SolarDB

Python API for the SolarDB photovoltaic dataset.

What is This?

This repository contains auxiliary code for accessing the SolarDB dataset. The SolarDB is a photovoltaic dataset consisting of 1-year of data for 16 solar power plants, including power production, weather, weather forecasts, and additional exogenous features. In total, it contains over 40M records.

The SolarDB dataset is a part of a currently in-review paper. For more information, see cphoto.fit.vutbr.cz/solar:

How to Install It?

The solardb package can be simply installed by following the subsequent instructions. First, solardb has the following dependencies:

• Python >= 3.7.1
• numpy
• pandas
• scipy
• scikit-learn
• sqlalchemy

Next, to install the solardb package and any dependencies, run the following:

conda create --name solardb -y python">=3.7.1" numpy pandas \
    scipy scikit-learn sqlalchemy
conda activate solardb
git clone https://github.com/PolasekT/SolarDB ./solardb
cd ./solardb && python ./setup.py install

How to Use It?

The solardb package only contains the code necessary for access to the various data sources. First, download the db files from the projects website. Next, to test if the installation is valid, run the following Python commands:

from solardb.db import SolarDBData

db = SolarDBData("./path/to/solardb.db")

print(f"Total power plant / inverters: {len(db.get_pp_info_df())}")
% Output: Total power plant / inverters: 274

For a quick-start, the SolarDBAssembler and SolarDBEvaluator may be used. First, we calculate history, weather, and prediction DataFrames for the prediction task of <10.5.2019, 11.5.2019) for power plant #8:

import datetime
from solardb.db import SolarDBAssembler

assembler = SolarDBAssembler(db)

history_df, weather_df, prediction_df = assembler.prepare_prediction(
    pp_id=8,
    dt_start=datetime.datetime(2019, 5, 10),
    dt_end=datetime.datetime(2019, 5, 11),
    weather_scheme="realistic", history_cnt=288,
)

print(f"DataFrame lengths: {len(history_df)}, {len(weather_df)}, {len(prediction_df)}")
# Output: DataFrame lengths: 288, 288, 288

Next, we calculate the predictions, filling the prediction_df DataFrame. In this example, our model just it with a constant value:

prediction_df.fillna(value=42, inplace=True)

Finally, we can evaluate the efficacy of the proposed prediction model:

from solardb.db import SolarDBEvaluator

evaluator = SolarDBEvaluator(db)

evaluation = evaluator.evaluate_prediction(
    prediction_df=prediction_df,
    history_df=history_df,
    weather_df=weather_df
)

print(f"Prediction error = {evaluation['power']['error'] * 100.0:.2f}%")
# Output: Prediction error = 99.14%

For additional uses, see code documentation in SolarDBData and the helper script solardb/run/solardb_main.py.

License

The code in this repository is licensed under MIT. For information about the SolarDB licensing, see the project website.