Pyomo.jl

This package is an interface to Pyomo, a Python package for nonlinear optimization and solving DAEs. More information is available at the Pyomo docs.

In particular, the purpose of this package is to interface Pyomo with Symbolics.jl. In this way, Symbolics can be used to build Pyomo expressions that can then be turned into Julia functions using Symbolics.build_function, and then to Python functions using PythonCall.pyfunc.

Please note that this repo is not affiliated with the Pyomo developers.

pyomo.environ is imported as pyomo, pyomo.dae is imported as dae, and pyomo.opt is imported as opt.

Symbolics Example

The motivation of this package is largely that there needs to be a way to write a Pyomo expression as a symbolic expression if one wants to compile symbolic models like those specified in ModelingToolkit to Pyomo. Below is a sketch of what this looks like.

using Pyomo, Symbolics

@variables t MODEL_SYM::SymbolicConcreteModel

model.t = dae.ContinuousSet(bounds = (0, 1))
model.x = pyomo.Var(model.t)
model.y = pyomo.Var(model.t)
model.dx = dae.DerivativeVar(model.x, wrt = model.t)
model.dy = dae.DerivativeVar(model.x, wrt = model.t)

# Use symbolic indexing to build Pyomo expressions returned by the functions.
# The type of variables in these equations is a Symbolic{PyomoVar}
prey_eq = model.dx[t] ~ 1.5*model.x[t] - model.x[t]*model.y[t]
predator_eq = model.dy[t] ~ -3*model.y[t] + model.x[t] * model.y[t]

prey_f = eval(Symbolics.build_function(prey_eq, MODEL_SYM, t))
pred_f = eval(Symbolics.build_function(pred_eq, MODEL_SYM, t))

model.deq1 = pyomo.Constraint(model.t, expr = Pyomo.pyfunc(prey_f))
model.deq2 = pyomo.Constraint(model.t, expr = Pyomo.pyfunc(pred_f))
...

Solvers

Pyomo.jl exposes a few different constructors for the discretizations in Pyomo:

• BackwardEuler(): 'dae.finite_difference' with 'BACKWARD'
• MidpointEuler(): 'dae.finite_difference' with 'CENTRAL'
• ForwardEuler(): 'dae.finite_difference' with 'FORWARD'
• LagrangeRadau(n): 'dae.collocation' with 'LAGRANGE-RADAU' and ncp = n
• LagrangeLegendre(n): 'dae.collocation' with 'LAGRANGE-LEGENDRE' and ncp = n