@dfaber it has already been described by several users, including me, or @Sarowe1990
A single hour of higher consumption than originally forecast destroys the entire consumption forecast!
The original forecast was for this evening and the coming night, with a constant consumption of approx. 0.7kW/h, just like last night and most of today.
Then, two hours before, a one-off increase in consumption and suddenly the forecast for the next 2 days is consistently high. This destroys the entire planning and energy is charged from the grid actually, which is absolutely unnecessary!
You need to improve this URGENTLY please, DESS is not usable like this!
Dynamic ESS uses consumption and solar yield forecasts to schedule ahead, when the forecasts are available. Depending on the historical data available prior, the accuracy of these forecasts can vary. As it stands now, optimal consumption forecasts need 28 days of past consumption data and optimal solar yield forecasts need 28 days of past solar data.
In my opinion, however, it only describes that at least 28 days of past solar and consumption data must be available. It does not state that this solar production and consumption must occur during active DESS. Why should this be the case? Both solar production and consumption are always available regardless of whether DESS is active or not.
And even when DESS is deactivated, a normally quite useful consumption forecast is actually displayed. But only as long as you don’t have any short-term unusual consumption. Because these immediately “destroy” the consumption forecast.
The consumption forecast only returns to normal once a few hours have passed after such short-term exceptional consumption. But if you have DESS active during this time, too much will be used from the grid due to the excessive consumption forecast.
Examples of short-term exceptional consumption:
Charging an electric car
Starting an electric fan heater
Switching on the tumble dryer
etc.
Exactly! And in these one, two or three hours until the consumption forecast returns to normal, DESS then charges from the grid, although this would often be absolutely unnecessary…
Just to let you know that we are looking into this. For the Node-RED implementation we already are testing with a change in the model that “forgets” about higher consumption hours quicklier.
If we don’t encounter unexpected behavior with that, we should be able to roll that out on the VRM implementation soon (probably tomorrow).
Another remark on the the system making bad choices based on the consumption forecast being off. We’ve ran simulations on this before and the actual impact of consumption forecast being off was not that significant (a few euro’s difference on a yearly base). That doesn’t mean that we shouldn’t strive to get it right, but it puts it a bit into perspective.
Thank you for taking on the problem. However, I would like to make 1-2 comments. I don’t think that’s enough, higher forecasts are forgotten more quickly. The PROBLEM is that high consumption in the short term has an impact on the entire coming period. 1-2 practical examples. If you cook now, the probability that the stove will be on in the next 1-2 hours is high. But that doesn’t mean that you’ll be cooking for the next 24 hours. And that’s exactly what the system does, it increases the total consumption for the entire time interval. It would be more correct to increase consumption for the next 1 or 3 hours, but leave the consumption further in the future the same. It would be similar with charging an electric car. With the current charging process, it is logical that it is very likely that this process will last a few hours, but not for the next 24 hours. Give it a few logical thoughts. Thank you. Another important point in my view is that many of us actively practice load management. This means that a lot of consumption is now switched on when PV production is high or prices are low. Perhaps in the future the logic could be integrated into the consumption forecast. Consumption forecasts that take into account the factors PV production, electricity price and Soc of the storage.
Another thought. I spend a lot of time on weather forecasts. Modern forecasting systems can predict the weather 5 days into the future very accurately. With longer forecasts, the accuracy decreases. In meteorology, the long-term average is included in the forecast. This means that the less accurate the forecast, the higher this average is included. The converse means that a weather forecast for the period in 4 weeks is actually only based on the long-term average. In relation to the consumption forecast in the DESS, this would mean: In the current hour there is high or low consumption, which means that for the next 1-2 hours there is a probability that consumption will also tend in this direction, but the further I look into the future, the greater the probability that it will tend towards the average of comparable days of the week. Please do not take this as general criticism. I am still enthusiastic about the basic idea of ​​the DESS.
Monday afternoon we switched the consumption forecast for the systems 234606, 310721, 440066, 202328 and 152012 to shorten the effect of high loads. Not sure if you noticed. If so, is this improving things for your situation?
I don’t know exactly whether my system is affected because I don’t know what to do with the system number. However, I haven’t noticed any extreme fluctuations for a few days. I just charged the electric car and at 5 p.m. in the latest forecast there was no extreme reaction. So far the values ​​are realistic. I’m still waiting for the 6 p.m. forecast
I have mostly letters at the end, so it didn’t make sense to me. However, I think I’ve noticed a different behavior in the forecast. I’m curious.
