Line disconnections while loading chronic scenarios

[NMegel]

Environment

• Grid2op version: 0.9.4
• System: windows 10

Bug description

Line disconnections while loading some chronic scenarios, meanwhile parameter NO_OVERFLOW_DISCONNECTION is activated This phenomenon keeps happening on other chronic scenarios (3, 5) but not on 0 and 5. The disconected line is not always the fourth as it happens in the following example. Suppressing chronic scenarios to change their position does not change anything to the outputs

How to reproduce

Run the following piece of code and ensure that you load l2rpn_2019 network files from ExpertOp4Grid repository (abstraction branch): https://github.com/mjothy/ExpertOp4Grid/tree/abstraction/alphaDeesp/ressources/parameters/l2rpn_2019

Command line

Code snippet

import grid2op
from grid2op.Parameters import Parameters

# Chose the number of the chronic to consider
chronic_scenario = 1 # Other disconnections: 3. No disconnection with 0 or 2

# Load environment
parameter_folder = r'.\alphaDeesp\ressources\parameters\l2rpn_2019' # Or any path to the network from ExpertOp4Grid repository (abstraction branch)
custom_params = Parameters()
custom_params.NO_OVERFLOW_DISCONNECTION = True
env = grid2op.make(parameter_folder, param = custom_params)

# Go to chronic scenario
env.set_id(chronic_scenario)
env.reset()

# Print line status
obs = env.get_obs()
print(obs.line_status)

Current output

[ True  True  True False  True  True  True  True  True  True  True  True True  True  True  True  True  True  True  True]

Expected output

[ True  True  True TrueTrue  True  True  True  True  True  True  True True  True  True  True  True  True  True  True]

[Tezirg]

Hello, thank you for reporting this issue, unfortunately I did not managed to reproduce it.

That data you are referring to is in a private repository, so I'm assuming it is a copy of the l2rpn_2019 dataset provided in this repository (We cannot support manually edited data) .

import grid2op
from grid2op.Parameters import Parameters

chronic_scenario = 1

# Load environment
parameter_folder = 'l2rpn_2019'
custom_params = Parameters()
custom_params.NO_OVERFLOW_DISCONNECTION = True
env = grid2op.make(parameter_folder, param = custom_params)

# Go to chronic scenario
env.set_id(chronic_scenario)
env.reset()

# Print line status
obs = env.get_obs()
print (obs.line_status)
print (grid2op.__version__)

My output is :

[ True  True  True  True  True  True  True  True  True  True  True  True
  True  True  True  True  True  True  True  True]
1.1.1

As you can see, I am using grid2op latest, it is recommended to stay up-to-date with the framework. Especially when we hit a major release version.

On top of my head, here are other sources for lines disconnections:

• Hard overflows: see Parameters.HARD_OVERFLOW_THRESHOLD
• Maintenance: Does not happen with l2rpn_2019
• Adversarial attacks : see Opponent Modeling, doesn't happen with l2rpn_2019

[Tezirg]

Update: Thank you for providing access to the dataset referenced in this issue.

Line disconnections are due to maintenances operations: cat ./alphaDeesp/ressources/parameters/l2rpn_2019/chronics/b/maintenance.csv | head -n 2

1_2_1	1_5_2	2_3_3	2_4_4	2_5_5	3_4_6	4_5_7	4_7_8	4_9_9	5_6_10	6_11_11	6_12_12	6_13_13	7_8_14	7_9_15	9_10_16	9_14_17	10_11_18	12_13_19	13_14_20
0.0	0.0	0.0	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0

However, there is something weird with this dataset, as the observation's maintenance-related variables does not seems to contains correct information. (Compared to our l2rpn_wcci_2020, l2rpn_neurips_2020_track2_small)

import grid2op
from grid2op.Parameters import Parameters
custom_params = Parameters()
custom_params.NO_OVERFLOW_DISCONNECTION = True
env = grid2op.make("./", param = custom_params)
env.set_id(1)
obs = env.reset()
print (obs.line_status)
print (grid2op.__version__)
print (obs.time_next_maintenance)
print (obs.duration_next_maintenance)

[ True  True  True False  True  True  True  True  True  True  True  True
  True  True  True  True  True  True  True  True]
1.1.1
[  7 105  56  32  38  12  63   4  27  99  67  39  16   7  30  62  30   3  10   1]
[  1   1   1 -31   1   1   1   1   1   1   1   1   1   1   1   1   1   1   1   1]

[NMegel]

Yes it was due to maintenance, thanks !