Solved

Why does my Vehicle to Grid (V2G) charger choose to import energy at midday instead of overnight, without me scheduling the charging? Kaluza AMA

  • 30 July 2020
  • 8 replies
  • 104 views

Userlevel 3

I have recently started to set a schedule for my Leaf to be charged early morning as I noticed that without a scheduled time, you were not charging it until around midday. That seems mad - surely you can buy energy cheapest in the middle of the night - so why wouldn’t you charge up my Leaf at 3am?

Also I suspect that now I have set a scheduled early morning time, the in-out flow is a little higher so my reward payments might be a bit higher?

Cheers

Julian

icon

Best answer by Tim_OVO 29 October 2020, 14:00

I’m very excited and happy to share this in-depth response from Josh, the Kaluza app product manager, let us know what you think! 

 

 

Hi @TeaTimeSoon 

 

This is a fascinating question and a good opportunity to talk about how our machine learning algorithms are crafted.

 

The main objective of the algorithm is to minimize costs, primarily to you as a user, but also to us as a supplier. 

 

If the user has a single rate tariff, this will be to minimize the cost of supplying the energy to the supplier. This the foundation for us building a commercially viable offering.

 

If the user has a multi-rate tariff, this will be to the user first (i.e. bias their off-peak period) and the supplier second (as above).

 

It follows that you incur cost when you import energy and reduce cost when you idle or export.

 

Much like trading on the stock market, the algorithm aims to import when the wholesale price is lowest and idle/export when the wholesale price is highest.

 

Intuition might suggest that a particular time of day is generally preferable to another, but there are some important things to consider:

 

  • Price volatility

    • While the shape of consumption may be consistent from one day to another, the price of the energy can vary considerably. One cannot assume because it cost £10/MWh on Tuesday at 10 am that it will cost £10/MWh on Wednesday at 10 am. In the same vein, energy available overnight is not always as cheap or desirable (from an environmental perspective) as one might think.

  • Local price

    • The algorithm doesn’t just use the national wholesale price when making its calculations, but rather the local wholesale price, which could result in one customer importing at the same time as another customer in a neighbouring postcode region exporting.

  • Dataset size

    • The algorithm ingests a massive amount of data when making these decisions, a volume that would far outstrip any single human’s capacity to consume and analyse.

    • While loose, my favourite analogy for this is using sat nav to navigate around my local area during rush hour. I used to be convinced that I knew better until I ran into the back of a traffic jam one too many times and learnt to trust that even when I thought I knew better, my sat nav had access to far more data than I could ever hope to have at that moment and could make much better decisions about the optimum route I should take than I could.

  • Frequency

    • The algorithm refreshes its decisions for all the thousands of devices connected to the platform every 2 minutes.

  • Forecast

    • Probably the most significant characteristic of the algorithm is its ability to look ahead using forecasts to make decisions about future events. Human intuition is most helpful about making decisions in the present, but the algorithm is able to look 8-10 hours ahead at what wholesale prices are likely to be and make decisions based on the losses and gains over that entire period. 

    • In contrast, humans by nature are poor at predicting the future, tend to be much more optimistic about future events when compared to reality, and are frequently convinced they made the right decision regardless (projection, optimism and confirmation bias respectively).

  • Long term vs. short term gains

    • Fundamentally, the algorithm has been directed to prefer gains over the longer term. If it foresees a peak in the future, it may in fact charge now at what might appear to be a high price point in order to reduce costs overall by being able to export more energy in the future.

 

When you set a ready by time, you’re applying an additional constraint the algorithm has to consider and this is given priority over gains that could be made across the entire day. Even more so with a multi-rate tariff which is given an even higher priority and could result in beneficial charging or discharging not taking place because it would fall outside your off-peak period.

 

If you’re interested in finding out more, it’s well worth watching the Alpha Go documentary. The makers demonstrate very clearly how a machine doesn’t think like a human and has no objection taking an approach we would consider wildly left field or goes against well established social norms or “universal truths”.

 

 

In summary, if you don’t have anywhere to rush off to and want to support the grid, consider using your charge range to constrain import/export instead of ready by times (Sheila_bell does this to great effect and demonstrates it in her 1-year post installation video). If you need to go further afield, just boost beforehand.

 

 

Given the above explanation, @philsquared I’m intrigued to understand what you mean by uneconomical. Uneconomical for whom? Uneconomical in what way?

 

@Transparent your point about being able to set more fine-grained constraints is an interesting one. However, I worry that the layperson would want charging triggered when there was only 100% green energy available (akin to a Zappi in Eco+ mode). While your supplier might sell you a 100% renewable energy tariff, until we stop burning fossil fuels for energy generation, there will always be some portion of brown energy in our wires. I’m concerned such a feature would result in an avalanche of support calls from angry customers querying why their batteries weren’t charged…

 

Thanks for such a great question @TeaTimeSoon 

 

Josh

View original

8 replies

Userlevel 2

I have recently started to set a schedule for my Leaf to be charged early morning as I noticed that without a scheduled time, you were not charging it until around midday. That seems mad - surely you can buy energy cheapest in the middle of the night - so why wouldn’t you charge up my Leaf at 3am?

Also I suspect that now I have set a scheduled early morning time, the in-out flow is a little higher so my reward payments might be a bit higher?

Cheers

Julian

 

I actually had the same thought. I had a 7am “charged” time from the “before times”, but I left it on even though we don’t need it in the mornings anymore because I had an intuition it would charge at uneconomical times, otherwise. Sounds like I was right. I might set it back to an even earlier time, too.

Userlevel 7
Badge +2

@TeaTimeSoon wrote:

I have recently started to set a schedule for my Leaf to be charged early morning as I noticed that without a scheduled time, you were not charging it until around midday. That seems mad - surely you can buy energy cheapest in the middle of the night - so why wouldn’t you charge up my Leaf at 3am?

Others may think differently.

There’s generally very little renewable energy available at 3am. So if customers want to be taking action to combat Climate Change, then they would be much more likely to require a mid-morning charge-cycle.

As I understand it, the main (only?) driver for the algorithm is the wholesale price of electricity, which changes by the minute.

What plans do Kaluza have to allow customers to set their own “balance” between price and energy-mix criteria?

Userlevel 3

@TeaTimeSoonwrote:

I have recently started to set a schedule for my Leaf to be charged early morning as I noticed that without a scheduled time, you were not charging it until around midday. That seems mad - surely you can buy energy cheapest in the middle of the night - so why wouldn’t you charge up my Leaf at 3am?

Others may think differently.

There’s generally very little renewable energy available at 3am. So if customers want to be taking action to combat Climate Change, then they would be much more likely to require a mid-morning charge-cycle.

As I understand it, the main (only?) driver for the algorithm is the wholesale price of electricity, which changes by the minute.

What plans do Kaluza have to allow customers to set their own “balance” between price and energy-mix criteria?

Wind energy is now the largest alternative energy source for the UK - please have a look at gridwatch.co.uk for a fantastic data presentation. So for example at 3am last night, wind was responsible for fulfilling 34.3% of the total UK electricity requirements (and even though I don’t have the wholesale energy costs to prove it, I would have thought the cost would have been pretty low just then). Wind does not stop at night, although it is of course intermittent in how much it contributes,

I would agree that the night before (the night of the 28th - morning of the 29th) wind was only around 7% at 3am, but my point remains, when the wind is blowing at 3am, why is Kaluza not getting such low cost energy that it makes sense to charge my car? This was not happening until I set a schedule for my car to be charged by 6.30am. 

Userlevel 7
Badge +2

I’m not disagreeing @TeaTimeSoon but, looking at their job-specs, I don’t think Fionn and Josh are best positioned to discuss that. We’d need someone like Conor Maher-McWilliams.

As I understand it, Kaluza trades electricity solely on the National (GB) market. The Kaluza Platform doesn’t “see” electricity which never makes its way onto the National Grid. I’m in an area where we have abundant supplies of wind and solar energy which can’t be sent to the rest of GB due to capacity constraints.

I can see this because the running statistics for Energy-Mix are available down to the level of the Bulk Supply Point (33kV output). There are considerable differences between the amount of renewable energy I receive and what the National Statistics state is being supplied into my region.

I suspect that the excess wind-energy in my locality would cost more if it were routed onto the National Grid, but I don’t know how that pricing is calculated.

 

The main reason I broached the general subject of a choice between price and “green” energy mix is because Josh refers to “If This Then That”. A similar approach would be required if customers want more choice as to where their energy is sourced from.

Conversely, that risks making the user-interface too complex…. and that is within Josh’s job-spec!

Userlevel 3

Thanks @Transparent - makes sense and useful info.

I am looking forward to the day when storage systems of whatever flavour (battery, air, gravitational, hydrogen) develop to the degree that the temporary excesses can be managed and properly balance out the supply/demand on the grid rather than switching on more Ccgt or interconnects.

Userlevel 7

I’m very excited and happy to share this in-depth response from Josh, the Kaluza app product manager, let us know what you think! 

 

 

Hi @TeaTimeSoon 

 

This is a fascinating question and a good opportunity to talk about how our machine learning algorithms are crafted.

 

The main objective of the algorithm is to minimize costs, primarily to you as a user, but also to us as a supplier. 

 

If the user has a single rate tariff, this will be to minimize the cost of supplying the energy to the supplier. This the foundation for us building a commercially viable offering.

 

If the user has a multi-rate tariff, this will be to the user first (i.e. bias their off-peak period) and the supplier second (as above).

 

It follows that you incur cost when you import energy and reduce cost when you idle or export.

 

Much like trading on the stock market, the algorithm aims to import when the wholesale price is lowest and idle/export when the wholesale price is highest.

 

Intuition might suggest that a particular time of day is generally preferable to another, but there are some important things to consider:

 

  • Price volatility

    • While the shape of consumption may be consistent from one day to another, the price of the energy can vary considerably. One cannot assume because it cost £10/MWh on Tuesday at 10 am that it will cost £10/MWh on Wednesday at 10 am. In the same vein, energy available overnight is not always as cheap or desirable (from an environmental perspective) as one might think.

  • Local price

    • The algorithm doesn’t just use the national wholesale price when making its calculations, but rather the local wholesale price, which could result in one customer importing at the same time as another customer in a neighbouring postcode region exporting.

  • Dataset size

    • The algorithm ingests a massive amount of data when making these decisions, a volume that would far outstrip any single human’s capacity to consume and analyse.

    • While loose, my favourite analogy for this is using sat nav to navigate around my local area during rush hour. I used to be convinced that I knew better until I ran into the back of a traffic jam one too many times and learnt to trust that even when I thought I knew better, my sat nav had access to far more data than I could ever hope to have at that moment and could make much better decisions about the optimum route I should take than I could.

  • Frequency

    • The algorithm refreshes its decisions for all the thousands of devices connected to the platform every 2 minutes.

  • Forecast

    • Probably the most significant characteristic of the algorithm is its ability to look ahead using forecasts to make decisions about future events. Human intuition is most helpful about making decisions in the present, but the algorithm is able to look 8-10 hours ahead at what wholesale prices are likely to be and make decisions based on the losses and gains over that entire period. 

    • In contrast, humans by nature are poor at predicting the future, tend to be much more optimistic about future events when compared to reality, and are frequently convinced they made the right decision regardless (projection, optimism and confirmation bias respectively).

  • Long term vs. short term gains

    • Fundamentally, the algorithm has been directed to prefer gains over the longer term. If it foresees a peak in the future, it may in fact charge now at what might appear to be a high price point in order to reduce costs overall by being able to export more energy in the future.

 

When you set a ready by time, you’re applying an additional constraint the algorithm has to consider and this is given priority over gains that could be made across the entire day. Even more so with a multi-rate tariff which is given an even higher priority and could result in beneficial charging or discharging not taking place because it would fall outside your off-peak period.

 

If you’re interested in finding out more, it’s well worth watching the Alpha Go documentary. The makers demonstrate very clearly how a machine doesn’t think like a human and has no objection taking an approach we would consider wildly left field or goes against well established social norms or “universal truths”.

 

 

In summary, if you don’t have anywhere to rush off to and want to support the grid, consider using your charge range to constrain import/export instead of ready by times (Sheila_bell does this to great effect and demonstrates it in her 1-year post installation video). If you need to go further afield, just boost beforehand.

 

 

Given the above explanation, @philsquared I’m intrigued to understand what you mean by uneconomical. Uneconomical for whom? Uneconomical in what way?

 

@Transparent your point about being able to set more fine-grained constraints is an interesting one. However, I worry that the layperson would want charging triggered when there was only 100% green energy available (akin to a Zappi in Eco+ mode). While your supplier might sell you a 100% renewable energy tariff, until we stop burning fossil fuels for energy generation, there will always be some portion of brown energy in our wires. I’m concerned such a feature would result in an avalanche of support calls from angry customers querying why their batteries weren’t charged…

 

Thanks for such a great question @TeaTimeSoon 

 

Josh

Userlevel 7
Badge +2

Has anyone else noticed that Indra has signed a deal to see the V2G chargers in Australia? Press story here.

They cost Au$10,000 (£5,400).

This presumably means that Kaluza’s computer running the Flex Platform will be wanting to charge vehicles at midday…. because they get a heap more sunshine than we do! :sun_with_face:

Userlevel 2

Thanks Josh (via @Tim_OVO),

Regarding prediction - “prediction is very hard - especially about the future” (sorry, couldn’t resist).

Great background, Josh - thanks. That all makes sense to me - and is what I expected (or, at least hoped). I suppose I was just surprised that it rarely seemed to favour imports overnight. but often during peak, or close to peak times (peak in terms of typical usage, rather than production, necessarily).

For now we’re back to regular usage, where my problem is in the other direction (I need two “ready by” times, and not being able to set the second one I often have to use boost, so not benefiting from the algorithm* during the day). If we end up locking down again (looking increasingly likely) I’ll try to be more trusting.

---

(*) I realise “algorithm” is not the technically correct word in the context of ML, but I’m sure you know what I mean.

Reply