Solvers¶

.optimize() solves with HiGHS by default, and the models in this guide solve in milliseconds. Real models get bigger: a large fleet, a year of quarter-hours, many scenarios. When a run gets slow, this page shows how to cap its time and accept a small gap, fix a seed for reproducible runs, or switch to a commercial solver.

Before tuning the solver, look at the model: its binaries grow with the number of steps times the number of scenarios. See Horizon and problem size, and the binaries table of the Mathematical Formulation.

Supported solvers¶

HiGHS is bundled as a core dependency, so it's always available. The other solvers require installing optional extras, and Gurobi and CPLEX also need a license.

Installation¶

Install the optional solvers as extras:

pip install odys[gurobi]
pip install odys[cplex]
pip install odys[scip]
pip install odys[gurobi,cplex,scip]  # all solvers

uv add odys[gurobi]
uv add odys[cplex]
uv add odys[scip]
uv add odys[gurobi,cplex,scip]  # all solvers

You only need to install the solvers you plan to use. If you request a solver that isn't installed, Odys raises OdysSolverError with a list of available solvers.

Selecting a solver¶

The snippets on this page solve the first system of Your first model:

from datetime import timedelta

from odys import AssetPortfolio, EnergySystem, FixedLoad, Generator, LoadProfile, Scenario

ccgt = Generator(name="ccgt", nominal_power=100.0, variable_cost=50.0)
factory = FixedLoad(name="factory")

energy_system = EnergySystem(
    portfolio=AssetPortfolio([ccgt, factory]),
    scenarios=Scenario(profiles=(LoadProfile(load=factory, values=[60, 90, 40, 70]),)),
    timestep=timedelta(hours=1),
    number_of_steps=4,
)

Pass a SolverConfig to EnergySystem.optimize():

from odys import SolverConfig, SolverName

result = energy_system.optimize(SolverConfig(solver_name=SolverName.HIGHS))

When you don't pass a SolverConfig, Odys uses HiGHS with default settings. To use another solver, install it and change solver_name, for example SolverName.GUROBI.

Configuration options¶

SolverConfig accepts these common options, automatically translated to each solver's native parameter names:

Options left at None keep each solver's own default, so presolve runs unless you set presolve=False. For example, to stop after a minute or within 1 % of the best possible plan, whichever comes first:

result = energy_system.optimize(
    SolverConfig(
        solver_name=SolverName.HIGHS,
        time_limit=60.0,
        mip_rel_gap=0.01,
        threads=4,
    )
)

A run that stops on its time limit or its gap may return a plan that's feasible but not proven optimal. result.termination_condition says why it stopped; see Solver status.

Solver-specific options¶

For parameters not covered by the common options, use the solver_options dict to pass raw solver-native parameters. For example, HiGHS's random seed and MIP feasibility tolerance:

result = energy_system.optimize(
    SolverConfig(
        solver_name=SolverName.HIGHS,
        solver_options={"random_seed": 42, "mip_feasibility_tolerance": 1e-7},
    )
)

With Gurobi, the same dict takes Gurobi's names, such as {"Method": 2, "Cuts": 2}.

solver_options is passed to the solver as it is, so it takes the solver's own parameter names, and its values override the translated common options when both set the same parameter.

Native parameter names¶

Each solver uses different names for the same option. Here's how the common options map:

In solver_options, always use the native names: a common name such as time_limit only works there for HiGHS, whose names happen to match. Check the solver's own documentation for the exact parameter names and valid values.

Availability check¶

Before solving, Odys verifies the requested solver is installed. If it's not available, you'll get an OdysSolverError listing all installed solvers:

from odys import OdysSolverError

try:
    result = energy_system.optimize(
        SolverConfig(solver_name=SolverName.GUROBI)
    )
except OdysSolverError as e:
    print(f"Solver not available: {e}")

This makes it safe to write code that tries multiple solvers or gracefully falls back to HiGHS.

Next steps¶

A run can also fail before the solver starts, because an input breaks a rule. Next, see what odys checks, in Validation and errors.