Skip to content

Latest commit

 

History

History
252 lines (226 loc) · 9.93 KB

NEWS.md

File metadata and controls

252 lines (226 loc) · 9.93 KB
Raw

ExaPF release notes

Version 0.11 (December 5th, 2023)

  • Remove the block-GMRES implementation
  • Remove custom BICGSTAB implementations
  • Use KrylovPreconditioners.jl for the block-Jacobi preconditioner
  • Add scaling for Krylov methods
  • Use KLU per default
  • Added new benchmark scripts for linear solvers

Version 0.10 (November 1st, 2023)

  • Adding extension support for CUDA.jl and AMDGPU.jl
  • Adding block-GMRES implementation

Version 0.9.3 (September 27th, 2023)

  • Add support for SCOPF problem (#274)

Version 0.9.2 (September 25th, 2023)

  • Add support to CUDA.jl 5.*

Version 0.9.0 (April 19th, 2023)

  • Update to KernelAbstractions 0.9 with respective API changes
  • Update to Krylov.jl 0.9, and CUDA.jl 4.1

Version 0.8.0 (August 16th, 2022)

  • API changes
    • Add new BlockPolarForm structure encoding block-structured power grid model
  • Bug fixes & improvements
    • add function load_case to import data directly from ExaData (#257)
    • test ArrowheadJacobian correctness with FiniteDiff (#263)
    • remove all custom SpMM kernels (#264)
  • Documentation
    • Add new tutorials in ExaPF for batch power flow and linear solver (#258)
    • Add doctests in expressionsn (#256 #261)

Version 0.7.2 (May 31th, 2022)

  • Bug fixes & improvements
    • Fix unitinialized arrays in NetworkStack (#248)
    • Fix finite-difference checks in tests (#251)

Version 0.7.1 (May 9th, 2022)

  • New features
    • Add BlockNetworkStack structure for batch evaluation (#241)
    • Add ArrowheadJacobian and ArrowheadHessian for stochastic OPF (#241)
    • define loads as parameters inside NetworkStack (#238)
  • Bug fixes & improvements
    • Migrate to KernelAbstractions 0.8 (#236)
    • Drop support for Julia 1.6
    • Migrate to Krylov.jl 0.8 (#239)
    • Update CI

Version 0.7.0 (March 1st, 2022)

  • API changes
  • New features
    • Add overlapping Schwarz-preconditioner (#220)
    • Full support for multi-generators (#217 #223)
    • Add new benchmark suite (#226)

Version 0.6.0 (November 19th, 2021)

  • New features
    • ExaPF now supports multiple generators per bus [experimental]
    • ExaPF now implements a batch power flow solver [experimental]
    • Add new functions network_operation and power_injection
    • Tests directory has been updated, and all tests are now implemented inside dedicated modules (#156). This allows to test directly any MATPOWER instance, simply as
    TestPolarFormulation(datafile, CPU(), Array)
    • Add benchmark script for power flow solver
  • API changes
    • All AbstractNLPEvaluator structures have been moved to ExaOpt (#191)
  • Bug Fixes & Improvements
    • ExaPF now supports CUDA.jl 3.*
    • IterativeSolvers.jl has been removed from the deps
    • Names have been homogenized in PolarNetworkState.
    • Fix power flow solver when phase shift is activated (#183)
    • ExaPF now supports CUDA.allowscalar(false) (#181)

Version 0.5.0 (April 14, 2021)

AutoDiff

  • New features
    • New AD AutoDiff.Hessian to compute the adjoint-Hessian-vector product of any function. (#99)
    • The submodule AutoDiff has been cleaned and its usage generalized. The constructor of AutoDiff.Jacobian has been moved in src/Polar/derivatives.jl. We could now compute the Jacobian of any constraint function implemented in src/Polar/Constraints/. (#106)
  • Code changes
    • Fix the performance of the seeding by moving it onto the host (#102)

PolarForm

  • API changes
    • Add a new NewtonRaphson object, storing all parameters associated with the Newton-Raphson algorithm. (#88) The signature of the powerflow solver ExaPF.powerflow becomes: 
      algo = ExaPF.NewtonRaphson(tol=1e-10, verbose=1)
ExaPF.powerflow(polar, jacobian_autodiff, buffer, algo)
      instead of 
      ExaPF.powerflow(polar, jacobian_autodiff, buffer; verbose=1, ε_tol=1e-10)
    • The constraints in PolarForm have been renamed, to improve clarity: 
      residual_function -> power_balance
state_constraint -> voltage_magnitude_constraints
power_constraints -> {active_power_constraints,reactive_power_constraints}
  • New features
    • Add support for lineflow constraints in the polar formulation. (#91) The constraint is implemented in src/Polar/Constraints/line_flow.jl, and shares the same signature as the other constraints: 
      m = size_constraint(polar, flow_constraints)
cons = zeros(m)
flow_constraints(polar, cons, buffer)
      The function flow_constraints adds explicit bounds on the absolute value of the complex power flowing through the lines: 
      |S_f|^2 <= S_max^2
|S_t|^2 <= S_max^2
      leading to 2 * n_lines additional constraints in the model.
    • Implement handwritten adjoints for polar formulation's functions (#107)
    • Add a new function adjoint! to compute the adjoint of any constraint. (#109)
    • We could now update inplace the load in PolarForm, with 
      setvalues!(model, PS.ActiveLoad(), values)
setvalues!(model, PS.ReactiveLoad(), values)
  • Code changes
    • The definition of the polar formulation has been moved inside src/Polar/, instead of src/model/Polar/. Now, src/model.jl implements only the generic templates to define a new mathematical model.
    • The constraints corresponding to the polar formulation are now implemented in src/Polar/Constraints 
      src/Polar/Constraints/
├── constraints.jl
├── active_power.jl
├── line_flow.jl
├── power_balance.jl
├── reactive_power.jl
└── voltage_magnitude.jl

Evaluators

  • API changes
    • ReducedSpaceEvaluator no longer stores explicitly the state x in its attribute, but implicitly. Now, the attribute buffer stores all the values needed for the computation of the callback. (#82)
    • The constructor for ReducedSpaceEvaluator has been updated, from 
      function ReducedSpaceEvaluator(
    model, x, u, p;
    constraints=Function[state_constraint, power_constraints],
    linear_solver=DirectSolver(),
    ε_tol=1e-12,
    verbose_level=VERBOSE_LEVEL_NONE,
)
      to 
      function ReducedSpaceEvaluator(
    model;
    constraints=Function[voltage_magnitude_constraints, active_power_constraints, reactive_power_constraints],
    linear_solver=DirectSolver(),
    powerflow_solver=NewtonRaphson(tol=1e-12),
)
  • New features
    • Add a new function ExaPF.optimize! to find the optimal solution of any problem specified as a AbstractNLPEvaluator, with signature: 
      ExaPF.optimize!(optimizer, nlp)
      By default, optimize! supports MathOptInterface solvers out-of-the box: 
      ExaPF.optimize!(Ipopt.Optimizer(), nlp)
      But other optimization procedures could be implemented as well.
    • Implement reduced Hessian for ReducedSpaceEvaluator, SlackEvaluator and AugLagEvaluator. (#93) The reduced Hessian is computed using AD (forward over reverse), with handcoded adjoints. The signature is: 
      hessian!(nlp, H, u)  # evaluate Hessian inplace
hessprod!(nlp, Hv, u, v)  # evaluate Hessian-vector product inplace
    • Add Hessian and Hessian-vector product in MOI wrapper (#100)
    • Bridge ExaPF with (ProxAL)[https://github.com/exanauts/ProxAL.jl]. A new evaluator ProxALEvaluator is introduced, allowing to formulate problems with (fixed) ramping constraints. (#76)
    • Add a new SlackEvaluator to reformulate automatically problems with inequality constraints with equality constraints, by adding additional non-negative slack variables. (#94)
    • Add a new FeasibilityEvaluator to automatically compute a feasible point for any AbstractNLPEvaluator.
  • Fixes
    • Fix type inference on Julia 1.6-rc1. (#98)

Maintenance

  • Update the structure of the test directory to separate the different functionalities: 
    test/
    ├── Evaluators
    │   ├── auglag.jl
    │   ├── interface.jl
    │   ├── MOI_wrapper.jl
    │   ├── powerflow.jl
    │   ├── proxal_evaluator.jl
    │   ├── reduced_evaluator.jl
    │   └── test_rgm.jl
    ├── iterative_solvers.jl
    ├── perf.jl
    ├── Polar
    │   ├── autodiff.jl
    │   ├── gradient.jl
    │   ├── hessian.jl
    │   ├── matpower.jl
    │   └── polar_form.jl
    ├── powersystem.jl
    └── runtests.jl
  • Update the documentation (#101 #134 #149)
  • Update the README.md to include a quickstart guide
  • Update CI to use Julia 1.6 by default
  • Update Krylov.jl to version 0.6.0. (#78)
  • Update KernelAbstractions.jl to version 0.5.
  • Update benchmark script
  • Deprecated scripts have been removed. (#87)
  • Fix typing of attributes in BlockJacobiPreconditioner
  • Fix non-deterministic bug issued by a wrong initilization of KA kernels (#145)

Version 0.4.0 (December 4, 2020)

  • CUDA.jl 2.0 support
  • New API
  • Objective with handcoded adjoints
  • Benchmark scripts (benchmark/benchmark.sh)
  • Krylov.jl BiCGSTAB (LinearSolvers.KrylovBICGSTAB)
  • Eigen BiCGSTAB (LinearSolvers.EigenBICGSTAB)
  • New non-linear evaluators for penalty algorithms (PenaltyEvaluator) and Augmented Lagrangian algorithms (AugLagEvaluator)
  • Reduced space gradient method (scripts/dommel.jl)
  • Penalty and Augmented Lagrangian algorithms (scripts/auglag.jl)
  • Updated documentation