Support parsing of v33 data

[nickrobinson251]

• close Support version 33 of the format #31
• close Support multiple owners (oi, fi columns) in Generators, Branches, Transformers #45

[codecov-commenter]

Codecov Report

Merging #47 (08cd885) into main (ce4e3a9) will increase coverage by 0.29%.
The diff coverage is 98.33%.

@@            Coverage Diff             @@
##             main      #47      +/-   ##
==========================================
+ Coverage   94.26%   94.56%   +0.29%     
==========================================
  Files           3        3              
  Lines         279      331      +52     
==========================================
+ Hits          263      313      +50     
- Misses         16       18       +2     

Impacted Files	Coverage Δ
src/types.jl	95.19% <97.14%> (-0.32%)	⬇️
src/parsing.jl	98.61% <98.82%> (-0.28%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ce4e3a9...08cd885. Read the comment docs.

[nickrobinson251]

This now parses a v33 file

However the changes on this branch reduce performance of the code on non-v33 files:

On this branch:

julia> using PowerFlowData

julia> @time parse_network("raw/realistic29.raw");
  1.348236 seconds (1.06 M allocations: 138.400 MiB, 2.10% gc time, 82.74% compilation time)

julia> @time parse_network("raw/realistic29.raw");
  0.219198 seconds (1.27 k allocations: 75.141 MiB, 10.57% gc time)

On main

julia> using PowerFlowData

julia> @time parse_network("raw/realistic29.raw");
  0.601938 seconds (331.93 k allocations: 90.757 MiB, 5.39% gc time, 63.28% compilation time)

julia> @time parse_network("raw/realistic29.raw");
  0.196851 seconds (1.13 k allocations: 69.327 MiB, 4.45% gc time)

i'm inclined to leave this to future work as this is already a huge PR and likely still needs some other cleanup (as well as tests)

[nickrobinson251]

Some timings, using a v33 file in https://github.com/NREL-SIIP/PowerSystemsTestData/
Running each one twice, so we can see with and without the compilation time of the first run.

PowerModels.jl (v0.18.3)

(with logging disabled via Memento.setlevel!(Memento.getlogger(PowerModels), "error"))

julia> @time PowerModels.parse_psse("PowerSystemsTestData/ACTIVSg70k/ACTIVSg70k.RAW");
 97.266027 seconds (695.57 M allocations: 15.219 GiB, 5.58% gc time, 2.52% compilation time)

julia> @time PowerModels.parse_psse("PowerSystemsTestData/ACTIVSg70k/ACTIVSg70k.RAW");
 97.628370 seconds (692.60 M allocations: 15.049 GiB, 6.54% gc time)

PowerSystems.jl (v1.15.1)

(with logging disabled via Logging.disable_logging(Logging.Error))

julia> @time PowerSystems.parse_psse("PowerSystemsTestData/ACTIVSg70k/ACTIVSg70k.RAW");
 86.812174 seconds (684.53 M allocations: 14.675 GiB, 6.07% gc time, 0.56% compilation time)

julia> @time PowerSystems.parse_psse("PowerSystemsTestData/ACTIVSg70k/ACTIVSg70k.RAW");
 96.440231 seconds (681.18 M allocations: 14.480 GiB, 6.52% gc time)

PowerFlowData.jl (this branch)

julia> @time parse_network("PowerSystemsTestData/ACTIVSg70k/ACTIVSg70k.RAW");
  1.584180 seconds (1.10 M allocations: 162.025 MiB, 82.08% compilation time)

julia> @time parse_network("PowerSystemsTestData/ACTIVSg70k/ACTIVSg70k.RAW");
  0.248645 seconds (1.19 k allocations: 96.073 MiB)

So right now we're >10x faster on first run, and >100x on subsequent runs. We also have >100x fewer allocations on first run, and 10,000x on subsequent runs. And overall allocate ~100x less memory.

[quinnj]

Dang those are some sweet performance improvements!