Survey of Non-Linear Optimization Modeling Layers in Julia

[ccoffrin]

For anyone who is following this thread, there is a new modeling layer in Julia called ExaModels, which has shown very good performance on the AC-OPF problem, see,

AC Optimal Power Flow in Various Nonlinear Optimization Frameworks

Thanks to recent contributions from @sshin23, a new rosetta-opf model has been added based on the ExaModels framework. As the detailed results below show, this implementation tends to be the fastest Julia AC-OPF implementation I have seen so far, specially at the largest scales. Amazing work @sshin23! This table also includes updates to the NLPModels to use the ConcreteStructs that were proposed in the SciML Benchmarks and an improvement to the Julia JIT warmup procedure and other small mainten…

You may be interested to check it out. In the same thread @sshin23 provides a nice introduction to ExaModels and the kind of use-cases where it shines.

AC Optimal Power Flow in Various Nonlinear Optimization Frameworks

Thanks, @ccoffrin for an amazing project and for sharing detailed results. Here’s a quick summary on how ExaModels works. What is ExaModels? ExaModels.jl is an algebraic modeling system embedded in Julia Language, which is tailored for what we call “SIMD abstraction”. By itself, ExaModels doesn’t have optimization solver capability. This means that you need to interface it with an optimization solver (e.g., Ipopt) to solve problems. ExaModels provide a user front-end to specify the nonlinear m…

@odow can you update the table in the first post to include an entry for ExaModels?