Error for non-connected and non-active production

[marota]

Error

PyPowsyblError: Generator 'GEN.P6G3': invalid value (0.0) for voltage setpoint (voltage regulator is on)

Can be reproduced using environment "env_dijon_v2_l2rpn" and replacing the init_state.json by the following:

{"set_bus": {"lines_or_id": {"GENISY611": 6, "GENISY615": 4, "GENISY614": 1, "GENISY613": -1, "GENISY612": 2, "GENISY616": 5}, 
                      "lines_ex_id": {"GENISY611": 6, "GENISY615": 4, "GENISY614": 1, "GENISY613": -1, "GENISY612": 2, "GENISY616": 5},
                      "loads_id": {},
                      "generators_id": {"GEN.P6G6": 3, "GEN.P6G1": 2, "GEN.P6G5": 1, "GEN.P6G3": 5, "GEN.P6G2": 4, "GEN.P6G4": 6},
                      "shunts_id": {}}
                      }

Code to reproduce

import grid2op 
from pypowsybl2grid import PyPowSyBlBackend
import pypowsybl as pp

def create_olf_rte_parameter() -> pp.loadflow.Parameters:
    return pp.loadflow.Parameters(read_slack_bus=False,
                                  write_slack_bus=False,
                                  voltage_init_mode=pp.loadflow.VoltageInitMode.DC_VALUES,
                                  transformer_voltage_control_on=True,
                                  use_reactive_limits=True,
                                  shunt_compensator_voltage_control_on=True,
                                  phase_shifter_regulation_on=True,
                                  distributed_slack=True,
                                  dc_use_transformer_ratio=False,
                                  twt_split_shunt_admittance=True,
                                  provider_parameters={"useActiveLimits": "true",
                                                       "svcVoltageMonitoring": "false",
                                                       "voltageRemoteControl": "false",
                                                       "writeReferenceTerminals": "false",
                                                       "slackBusSelectionMode" : "NAME",
                                                       "slackBusesIds" : "N.SE1P1_0#0"})

from grid2op.Backend import Backend
def create_pypowsybl_backend() -> Backend:
    lf_parameters = create_olf_rte_parameter()#create_olf_parameter()#create_olf_rte_parameter()
    return PyPowSyBlBackend(n_busbar_per_sub=6,
                            lf_parameters=lf_parameters)

env_name="env_dijon_v2_l2rpn
PATH = "./"+env_name

backend=create_pypowsybl_backend()
env=grid2op.make(PATH,backend=backend, n_busbar=6,allow_detachment=True)

obs=env.reset()
do_nothing=env.action_space({})
obs.simulate(do_nothing,time_step=0)

[geofjamg]

We try to apply a 0 target v to a connected generator with voltage control on: IIDM model does not allow this.
So here what is the expected behaviour of a 0 target v coming from grid2op ? Gettting the generator out of voltage control ? And when a target v > 0 will came back, reactivate the voltage control ?

[geofjamg]

We try to apply a 0 target v to a connected generator with voltage control on: IIDM model does not allow this. So here what is the expected behaviour of a 0 target v coming from grid2op ? Gettting the generator out of voltage control ? And when a target v > 0 will came back, reactivate the voltage control ?

Thinking again, there is 2 ways to interpret a 0 target v:

• the generator does not control voltage but is still connected and is "PQ" inside the LF
• the generator is disconnected

Which one should I implement ?

[BDonnot]

Hello,

There is currently a bug in the implementation (that I will solve shortly) in grid2op. But this bug lead to something quite unexpected.

When we do "obs.simulate" we suppose that every information needed for the simulation is present in the observation. This used to be true when loads and generators could not be disconnected.

The situation now is that some generators (in this example) are disconnected in the observation. Now suppose that someone tries to reconnect one. The previous voltage "setpoint" of the generator was 0. (as per grid2op) so when you reconnect it the setpoint is still 0. In the normal environment, this is handled by applying the last known voltage. But the last known voltage is not part of the observation. Only the present one...
So what do we do ?

1. we add some information about the "last known stuff" for all observation attributes (gen_p, gen_v, load_p, load_q, shunt_p, shunt_q, storage_p and maybe shunt_bus and topo_vect)
2. we specifically require for every "generator reconnection action" to provide the new voltage setpoint (or we raise an illegal action)
3. we require for each generator reconnection action IN THE SIMULATE (and not in the step) to provide the new voltage setpoint

[geofjamg]

This is one of the reason that for instance in IIDM we have 2 attributes for the same physical value target_v and calculated_v: one is the input and the other one is the output. So we have a voltage input for the generator ready whatever it is connected or not.

[BDonnot]

It makes much more sense in the iidm way. It's the same in lightsim2grid too.

I think what you suggest is to keep the "last known values" also in the observation (this would be the "target_v")

[geofjamg]

It makes much more sense in the iidm way. It's the same in lightsim2grid too.

I think what you suggest is to keep the "last known values" also in the observation (this would be the "target_v")

Yes, last known prod_v would do the same and it would be more convenient for users that forcing to provide a value of prod_v when reconnecting.