nkwasey/Bui-Hydro-DL

This repository provides a modular, research-grade pipeline for short-term hydropower generation forecasting* at the Bui Dam (Ghana), using advanced deep learning models—including LSTM, Transformer, Seq2Seq, and their tunable variants. # Hydropower Generation Forecasting at Bui Dam --- ## Project Overview This repository provides a **modular, research-grade pipeline** for **short-term hydropower generation forecasting** at the Bui Dam (Ghana), using advanced deep learning models—including LSTM, Transformer, Seq2Seq, and their tunable variants. Key features: - **End-to-end reproducible workflow**: from raw data QC to explainability and uncertainty analysis. - **Modular codebase** with robust configuration, CLI, and comprehensive documentation. - **Feature ablation and model comparison**: supports scientific experimentation and publication-ready output. - **Research-grade diagnostics:** SHAP feature importance, MC Dropout uncertainty, skill tables, and publication-ready figures. --- ## Quickstart **1. Clone the repository:** ```bash git clone github.com cd Bui-Hydro-DL ```` **2. Install dependencies (recommended: use conda or \[venv]):** ```bash conda env create -f environment.yml conda activate bui-forecast # or pip install -r requirements.txt ``` **3. Prepare data:** * Place provided CSVs (`Exp_A_MetOnly.csv`, `Exp_B_MetHydro.csv`) in the `data/` directory. **4. Run your first analysis:** ```bash # Run a data QC notebook (see Notebooks section below) jupyter notebook notebooks/00_Data_Exploration_and_QC.ipynb # Or train a model python train.py --config configs/your_experiment.yaml ``` --- ## Project Structure ```text LSTMv0.3.0/ ├── .gitignore ├── README.md ├── environment.yml ├── requirements.txt ├── config.py # Global configuration and paths ├── models.py # All model architectures ├── data.py # Data loading, cleaning, feature engineering ├── evaluation.py # Metrics, SHAP, confidence intervals ├── inference.py # Model inference and uncertainty quantification ├── collinearity.py # Multicollinearity detection and pruning ├── cv.py # Time-series cross-validation u …