MOEA Usage

MOEA Usage#

Setup of an experiment is carried out in EnergyPLAN, which provides a handy GUI. When a scenario has been configured, experimental parameters are stored in a data file, which path is required to run the algorithm. The decision variables are a subset of the parameters in the input file and the algorithms will change only those values, leaving all the other parameters unchanged.

Define a model#

The word model is used to indicate the problem to be optimized, which is an model of reality defined by the user. A new model should be declared in a new file in the folder moea/models. Each module (i.e., a file in the folder models) can contain several models. To create a new model, declare a new class that inherits from the BaseModel class in moea.models.base_model. A good practice is to group models in modules by similarity of the concepts that they are intended to model.

When a new model is created, to make it available to the function get_model, it must be added to the list of models in the __init__.py file of the models package.

Define an algorithm (optional) (work in progress)#

Custom algorithms can be defined by the user.

TODO...

The list of available algorithms is

PS> moea list-algorithms

When a new algorithm is created, to make it available to the function get_algorithm, it must be added to the list of models in the __init__.py file of the algorithms package.

MOEA interface#

A user can interact with the MOEA via CLI or declaratively in a script. CLI is practical when several runs of the same optimization problems are required and we want to automate this task. Interaction via scripting is the preffered way when developing a new model of an energy system, e.g., to run the algorithm from a script, or when using the MOEA in a Jupyter notebook.

Command line interface#

If the Python environment was created without errors, the application moea should be accessible from the command line when the environment is active.

A minimal guide to the use of moea is available via

PS> uv run moea --help

and help for each of the command can be consulted using

PS> uv run moea COMMAND --help

Run the algorithm#

The run command requires three (mandatory) arguments

  1. the name of the algorithm, which must correspond to the name of the algorithm class,

  2. the name of the model to be optimized, which must correspond to the name of the model class, and

  3. the name of the file containing the parameters for a scenario (w/o the .txt extension does not matter).

The syntax of the command is the following

PS> uv run moea run ALGORITHM MODEL DATA_FILE_NAME

which runs the algorithm with default parameters. The list of optional MOEA parameters, e.g., number of individuals in the initial population, number of generations, etc., can be listed using the --help flag for the run command.

Declarative approach#

The declarative approach requires at least to:

  1. Recall (or declare a new) problem class and initialize it;

  2. Recall (or declare a new) algorithm class and initialize it;

  3. Run the algorithm using the PyMOO function minimize.

Declare a new problem#

import pandas as pd

from moea.models.base_model import BaseModel
from moea.energyplan.energyplan163 import EnergyPLAN163


class MyModel(BaseModel):

    energyplan_version = EnergyPLAN163

    def __init__(self, data_file: str | Path, **kwargs):
        # Define variables here

        self.vars = pd.DataFrame(
            columns=['name', 'lb', 'ub']
        )

        # Initialize the parent class
        super().__init__(
            n_var=len(self.vars),
            n_ieq_constr=3,
            n_obj=2,
            xl=self.vars['lb'].values,
            xu=self.vars['ub'].values,
            data_file=data_file,
            **kwargs
        )

    def _evaluate(self, x, out, *args, **kwargs):
        self.energyplan.run(
            inputs=...
        )

        # Parse the output file
        Y = self.energyplan.read_values(
            "CO2-emission (corrected)",
            "TOTAL ANNUAL COSTS",
            ("Maximum", "import"),
            ("Annual", "heat3- balance"),
            ("Minimum", "stab.- load"),
        )

        # Store objective function values
        out["F"] = np.array(Y[:, :2])

        import_constr = ...

        heat_constr = ...

        out["G"] = np.column_stack([
            import_constr,
            stable_load,
        ])

Declare a new algorithm#

Optimize using minimize#

Contents