HybridACPowerFlow #97
Conversation
…, and tests are passing.
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
Remaining comments which cannot be posted as a review comment to avoid GitHub Rate Limit
JuliaFormatter
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
[JuliaFormatter] reported by reviewdog 🐶
PowerFlows.jl/test/test_powerflow_data.jl
Line 23 in 9a9fc38
JuliaFormatting suggestions Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
src/nlsolve_powerflow.jl
Outdated
| # update jacobian. | ||
| J(nlCache.x) | ||
|
|
||
| converged = |
There was a problem hiding this comment.
move convergence check to a function
There was a problem hiding this comment.
Implemented here
…nto lin-solve-cache
|
Why did you call this Hybrid? |
|
I called it |
| cache.K._symbolic = C_NULL | ||
| @assert cache.rf_symbolic[] == C_NULL | ||
| end | ||
| if cache.K._numeric != C_NULL |
There was a problem hiding this comment.
Is accessing the _symbolic and _numeric fields of K part of KLU's public interface? If not (and my guess is not), find a way to avoid
There was a problem hiding this comment.
This was a subject of discussion with @rbolgaryn. We're doing things this way (directly calling C wrappers, using private fields) for performance reasons: KLU's public interface was not written in a non-allocating manner. That being said, I have not actually measured the performance difference.
There was a problem hiding this comment.
Maybe @jd-lara and @daniel-thom want to weigh in here but I'd be veeery wary of doing this. I guess you could check the Git history to find out how stable in practice some of these things are, but typically when a field starts with an underscore it's to explicitly signal that it should not be relied upon externally….
There was a problem hiding this comment.
Alternatively, how hard would it be to write our own wrapper to replace KLU.jl?
There was a problem hiding this comment.
This is how it looks in the KLU.jl package:
function klu_factor(Ap, Ai, Ax, Symbolic, Common)
ccall((:klu_factor, libklu), Ptr{klu_numeric}, (Ptr{Int32}, Ptr{Int32}, Ptr{Cdouble}, Ptr{klu_symbolic}, Ptr{klu_common}), Ap, Ai, Ax, Symbolic, Common)
end
function klu_solve(Symbolic, Numeric, ldim, nrhs, B, Common)
ccall((:klu_solve, libklu), Cint, (Ptr{klu_symbolic}, Ptr{klu_numeric}, Int32, Int32, Ptr{Cdouble}, Ptr{klu_common}), Symbolic, Numeric, ldim, nrhs, B, Common)
end
function klu_refactor(Ap, Ai, Ax, Symbolic, Numeric, Common)
ccall((:klu_refactor, libklu), Cint, (Ptr{Int32}, Ptr{Int32}, Ptr{Cdouble}, Ptr{klu_symbolic}, Ptr{klu_numeric}, Ptr{klu_common}), Ap, Ai, Ax, Symbolic, Numeric, Common)
end
| @@ -366,28 +366,28 @@ function _update_F!(F::Vector{Float64}, Sbus_result::Vector{Complex{Float64}}, | |||
| return | |||
| end | |||
|
|
|||
| function _update_dSbus_dV!(rows::Vector{Int64}, cols::Vector{Int64}, | |||
| function _update_dSbus_dV!(rows::Vector{Int32}, cols::Vector{Int32}, | |||
There was a problem hiding this comment.
Where did we land on Int64 vs. Int32?
…alculations, filter init voltage guess, implement loss factors for HybridPowerFlow, some additional minor changes. TODO: figure out how to handle the PSSE roundtrip testsn that are failing because of the voltage guess correction.
…rFlows.jl into rb/lin_solve_cache_new_J
Newton AC PF enhancements: loss factor calculation, progress bar option, better filtering of initial guesses, and some code style tweaks for the new `HybridPowerFlow`.
With @rbolgaryn:
LinearSolverCachefor all linear solvers, so there's a uniform interface for symbolic and numeric factoring. Right now, there's 2 subtypes:KLULinSolveCache(used in code) andLULinSolveCache(mainly for testing purposes).HyrbidACPowerFlowinnlsolve_powerflow.jl: a simple Newton linear search, storing the residual and Jacobian as functors, using KLU to invert the Jacobian. This is faster thanNLSolveACPowerFlowby about 25%, though it's a bit less robust (errors at non-invertible J, convergence issues: may want to switch to a trust region method eventually).HyrbidwhereverNLSolveACPowerFlowappears in the tests. These all passed, except for a pair (intest_psse_export.jl: I've commented them out) whereHybridfails to converge. (NLSolve.jl's newton method also fails to converge for that same data, so it's limitation of the method, not a bug.) Also addedtest_klu_linear_solver.jl.