actual_dispatch added to target_dispatch when the previous target_dispatch is 0 but actual_dispatch is not

[richardwth]

Environment

• Grid2op version: 1.7.2
• System: ubuntu18.04

Bug description

When obs.target_dispatch is 0 and obs.actual_dispatch is not 0 at time t, the obs.actual_dispatch will be added to obs.target_dispatch given an redispatch action at the next step.

How to reproduce

from grid2op import make
from lightsim2grid import LightSimBackend

backend = LightSimBackend()
env = make(".../L2RPN_wcci2022_starting_kit/input_data_local", backend=backend)

env.set_id(0) 
obs = env.reset()
env.fast_forward_chronics(10)
obs = env.get_obs()
gen_id = 2

# Step 11
act = env.action_space()
act.redispatch = [(gen_id, -3.5)]
obs, reward, done, info = env.step(act)
print('step {}: id {}, gen {:.1f}, actual {:.1f}, target {:.1f}, forecast {:.1f}'.format(
    obs.current_step, gen_id, obs.gen_p[gen_id], obs.actual_dispatch[gen_id], obs.target_dispatch[gen_id], obs.get_forecasted_inj()[0][gen_id]))

# Step 12
act = env.action_space()
act.redispatch = [(gen_id, +3.5)]
obs, reward, done, info = env.step(act)
print('step {}: id {}, gen {:.1f}, actual {:.1f}, target {:.1f}, forecast {:.1f}'.format(
    obs.current_step, gen_id, obs.gen_p[gen_id], obs.actual_dispatch[gen_id], obs.target_dispatch[gen_id], obs.get_forecasted_inj()[0][gen_id]))

# Step 13
act = env.action_space()
act.redispatch = [(gen_id, +1.5)]
obs, reward, done, info = env.step(act)
print('step {}: id {}, gen {:.1f}, actual {:.1f}, target {:.1f}, forecast {:.1f}'.format(
    obs.current_step, gen_id, obs.gen_p[gen_id], obs.actual_dispatch[gen_id], obs.target_dispatch[gen_id], obs.get_forecasted_inj()[0][gen_id]))

Current output

step 11: id 2, gen 342.7, actual -3.5, target -3.5, forecast 348.6
step 12: id 2, gen 346.2, actual -2.5, target 0.0, forecast 353.8
step 13: id 2, gen 349.7, actual -1.4, target -1.0, forecast 355.7

Expected output

step 11: id 2, gen 342.7, actual -3.5, target -3.5, forecast 348.6
step 12: id 2, gen 346.2, actual -2.5, target 0.0, forecast 353.8
step 13: id 2, gen 349.7, actual -1.4, target 1.5, forecast 355.7

[BDonnot]

Hello,

Thanks for reporting this issue.

This is indeed an incorrect behaviour and the setpoint should indeed be +1.5 (as you mentionned) after the third action here. I'll see where the bug occurs and try to fix it "asap" (though i'm not sure when)

Benjamin