Thanks ... thanks a lot! ... and wow! So the sample submission is in 15m intervals and the outages are 1h intervals?

My goodness, it is like cutting logs of wood with a scalpel. We need a chainsaw for this one ...

🤣🤣

Hi Julius ... nope, this is not funny ... it is a disaster.

Never mind optimising energy sources, no fun with mathematics for us as this challenge purports to be ... rather we are stuck in a literal nightmare where we have to guess if the sun will shine in 15m intervals for the next week in 10 different unknown locations and, guess what, if you guess wrong your submission is rejected.

Yes, This one seemed like a gamble to me, thus I stopped after a while.

What is interesting is that you have to model such that, delta of total energy and solar energy is almost never under predicted.

But I was not able to achieve this, the better the prediction error, the more such cases of underprediction, the worse the predictions, the higher the score on leaderboard :)

I am excited to look at the top solutions.

your strategy discharges the batteries too much, @skaak in the test period, not funny yea, i know from experience😅

🤣🤣🤣. i AM JUST LAUGHING IT HARD WITH ALL YOUR COMMENTS. In fact i can't wait to see those winining solution strategies.

anyone cannot trust his solution that much, we don't know what happen in private data so i have 0.01 chance even to feasible solution both on public and private data.

I joined this one late ... really just to do something while waiting for my models on the traffic forecasting and malaria detection to finish. This one seemed so nice, like a cross word puzzle, and I now have this real nice implementation to switch between power sources which I really want to work on some more. I created just the simplest baseline forecast for the solar production and power consumption, but nothing works and I spend more and more time on that and no time at all on the optimiser.

This sucks the fun from this one in a big way for me. By now I have pages of unfeasable strategy and I just work on the stupid baseline.

Oh well ... but look at all the replies here ... it reads like the whos who in zindi. Thanks guys! <3 <3 <3

I will probably continue this conversation on the traffic forecast discussions, where many of you are also playing, but over there also I think it is a bit of a gamble, because the test set is not random and as a result, that one can easily shake up big time in private ... tbc

If they had penalized the infeasible cases in the objective function, the problem would have been solved.

Exactly @ademakgul, heavy penalties on scores rather than the unfeasible error

I'm facing the same issue. I doubt my prediction has a high margin of error. I've tried different error measures (MAE, MSE, etc.) for better results, but any strategy I try to create, including most changes in the diesel motor state, does not alter the outcome. Each time, I get the message "The strategy is unavailable".

What is your best MAE,MSE ?

I've trained each site individually for both consumption and solar energy forecasting. For example, for the first site, I achieved an approximately 0.034 MAE