kit-iai / rapidpf

Matlab code to generate distributed power flow problems.

RapidPF - Rapid prototyping for distributed power flow problems

The official documentation can be found here.

RapidPF is Matlab code that facilitates the laborious task of formulating distributed power flow problems.

The mathematical details of the code are described in this arXiv preprint.

rapidpf's Issues

Apply GN/LM methods for solve nonlinear least squares problems (local step)

Currently, the NLP problems in local step are solved as a standard optimization, although the problem is formulated as a least-sqaures problem.

In order to gain more benefit from LS structure, i.e., saving computing time and increasing robustness, Gauss-Newton method and Levenberg Marquadt method will be implemented in rapidPF.

Lower/upper bounds for OPF

ensure compatibility OPF<->PF

the OPF code should not break the working PF code.

Validation for OPF

There are several things we need to validate:

• Does the size of the local state make sense?
• Do all local functions work for the assumed size of the state?
• Is the size of the upper/lower bounds correct?
• etc.

Add subsystem consisting of exactly one generator bus

When trying to add a subsystem consisting of exactly one node, an error occurs in shift_numbering.m

files_issue.zip

Initial conditions for OPF

Generate initial condition for OPF formulations.

Limited GSK value for 300-subsystem

In the TSO-DSO test, The maximum GSK of case300 file from MATPOWER is limited to 0.55. Consequently, simulation result of the 4662-bus system, which includes 2 300-bus systems, shows a limitation of GSK value (also 0.55). The main factors remain unknown.

Possible solution:

• Change connection buses of 300-subsystem

• parititon the 300-subsystem into smaller regions

GSK: a parameter describes the percent of generation shifted from transmission system(s) to distribution system

Numerial Hessian for OPF-Lagrangian

Unfortunately, we can't use matpower's built-in functionality for the Hessian of the Lagrangian.

We need to compute it numerically as the Jacobian of the gradient.

Migrate CI Workflow to Github Action

Migrate the CI Workflow for the documentation to Github Actions

https://github.com/KIT-IAI/rapidPF/blob/49b1b7b58c6919a7346bb4dca02d422beaa51c62/.gitlab-ci.yml

MATPOWER Hessian for Lagrangian

Include MATPOWER hessian function for Lagrangian.
This should be possible by setting the lambda and entries corresponding to the copy nodes to zero.

Termination Conditition of ALADIN Toolbox

The default tolerance parameter is set to 10^{-10}, but some simulations shows a comparable lower accuracy. Main factor might lay in termination condition of ALADIN Toolbox.

Termination condition of ALADIN Toolbox differs from original ALADIN paper in 2 ways:

• Termination check is activated right after NLP local-step in ALADIN paper. In contrast to the original paper, Toolbox uses a while-loop for iterative, which means the termination conditions in the Toolbox acts like a pre-condition for the while-loop.

• The sufficitent and necessary conditions of ALADIN paper consists of feasibility condition

  

  and duality condition

  

  However, there is only feasibility condition in the Toolbox. Lacking duality condition, or more specifically, lacking consideration of penalty parameter might be the main factors for comparable lower accuracy.

This topic needs further investigated.

Lagrangian multipliers in solving PF problem

Currently, it has been observed that all lagrangian multipliers converge to zeros for some cases.

The reason might be: at solution, where the first order necessary condition holds, the residual value, i.e., the value of cost function, is equal to zero.

Nevertheless, we need to investigate the Nullspace of matrix A carefully, before we make a conclusion.

Dimensions for OPF

Extend to optimal power flow

The whole functionality should also work for OPF problems.

Turn off generators at copy buses

For the OPF problem it is convenient to simply turn off the generators at the copy buses (#3). We can do the same, however, for the PF problem. Hence, the functionality should be moved to the case file generator, and not to the OPF parser.

Generate consensus matrices for OPF