predict-element-instead-of-sequence

This is the code repository for the article "Predict the element instead of the sequence: ResNet surrogate method for very accurate predictions of hourly building energy" in the Applied Energy journal.

Installation

To install and test the code yourself, follow these steps:

1. Clone this repository to your local disk.
2. Install EnergyPlus V24-1-0. NOTE: Make sure that the path to the EnergyPlus executable is correctly defined in the run method of the Building class in pred_elem_seq/simulation.py. The path is set to C:/EnergyPlusV24-1-0/energyplus.exe by default.
3. Install the uv Python package and project manager.
4. Run the following two commands in the predict-element-instead-of-sequence root directory:
   1. uv venv
   2. uv pip install -r pyproject.toml.

Running the code

Even though it is not common practice, we have organized the various implementations of our method into pytest functions. The arguments of the pytest functions are organized in the tests/conftest.py module as pytest fixtures. Feel free to modify the values of the fixtures to test different arguments (such as surrogate model hyperparameters). Use your favourite IDE (or use the command line) to run the main pytest functions explained below.

Main pytest functions

1. tests/test_surrogate.py:

   1. test_get_surrogate

   This function uses the ann and ann_datasets_train pytest fixtures to train, test, and save a surrogate model of a building energy model (BEM) with only one weather file in the design space.

   The ann fixture is an instance of the AnnConfig dataclass that contains all the metadata necessary to clearly define the type and characteristics of the surrogate model (e.g., number of hidden layers), the details of the BEM (e.g., building parameters in the surrogate design space), and details of the surrogate model training and testing datasets (e.g., number of samples).

   The ann_datasets_train fixture is an instance of the AnnDatasets class that is used to generate the surrogate model training and testing by simulating the BEM through a design of experiment.

   2. test_get_surrogate_weather

   This function uses the ann and ann_datasets_weather_train pytest fixtures to train, test, and save a surrogate model of a building energy model (BEM) with multiple weather files in the design space.

   The ann_datasets_weather_train fixture is also an instance of the AnnDatasets class, like ann_datasets_train, but it also uses weather_list as an argument, which is a list of .epw file paths that will be randomly selected during the simulations of the BEM.