Time Series (doc in progress)

The documentation of this section is in progress. It is rather incomplete for the moment, and only expose the most basic features.

If you are interested in collaborating to improve this section, let us know.

Goal

This class aims to make faster (and easier) the computations of the current flows (measured in Amps) at a certain side of a powerline / transformer when the topology is not modified.

It can be used as:

from lightsim2grid import TimeSerie
import grid2op
from lightsim2grid.lightSimBackend import LightSimBackend

env_name = ...
env = grid2op.make(env_name, backend=LightSimBackend())

time_series = TimeSerie(env)
res_p, res_a, res_v = time_series.get_flows(scenario_id=..., seed=...)

# we have:
# res_p[row_id] will be the active power flows (origin side), on all powerlines corresponding to step "row_id"
# res_a[row_id] will be the current flows, on all powerlines corresponding to step "row_id"
# res_v[row_id] will be the complex voltage, on all bus of the grid at step "row_id"

For now this relies on grid2op, but we could imagine a version of this class that can read to / from other data sources (for now please use the more basic lightsim2grid.timeSerie.Computers for such purpose)

Importantly, this method is around 13x faster than simulating “do nothing” (or “one change then nothing”) with grid2op (see section Benchmarks (Time Series) )

Note

A more detailed example is given in the examples\time_serie.py file from the lightsim2grid package.

Warning

Topology and injections

The topology is taken from the initial provided grid and cannot be changed when evaluating a given “time serie”.

Then, the call to time_series.compute_V(scenario_id=…, seed=…) will only read the injections (productions and loads) from grid2op to compute the voltages.

Note

As this class calls a long c++ function, it is possible to use the python Threading module to achieve high efficient parrallelism. An example is provided in the examples\computers_with_grid2op_multithreading.py file.

Benchmarks (Time Series)

Here are some benchmarks made with:

date: 2026-04-21 09:05 CEST
system: Linux 6.8.0-60-generic
OS: ubuntu 22.04
processor: 13th Gen Intel(R) Core(TM) i7-13700H
python version: 3.13.5.final.0 (64 bit)
numpy version: 2.3.5
pandas version: 2.3.3
pandapower version: 3.4.0
pypowsybl version: 1.15.0
grid2op version: 1.12.4.dev0
lightsim2grid version: 0.13.1
lightsim2grid extra information:
- klu_solver_available: True
- nicslu_solver_available: True
- cktso_solver_available: True
- compiled_march_native: False
- compiled_o3_optim: True

This benchmark is available by running, from the root of the lightsim2grid repository:

cd benchmarks
python3 time_serie.py

For this setting the outputs are:

For environment: l2rpn_neurips_2020_track2
Total time spent in "computer" to solve everything: 0.03s (21834 pf / s), 0.05 ms / pf)
    - time to pre process the injections: 0.00s
    - time to perform powerflows: 0.02s (23675 pf / s, 0.04 ms / pf)
In addition, it took 0.00 s to retrieve the current from the complex voltages (in total 20703.1 pf /s, 0.05 ms / pf)

Comparison with raw grid2op timings
It took grid2op (with lightsim2grid): 0.31s to perform the same computation
    This is a 11.3 speed up from TimeSerie over raw grid2op (lightsim2grid)
It took grid2op (with pandapower): 6.47s to perform the same computation
    This is a 232.6 speed up from TimeSerie over raw grid2op (pandapower)
All results match !

In this case then, the TimeSerie module is more than 15 times faster than raw grid2op.

Time Series (doc in progress)

Goal

Benchmarks (Time Series)

Detailed usage