Clarification on current-based sensitivities for voltage and transformer loading in EV2Gym

[bumj00712]

Hi EV2Gym developers,

I’ve confirmed from the environment code that the actions in EV2Gym directly control current (through scaling with max_charge_current or max_discharge_current), which is then converted to power for the load or generator elements.

I’d like to clarify one point regarding grid sensitivity analysis for voltage and transformer loading constraints:

Given that the control variable is current, would it be more accurate within EV2Gym to:

1. Compute sensitivities directly as ∂V/∂I and ∂(Loading%)/∂I by perturbing current magnitudes; or
2. Continue using ∂V/∂P and ∂(Loading%)/∂P, and convert them via
   ∂I∂P​=V* (phases)^1/2* PF
   In other words — does the internal model handle current-to-power conversion linearly enough that both approaches are effectively equivalent, or would direct current perturbation yield more stable and realistic responses in the EV2Gym grid simulation?

[StavrosOrf]

Hello,

Thank you for your continous engagement with EV2Gym!

The current appraoch in EV2Gym is quite simple, we assume that we control the charging current but we dont take into account the bus voltage to calculate the charge power (we assume that the voltage is always nominal at the charger level). Then after the charging power is calcualted we add it up ass another active power element.

It would be interesting to see your suggested approach implemented. We would appreciate if you could implement it and make a pull request with the changes, so everyone could benefit by these improvements.

[bumj00712]

I’m integrating EV2Gym EV logic into a pandapower/CIGRE workflow and want to confirm the intended grid-side representation of EV charging/discharging power.

From the code:

In ev2gym/models/grid.py, step() adds actions only to active_power before running the Laurent/tensor power flow; reactive_power is computed as base_load × pf and is not updated to reflect EV charging/discharging. This makes EV impact P-only in the tensor PF.
EV_Charger.step() (ev2gym/models/ev_charger.py) produces current_power_output in kW; there’s no explicit Q calculation.
The Gurobi baselines (e.g., baselines/gurobi_models/v2g_grid.py) aggregate EV charging/discharging to transformer power and couple it to the voltage model via total active power per bus; Q is supplied externally.
Question:

For users integrating EV2Gym into external grid solvers (e.g., pandapower), should EV charging/discharging be represented as:
P-only injection/withdrawal at the bus; or
P+Q with a specified power factor (e.g., q = p · tan(arccos(pf))), applied for both charging and V2G discharging?
Follow-ups:

If P-only is the intended assumption (consistent with “assume nominal voltage at the charger and add as active power”), do you discourage assigning any reactive power for EVs in external solvers?
If P+Q is acceptable/preferred in external solvers, do you recommend a fixed PF (which value?), or any guidance on how to choose PF by mode (charge vs discharge) or current level?
For consistency with EV2Gym’s internal assumptions and the paper cited in your repo, which approach would you consider “correct” for reproducing results?
Thanks a lot for the guidance!

[StavrosOrf]

Questions:

1. For users integrating EV2Gym into external grid solvers (e.g., pandapower), should EV charging/discharging be represented as:
   P-only injection/withdrawal at the bus; or P+Q with a specified power factor (e.g., q = p · tan(arccos(pf))), applied for both charging and V2G discharging?

Common EV chargers have a very small reactive power component as shown in Figure 4 of https://www.nature.com/articles/s41597-025-05524-5, so this is why we didn't include it in the grid voltage calculation. In your case, you can also include this reactive power component to improve realism in the grid solver.

However, keep in mind there are reactive power-injecting chargers that can convert a higher amount of active power to reactive to support grid use cases, such as voltage support, etc.

2. If P-only is the intended assumption (consistent with “assume nominal voltage at the charger and add as active power”), do you discourage assigning any reactive power for EVs in external solvers?

No, you can add a small reactive power component similar to the paper https://www.nature.com/articles/s41597-025-05524-5

3. If P+Q is acceptable/preferred in external solvers, do you recommend a fixed PF (which value?), or any guidance on how to choose PF by mode (charge vs discharge) or current level?

Yes the most realistic assumption is to assume a different efficiency/ PF for charging and discharging. Look for example Figure 7 of the paper https://www.nature.com/articles/s41597-025-05524-5

6. For consistency with EV2Gym’s internal assumptions and the paper cited in your repo, which approach would you consider “correct” for reproducing results?

To reproduce the results, you can use the simplest assumption we also used in the paper. However, since you have a more powerful grid solver, I suggest you also add the reactive power component to improve modeling accuracy.