PV Power Forecasting using Ensemble Learning

This project forecasts photovoltaic (PV) power generation using a combination of statistical, deep learning, and transformer-based models. Predictions from multiple models are combined using a simple mean ensemble to improve accuracy.

📌 Features

• Multiple Models Implemented:
  • ARIMA & Seasonal ARIMA (SARIMA)
  • Vector AutoRegression (VAR)
  • Long Short-Term Memory (LSTM) networks
  • Convolutional Neural Networks (CNN) for time series
  • Multi-Layer Perceptron (MLP)
  • Transformer models for sequence forecasting
• Ensemble Learning — combines outputs from multiple models for robust predictions.
• Data Visualization — publication-quality plots using scienceplots.
• Model Saving — trained models saved in .h5 format.

📂 Dataset

The project uses PV power generation data from PVGen.csv:

• Columns: Time-stamped readings of PV generation.
• Sampling: Subset of the first 190,000 readings is used for training/testing.
• Preprocessing: Date extraction, train/test splitting, normalization.

🛠 Installation

Install dependencies:

pip install pandas numpy matplotlib seaborn scikit-learn tensorflow pmdarima scienceplots statsmodels

If running in Google Colab, upload the dataset and install missing packages:

!pip install scienceplots pmdarima

🚀 Usage

1. Open PV_Forecating_Using_Ensemble_Learning.ipynb.
2. Run all cells in sequence:
   • Load and preprocess data
   • Train individual models
   • Generate forecasts
   • Evaluate and visualize results
3. Saved models will appear as .h5 files.

📊 Evaluation

Models are compared using Mean Squared Error (MSE) on the test set.
Example comparison:

• Simple Mean Ensemble
• CNN
• MLP
• ARIMA
• LSTM
• VAR
• Transformer

📈 Example Output

• Line plots: Actual vs. predicted PV generation.
• Bar charts: MSE scores for each model.
• Saved Plots: .svg format for high-quality publication-ready figures.

📜 License

This repository is licensed under the MIT License — see the LICENSE file for details.