noahrhodes
commented
7 months ago Thank you for sharing! Regarding the convergence issues, we can give some details that may be helpful to you and other people who have had similar issues.
We've found that our approach of turning the GIS line data into powerlines with R, X, and B values resulted in some very short lines and some longer lines with a wide range of parameter values. This is also possible in real grid models, as bus ties or short transmission lines connecting two sides of a substation can be modeled as very low impedance branches. However, this results in very small impedance values which may pose challenges for the stability of numerical methods.
When creating CATS, we tested it by solving the optimal power flow (OPF) problem formulations in the PowerModels.jl package (using Julia/JuMP). While we have not used Pandapower, some other users have also had numerical/convergence issues with CATS. Typically, this arises when numerical methods like Newton-Raphson are used to solve power flow, rather than using optimization with an optimization solver like Ipopt, HiGHS, or Gurobi. Optimization solvers can scale parameters to mitigate numerical issues (either by default or by setting solver attributes), or use higher precision number formats but note that this may result in longer solve times.
PowerWorld is a commercial OPF solver, and it has a zero branch resistance (ZBR) threshold, where it will combine buses connected by branches below this threshold, thus eliminating small impedance lines. The default threshold in PowerWorld is 2e-4.
Another user group, using optimization in Python/Pyomo, has found the following methods to be helpful:
- Remove buses with node degree 2 that do not contain generation, load, or reactive power compensation.
- Remove lines with impedance values below some threshold (as described above in PowerWorld). It is also possible to disregard line losses below a certain proportion of line capacity, thus imposing thresholds on resistance that are not the same for all branches.
We are debating whether to modify CATS to address this challenge. We will keep the current grid model as is, to maintain accuracy with the GIS data of California's actual electric infrastructure. However, we may consider creating a 'simplified' model that merges buses and omits short powerlines to improve the numerical qualities of the model.
Let us know if you have any thoughts and how you are considering using this model!
RJEllis-HTC
Developers and anyone else who may pass by
Firstly thanks for the effort in designing this network it is very useful
I have used the built in functions of Pandapower to convert the MATPOWER case to a Pandapower net, which appears to work with no problems - the net looks like any other pandapower net. https://pandapower.readthedocs.io/en/v2.13.1/converter/matpower.html
I am not expecting a fix etc... as this is designed for MATPOWER, but I wondered if anyone related to the project had any success with conversion and running in Pandapower.
AC powerflow does not converge, DC powerflow does not appear to work as expected. Built-in OPF algorithms are also problematic with convergence issues.