might actually make sense

The key benefits of this kind of investment are:

  • It makes the most of all the renewable energy we generate.
    • By time-shifting the energy from when it is generated to when we need it, it allows renewable energy to be sold at a higher price and improves the economics of all renewable generation
  • The capital costs are predictable and, though large, are not extreme.
  • The capital generates an income within a year of commitment.
    • In contrast, the 3.2 GW nuclear power station like Hinkley Point C is currently estimated to cost about £20 billion but does not generate any return on investment for perhaps 10 years and carries a very high technical and political risk.
  • The plant lifetime appears to be reasonable and many elements of the plant would be recyclable.
  • If distributed into 50 separate Hornsdale-size facilities, the battery would be resilient against a single catastrophic failure.
  • Battery costs still appear to be falling year on year.
  • Spread across 30 million UK households, the cost is about £6 per year.


I performed these calculations for my own satisfaction. I am aware that I may have missed things, and that electrical grids are complicated, and that contracts to supply electricity are of labyrinthine complexity. But broadly speaking – more storage makes the grid more stable.

I can also think of some better modelling techniques. But I don’t think that they will affect my conclusion that a grid scale battery is feasible.

  • It would occupy about 50 football pitches worth of land spread around the country.
  • It would cost about £2 billion, about £6 per household per year for 10 years.
    • This is one tenth of the current projected cost of the Hinkley Point C nuclear power station.
  • It would deliver benefits immediately construction began, and the benefits would improve as the facility grew.

But I cannot comment on whether this makes economic sense. My guess is that when it does, it will be done!


Data came from Templar Gridwatch


%d bloggers like this: