Machine-learning demand generation from EULP dataset.
This repository consists of 4 main types of scripts. There are variations for these scripts to reflect different cases of application. Below each type of script is defined and
Importantly, the first script depends on the previous downloading of the metadata files, which contain important description on the buildings used in the simulations. The scripts that perform this task are: download_commercial_metadata.py and download_residential_metadata.py.
Across the multiple scripts, the file pds_database_address.txt is needed to access the datalake with the primary information that is then downloaded.
The first type of script the needs to be run is the one with the root name "download_parquets". These scripts will access the datalake and download specific files indicated in the parameterization of the file (thus having different script variations). The scripts will create a folder named with the root "parquet_downloads" containing the multiple parquet files.
The second type of script simplifies and cleans the individual-downloaded parquet files and will generate synthesized parque files named "simple_parquet".
The third type of script has the root name "process_hp_data". This script applies the machine learning algorithms based on the "simple_parquet" dataset generated in the previous step. It then generates the parameters of the machine learning models that then enable a visualization of the system.
The fourth type of script is "visualize_model_apply.py", that depends on viz_config.yaml (one example configuration file is uploaded for reference). This model takes the parameters of the trained model, and it generates multiple archetypes accordingly. It generates plots that illustrate how the model generates a curve as a function of the temperature. This script is preferably applied in specific folders where the machine learning models are applied to specific datasets (e.g., electric backup, fuel backup, and efficient heat pump), generating a result named aggregated_heating_demand_periodic.csv. NOTE: the use would need to set this folders up manually. Other visualization scripts that perform similar tasks are also in the repo. For example, viz_model_append.py generates viz_model_appended.csv as a result of processing the results from the specific folders of each case of interest.