Solar Panel Tilt Optimization using Machine Learning & Deep Learning

Project Overview

This project focuses on optimizing solar panel performance by predicting POA Global Irradiance, Surface Tilt, and Surface Azimuth to maximize energy generation. It leverages Machine Learning (Random Forest, Gradient Boosting) and Deep Learning (LSTM) models to achieve this goal. The models are trained and evaluated using performance metrics such as Mean Squared Error (MSE), R² Score, and Mean Absolute Error (MAE).

The notebook follows a structured approach:

• Data Preprocessing: Cleaning and preparing datasets.
• Feature Engineering: Selecting key features for predicting POA Global Irradiance, Surface Tilt, and Surface Azimuth.
• Model Training: Implementing and training ML/DL models.
• Model Evaluation: Assessing performance with metrics.
• Results & Analysis: Comparing model outcomes.
• Conclusion: Summarizing findings and suggesting improvements.

Key Findings

• Random Forest outperformed other traditional ML models with the lowest error and highest R² scores for predicting POA Global Irradiance, Surface Tilt, and Surface Azimuth.
• Gradient Boosting showed high accuracy but was slightly less effective than Random Forest across these predictions.
• LSTM achieved the lowest test loss, excelling in sequential learning, though Surface Tilt predictions require further refinement.

Dependencies

This project requires the following Python libraries:

• pandas
• numpy
• seaborn
• matplotlib
• tensorflow (with GPU support recommended)
• scikit-learn

Note: The datasets (2017_2019.csv, solar_angles_dataset.csv, solar_forecasting.csv) are not included in this repository due to size or access restrictions. Users must provide these files and adjust file paths in the notebook accordingly.

Model Details

Machine Learning Models

• Random Forest: Best-performing model for accuracy and stability in predicting POA Global Irradiance, Surface Tilt, and Surface Azimuth.
• Gradient Boosting: High accuracy, slightly less effective than Random Forest for these predictions.

Deep Learning Model

• LSTM:
  • Architecture: 64 LSTM units (ReLU) → 32 Dense units (ReLU) → 3 output neurons (for POA Global Irradiance, Surface Tilt, Surface Azimuth).
  • Optimization: Adam optimizer and MSE loss.
  • Challenges: Captures temporal dependencies but requires adjustment for negative Surface Tilt predictions.

Results

• Random Forest: Most reliable for predicting POA Global Irradiance, Surface Tilt, and Surface Azimuth.

  • Test Metrics:
    • POA Global Irradiance MSE: 17.6038
    • POA Global Irradiance R²: 0.9997
    • Surface Tilt MSE: 0.0444
    • Surface Tilt R²: 0.9999
    • Surface Azimuth MSE: 0.0144
    • Surface Azimuth R²: 0.99997

• Gradient Boosting:

  • High accuracy but slightly outperformed by Random Forest in predicting POA Global Irradiance, Surface Tilt, and Surface Azimuth.
  • Test Metrics:
    • POA Global Irradiance MSE: 88.6979
    • POA Global Irradiance R²: 0.9986
    • Surface Tilt MSE: 0.4066
    • Surface Tilt R²: 0.9991
    • Surface Azimuth MSE: 0.2972
    • Surface Azimuth R²: 0.9993

• LSTM: Promising for sequential predictions of POA Global Irradiance, Surface Tilt, and Surface Azimuth, with:

  • Test Loss: 0.4816
  • Test MSE: 0.4816
  • Test MAE: 0.5092

Future Enhancements

• Develop hybrid models (e.g., Random Forest + LSTM) for improved generalization in predicting POA Global Irradiance, Surface Tilt, and Surface Azimuth.
• Fine-tune LSTM to eliminate negative Surface Tilt predictions.
• Incorporate additional data (e.g., weather conditions) for enhanced accuracy.

Conclusion

This project demonstrates the effectiveness of ML and DL in predicting POA Global Irradiance, Surface Tilt, and Surface Azimuth to optimize solar panel orientation. Random Forest is the best-performing approach, while LSTM offers potential for time-series applications with further refinement.

License

This project is licensed under the MIT License. See LICENSE for details (if applicable).