Solar PV/Battery forecast taken a bit too far

[S2D2]

I've been interested in PV for a while but put off by the variance of estimated savings making it difficult to determine how useful it would be to my specific house. PVGIS is a fantastic tool, but how much of that generation could I actually use? "Around 50%" wasn't specific enough for me but fortunately, my smart meter was recently updated to connect to DCC, so it was time to go down that rabbit hole, results shared here to prompt debate and so others can point out errors and maybe explore a similar study on their own property.

 

Firstly, hourly data for the proposed install location was downloaded from PVGIS and averaged over the 2005-2020 period available: https://re.jrc.ec.europa.eu/pvg_tools/en/. This average was then dumped into an InfluxDB database

Next, half-hourly consumption data from my smart meter is available via a DCC connected company, with 13 months of data available. There are a number available, I chose Bright as they have API support: https://play.google.com/store/apps/details?id=uk.co.hildebrand.brightionic. This data was pulled down with a python library into the InfluxDB database: https://github.com/cybermaggedon/pyglowmarkt. Note I had to pull 10000 minutes at a time as there seems to be a limit on request size.

 

Now I have all the data, there's a few options on what to do with it. Firstly, Grafana was used to support detailed interrogation of the data as well as an in-built method for summing timeseries data. Below pictures are for a 3kW system mounted vertically on a ~SSE facing house wall:

 

 

Annual utilised PV is just the min of usage/generation for each hour in the time series. The result is... disappointing, suggesting if I'd had this system installed over the last year I'd only have used 40% of the generated power. As with most houses, the culprit is significant evening usage - computers, TVs, dishwasher etc. I can shift some of the usage to mid day, but not enough to make a significant enough impact. Inspecting a random day (6th September) confirms an increase in PV size would not solve the problem either:

 

 

Area on this graph is kWh so it's easy to see a lot of wasted generation mid day (between the blue and orange lines) followed by significant grid usage in the evening (between the green and blue lines). Note that electicity consumption is quite low as my heating and hot water is from mains gas. Ideally this excess could be dumped into a hot water tank but we have a combi boiler and I'm struggling to make the sums add up once you factor in even more initial investment for a tank etc. Therefore I'm ignoring offsetting gas consumption and purely looking at the significantly more expensive unit rate electricity consumption.

 

That aim then brings us to battery storage. I couldn't find a way to simulate state in Grafana so used the InfluxDB Python bindings to set up a basic charge/discharge simulation across the year. It does not factor in battery efficiency due to me being lazy which will slightly skew figures. This simulation takes three parameters: PV system power in kW, battery capacity in kWh and the unit electricity rate to calculate savings. Hard to predict for a long term investment so I just used the current 27.09p/kWh rate. It's then possible to experiment with proposed systems:

 

$ ./battery_sim 0 0 0.2709
0.00kWh utilised and 0.00kWh stored out of 0.00kWh generated. Total 0.00kWh (0%). From grid: 2280.95kWh. Annual Saving £0.00

$ ./battery_sim 2 0 0.2709
750.32kWh utilised and 0.00kWh stored out of 1424.67kWh generated. Total 750.32kWh (53%). From grid: 1530.63kWh. Annual Saving £203.26

$ ./battery_sim 3 0 0.2709
863.03kWh utilised and 0.00kWh stored out of 2137.00kWh generated. Total 863.03kWh (40%). From grid: 1417.92kWh. Annual Saving £233.79
$ ./battery_sim 3 5 0.2709
863.03kWh utilised and 998.04kWh stored out of 2137.00kWh generated. Total 1861.07kWh (87%). From grid: 419.89kWh. Annual Saving £504.16
$ ./battery_sim 4 5 0.2709
930.05kWh utilised and 1102.64kWh stored out of 2849.34kWh generated. Total 2032.69kWh (71%). From grid: 248.26kWh. Annual Saving £550.66
$ ./battery_sim 4 9.5 0.2709
930.05kWh utilised and 1151.73kWh stored out of 2849.34kWh generated. Total 2081.78kWh (73%). From grid: 199.17kWh. Annual Saving £563.96
$ ./battery_sim 5 9.5 0.2709
970.84kWh utilised and 1211.10kWh stored out of 3561.67kWh generated. Total 2181.94kWh (61%). From grid: 99.01kWh. Annual Saving £591.09
$ ./battery_sim 5 13.5 0.2709
970.84kWh utilised and 1221.01kWh stored out of 3561.67kWh generated. Total 2191.85kWh (62%). From grid: 89.10kWh. Annual Saving £593.77
$ ./battery_sim 7 13.5 0.2709
1018.43kWh utilised and 1268.62kWh stored out of 4986.34kWh generated. Total 2287.04kWh (46%). From grid: 0.00kWh. Annual Saving £619.56

 

Quite easy to spot the return on investment of the PV and battery capacity is tightly coupled, no point having loads of generation you can't store or massive storage with no excess generation. This is where I've stopped for now, it'd be trivial to go one step further and optimise the system for most cost-effective setup but that would need a cost function for £/kW solar and £/kWh storage (installed) that I haven't bothered to put together yet.

 

Note that the upgrade to a Tesla Powerwall (13.5kWh) is very cost-ineffective, but could technically provide 100% of my usage with a 7kW solar array. The 5kWh battery with 3kW PV seems a sweet spot for me and uses 87% of the generated power but I'm unsure on the costs of such a system yet. To go ahead I'd prefer a maximum 10 year ROI which sets pretty tight budgets, ~£2k for a 2kW system with no battery or ~£5k for a 3kW PV 5kWh battery system. This obviously rules out MCS, but has anyone got close to this with self installation and connection by an electrician?

 

Comments and questions welcome, does this line up with actual performance people have seen from their installs?

Edited September 8, 2022 by S2D2

[Ferdinand]

Octopus Agile Outgoing Tariff may help. 

[Radian]

1 hour ago, S2D2 said:

Area on this graph is kWh

Oh, so what happens when you put on the kettle and the oven at the same time? Your PV clips out at 3kW so you're not necessarily seeing full utilisation even though the half-hour averages make it seem like you are.

[S2D2]

3 minutes ago, Radian said:

Oh, so what happens when you put on the kettle and the oven at the same time? Your PV clips out at 3kW so you're not necessarily seeing full utilisation even though the half-hour averages make it seem like you are.

This is one of a number of limitations of the abstracted model which I'm ignoring completely and crossing fingers that the summed effect isn't significant. The data is simply not available at a high enough fidelity to model such events. Another is averaging PV output data, which not insignificantly undermines the battery simulation accuracy as the battery has a min/max which causes clipping.

[Radian]

It may be too late in the day for you but you could get a CT clamp meter (emonpi or homebrew something else up with an ESP?) and have all the resolution your heart desires. I measure real power over every mains cycle (for a PV divert) and update a database 10 times a minute (to see things like kettles going on and off). Admittedly this doesn't help with the PV database but things move at a more stately pace in the sky.

[S2D2]

Yep increased fidelity will improve accuracy if anyone else wants to try this. It's always annoyed me that I can't easily ping the smart meter for a reading, the DCC data is often ~8 hours late so you can't do any real time control with it, but it's good enough for basic calculations like this.

[S2D2]

1 hour ago, Ferdinand said:

Octopus Agile Outgoing Tariff may help. 

Interesting idea at the moment but difficult to project far into the future I guess. It requires MCS too but at current rates might cover that additional cost. Thanks!

[Radian]

I'm lucky that I've got a HW immersion. Could you not preheat the cold feed into the combi with a smallish tank? Compared with batteries water is a very cheap energy store. However, I'm also keen to do some small scale off-grid storage to be used as a 'glorified' UPS. I think a <£1500 investment in battery and charger/inverter could make sense in this role.

[S2D2]

I've seen the pre-combi tank recommended on here but don't really have a handle on installation costs. Can anyone who has done it give a ballpark figure? Because mains gas is still relatively cheap (ironically) it only saves about £90/year on the 3kW system. Add in annual maintenance of UVC and the ROI seems big, but I'm making unfounded assumptions and should price it up instead. 

[S2D2]

On 08/09/2022 at 16:23, Radian said:

Oh, so what happens when you put on the kettle and the oven at the same time? Your PV clips out at 3kW so you're not necessarily seeing full utilisation even though the half-hour averages make it seem like you are.

I revisited this as the PV data can be interpolated to 30m values to see what the impact of improved fidelity was. Around 5% reduction in utilised PV by moving from 1h windows to 30m windows. This trend would continue as the fidelity increases, so there's probably at least another 5% reduction in utilised PV missed by this model.

 

Much more significant is the error introduced by averaging PV data. If I instead take only the 2020 timeseries utilised PV drops by around 23%. It's intuitive as to why, I was smoothing out daily irregularity by taking an average, which has a big impact on utilisation if we look at the same day as my original image:

 

 

Poor midday sun causing a large draw from the grid. This helps support a case for batteries; even when shifting load to mid day you might just get unlucky with PV generation, I don't want to have to predict the cloud patterns before setting off the dishwasher.

 

As a result, utilisation for a 3kWp system with no battery drops even further to 31% making it very hard to justify any kind of ROI on PV only when export rates are either 0 or require thousands extra for MCS. I added a crude pre-combi water tank model and with such low utilisation it does become just about viable based on some estimated costs, still keen to hear from anyone who has installation costs for this. More accurate information could be gained by running each year's PV data in turn and averaging the outputs.

 

So with batteries having a few advantages, including offsetting the much higher electricity cost over gas, I went down the route of optimising such a system. With higher PV generation a pre-combi tank could be added as well, but I am space limited on the number of panels I can install. Some rough equipment/install cost estimates (Not MCS) and 133 data points later gives me the following optimisation space:

 

 

Ignore Excel's bizarre labelling, y-axis is Battery (kWh). The optimal value being around 3.65kWp PV with a 4.56kWh battery. Adding a tank to this system would only save ~£79 a year so fairly hard to justify the extra expense. Does anyone have reference quotes for a similar system to double check my cost functions?

[SteamyTea]

@S2D2

 

Can you reduce your non daylight usage?

 

May be worth adding in some margins of error, and are your 3 series all on the same time axis? i.e. GMT OR BST, not a mixture.

 

May also be worth reading this.

Edited September 14, 2022 by SteamyTea

[S2D2]

10 minutes ago, SteamyTea said:

@S2D2

 

Can you reduce your non daylight usage?

 

May be worth adding in some margins of error, and are your 3 series all on the same time axis? i.e. GMT OR BST, not a mixture.

 

May also be worth reading this.

Baseline usage is already as low as I can get it, 30W constant around 50w including fridge freezer. Some of the peaks can be shifted - dishwasher, washing machine, but I've highlighted issues with that above, plus it's not really something I want to have to think about and would reduce the ROI for that benefit. I don't have any error margin data for either dataset, you could estimate for PV but it would be gigantic given how variable the data is.

 

All data stored UTC and in the images above displayed in BST so accurate between series.