analyse Dvorkin et al paper, present short analysis/comparison about powermodelsprivacypreserving.jl

[MingDing2019]

Which parameters are they trying to obfuscate?
[Ming]: The power flows f in the optimal OPF solution (see (3b)). Then, g and u need to be adjusted accordingly using (3a) and (3c).

Which data do they need? Is it the same?
[Ming]: The authors used impedance z = r + jx, while in our program we used y = g + jb.

Gaussian vs Lapacian? What are the implications?
[Ming]: Gaussian implies a slightly relaxed version of differential privacy. The impact on implementation is minimum since it is just an alternative distribution for noise.

Shunt admittance obfuscated?
[Ming]: It does not seem to be obfuscated. See (3c).

Methodology? How does it compare?
[Ming]: For a fair comparison, I suppose we should use the same form of noise for both methods and choose the same privacy budget parameters. Then we can compare utility in terms of operating cost.

[davidsmith2020]

It is more different than this (re comments above, I think) -- distribution low-voltage customer, power flows and power outputs, rather than transmission side admittance. I think shunt admittance is implicitly obfuscated. It is very different, and Gaussian required therein due to affine requirement (and no post-processing therein needed)

[afeutrill]

• Which parameters are they trying to obfuscate?
  • Power flow, enforces some affine perturbations from the random vector
  • Power output
  • Voltage magnitude
• Which data do they need? Is it the same?
  • Not sure if the data is contained in the branch information or not
• Guassian vs Lapacian? What are the implications?
  • Gaussian
  • Not sure of the implications exactly, would have smaller tails than Laplace I think
• Shunt admittance obfuscated?
  • Not that I can see, shunt not mentioned
• Methodology? How does it compare?
  • Quite different methodology. Seems to be more high level, or general just looking really at perturbing the power flow
• CC-OPF Minimises cost of transmission
• ToV-CC-OPF minimises the above plus some weighted standard deviations
• TaV-CC-OPF minimises like CC-OPF plus weighted sum of difference between original standard deviation and resulting standard deviation