🌪️ Wind Forecasting Project

🚀 Project Overview

This project focuses on wind forecasting using various deep learning models. It includes data preprocessing, model implementation, and training scripts for different architectures such as STTRE and Spacetimeformer.

📜 Full Project Title

Ultra-Short-Term Probabilistic Spatio-Temporal Modeling of Wind Farm Dynamics and Disturbances for Wake Steering Control

This open-source framework aims to predict wind speed and direction in the short term, specifically tailored for wake steering control applications in wind farms.

📚 Table of Contents

• 🌪️ Wind Forecasting Project
  • 🚀 Project Overview
    • 📜 Full Project Title
  • 🛠 Setup
    • Jupyter Notebook Collaboration
    • Data
    • Environment Setup
  • 🧠 Models
  • 🔄 Preprocessing
  • 🖥️ Running Jobs
  • 📂 Project Structure
  • 📋 Usage

🛠 Setup

Jupyter Notebook Collaboration

To ensure consistent handling of Jupyter notebooks in this repository, please follow these steps after cloning:

1. Install nbstripout using Mamba:

   mamba install nbstripout

2. Set up nbstripout for this repository:

   nbstripout --install --attributes .gitattributes

Data

The examples folder contains scripts for downloading and processing input data:

examples/download_flasc_data.py

This script downloads the SMARTEOLE wake steering experiment data and extracts it to the inputs folder.

Environment Setup

The install_rc folder contains scripts and YAML files for setting up Python environments for both CUDA and ROCm:

• install.sh: Script to create and activate the environments
• wind_forecasting_cuda.yaml: Conda environment for CUDA-based setups
• wind_forecasting_rocm.yaml: Conda environment for ROCm-based setups
• export_environments.sh: Script to export environment configurations

To set up the environments, run:

install_rc/install.sh

🧠 Models

The wind_forecasting/models directory contains implementations of various forecasting models:

• STTRE
• Spacetimeformer
• Autoformer
• Informer2020

Each model has its own subdirectory with specific implementation details and training scripts.

🔄 Preprocessing

The wind_forecasting/preprocessing folder contains scripts for data preprocessing:

• preprocessing_main.ipynb: Main script for loading and preprocessing data
• load_data.sh: Script for loading the data in the HPC
• data_loader.py: Contains methods for loading data
• data_inspector.py: Methods for plotting and analysing data
• data_filter.py: Methods for filtering and arranging data

🖥️ Running Jobs

The rc_jobs folder contains SLURM scripts for running jobs on HPC environments:

• job.slurm: General job script for NVIDIA GPUs
• job_amd.slurm: Job script for AMD GPUs
• job_preprocess.slurm: Job script for preprocessing data

To submit a job from the HPC to the cluster, use:

sbatch rc_jobs/job.slurm

📂 Project Structure

Click to expand

wind-forecasting/
├── examples/
│   ├── download_flasc_data.py
│   ├── SCADA_SMARTEOLE_Overview.ipynb
│   └── inputs/
│       ├── awaken_data
│       └── SMARTEOLE-WFC-open-dataset
├── lut/
├── rc_jobs/
│   ├── estimate_job_start_time.sh
│   ├── job.slurm
│   ├── job_amd.slurm
│   └── job_preprocess.slurm
├── install_rc/
│   ├── export_environments.sh
│   ├── install.sh
│   ├── setup.py
│   ├── wind_forecasting_cuda.yaml
│   └── wind_forecasting_rocm.yaml
├── wind_forecasting/
│   ├── models/
│   │   ├── Autoformer/
│   │   ├── Informer2020/
│   │   └── spacetimeformer/
│   ├── preprocessing/
│   │   ├── preprocessing_main.ipynb
│   │   ├── data_loader.py
│   │   ├── data_inspector.py
│   │   ├── data_filter.py
│   │   └── load_data.sh
│   ├── postprocessing/
│   └── run_scripts/
│       ├── train_informer.py
│       └── train_spacetimeformer.py
├── .gitignore
├── .gitattributes
├── .gitmodules
├── STTRE.ipynb
├── STTRE.py
└── README.md

📋 Usage

1. Clone the repository and set up the Jupyter notebook collaboration as described in the setup section.
2. Download the required data using the script in examples or use your own data.
3. Set up the appropriate environment (CUDA or ROCm) using the scripts in the install_rc folder.
4. Preprocess the data using the script in the wind_forecasting/preprocessing folder.
5. Train and evaluate models using the scripts in the wind_forecasting/models directory.
6. For running jobs on HPC environments, use the SLURM scripts provided in the rc_jobs folder.

---