CO2

[rainbownmm]

CO2

[rainbownmm]

I have just done some research on this and you see that the project does not use the RL model for prediction, right? You just calculated his loss value, maybe I misunderstood it, I hope you can reply, thanks!

[enjeeneer]

@Puzzledyy Sorry I'm just replying, I completely missed this. The CO2 data is taken from here, and matches the true data for the year simulated by Energym. It's appended to the state space and fed to the agent at each timestep. PEARL does predict grid CO2 in its rollouts, but of course, these predictions are conditioned on the grid CO2 data it has seen in previous states.

I'm updating the paper to make this more clear, and I should really add this to the README because you're not the first person to raise this. Thanks for checking out the repo!