search

JieFeng-cse / Stable-DDPG-for-voltage-control

Official implementation for the paper
MIT License
31 stars 4 forks source link

Results dont coincide with paper #3

Closed HafizRiaz closed 1 month ago

HafizRiaz commented 1 month ago

Dear Author, after running the provided repo, required results could not be obtained. Plz help me to get the results

JieFeng-cse commented 1 month ago

Thank you for your interest in our work. The checkpoints and test code are provided in this repo. Please make sure you use the metrics we used to get the same result. While exact results may be slightly different due to randomness, almost the same results can be reproduced with test.py. Specifically, to reproduce the table result, use the following functions in test.py

test_suc_rate('safe-ddpg',step_num=100) #safe-ddpg
test_suc_rate('linear',step_num=100)
test_suc_rate('ddpg',step_num=100)

which gives the table results, note that those are calculated only considering the cases where the voltage is successfully restored. You can check the comments in the code to get the result considering all 500 tested trajectories.

image

I just tested all checkpoints, they align well with our table results. I acknowledge that training with DDPG can be quite tricky. While the hyperparameters I used are presented in the repo, those hyperparameters for training may still yield a different training trajectory due to the randomness of RL and probably the difference in machines or environments. As a result, it might require tuning the hyperparameters to get similar or better results if you train from scratch.

HafizRiaz commented 1 month ago

Thanks for your valuable response. Actually i want to learn from your Repo about, how to run a time series power flow for 3 phase, 13 bus unbalanced system? You have modeled it using OPenDSS, is there any guideline to midel it using Pandapower? i also found your query to the author of pandapower repo while searching on web for the same result. Waiting for your positive response.

JieFeng-cse commented 1 month ago

Hi Hafiz, I indeed tried to do that as Pandapower is easier to use. However, I did not find a clear solution for that, as a result, I used OpenDSS instead. It would be more efficient to directly ask the developers of Pandapower for a specific guideline. Thank you.

HafizRiaz commented 1 month ago

Thanks for your kind response. Let me know that if I want to edit your code for a different reward function or some other system ?

Get Outlook for Androidhttps://aka.ms/AAb9ysg

From: Jie Feng @.> Sent: Tuesday, August 27, 2024 11:07:45 PM To: JieFeng-cse/Stable-DDPG-for-voltage-control @.> Cc: Mehboob Riaz @.>; Author @.> Subject: Re: [JieFeng-cse/Stable-DDPG-for-voltage-control] Results dont coincide with paper (Issue #3)

Hi Hafiz, I indeed tried to do that as Pandapower is easier to use. However, I did not find a clear solution for that, as a result, I used OpenDSS instead. It would be more efficient to directly ask the developers of Pandapower for a specific guideline. Thank you.

— Reply to this email directly, view it on GitHubhttps://github.com/JieFeng-cse/Stable-DDPG-for-voltage-control/issues/3#issuecomment-2313205727, or unsubscribehttps://github.com/notifications/unsubscribe-auth/BAK6YL7SE6C55MFMEOBX2OTZTS57DAVCNFSM6AAAAABM77R6O6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMJTGIYDKNZSG4. You are receiving this because you authored the thread.Message ID: @.***>