Windfarm Prediction

Description

This repository contains Python code for processing and analyzing wind farm data using machine learning techniques. The provided scripts and notebooks facilitate data preprocessing, feature selection, and LSTM-based time series prediction.

Download Raw Data

British Wind Farm Data

Go to https://zenodo.org/record/5841834#.ZEajKXbP2BQ

Download Kelmarsh_SCADA_2016_3082.zip. (Any other year is fine too)

Unzip the data and copy the corresponding turbine data into the folder /src/raw_data

Brazilian Wind Farm Data

Go to https://zenodo.org/record/1475197#.ZD6iMxXP2WC

Download UEPS_v1.nc

Copy the file into the folder /src/raw_data

How to run

First, install dependencies

# clone project   
git clone https://github.com/Felix6464/windfarm_forecasting_boette.git

# install project   
cd windfarm_forecasting 
pip install -e .   
pip install -r requirements.txt

Next, navigate to any file and run it.

windfarm_britain_dataprocessing.ipynb

• Reads raw data from the British wind farm dataset into a pandas dataframe.
• Imports the datapreprocessing.py file with data processing functions.
• Splits the data into target and input datasets.
• Performs min-max normalization and removes NaN values.
• Allows feature selection using various mechanisms.
• Saves a new CSV file with the data of the selected important features to ./preprocessed_data

windfarm_brazil_dataprocessing.ipynb

• Processes wind farm data from an nc file.
• Similar data processing steps as in windfarmbritaindataprocessing.ipynb.

lstm_training.py

• Trains an LSTM model for time series prediction.
• Reads the preprocessed CSV file and performs data normalization.
• Splits the data into train and test sets with specified windows.
• Shapes the data into time series splits for model training.
• Sets hyperparameters and initializes the model.
• Uses teacher-forcing during training to learn temporal dependencies.
• Saves model-related information for later evaluation and plotting.
• in order to change the model that is being used you have to change the respective import of the model
• and specify the class of model initalization in the main function (line 122)

LSTM_enc_dec.py

• Contains the neural network architecture and training loop for LSTM models.
• Includes a function for preparing input data compatible with time series prediction.

utility_functions.py

• Contains two minor functions:
  • Loading text as JSON.
  • Appending two pandas dataframes.

plot_saved_model.py

• Loads a saved model and preprocessed data from files.
• Plots the time series of the first variable (target variable) for training and test data.
• Creates two plots showing the convergence of training and test loss over the specified range of epochs.
• Imports plots.py, which contains individual functions for the plots.

lstm_testing.py

• Loads a saved model and a preprocessed dataset (not the dataset the model was trained on).
• Evaluates the trained model on an unseen dataset.

global_model_approach.ipynb

• Appends two preprocessed datasets to create a larger dataset.
• Increases variability in the data and potentially improves performance on unseen datasets.