Solar Forecast

A deep learning–based project built to improve short-term solar power prediction. Trained on over 26,000 points of real-world weather and generation data, the final system combines Transformer and TCN architectures to model both short- and long-term patterns in solar output. From data cleaning and feature engineering to model tuning and ensemble development, resulted in an R² score of 0.9488. This study helps demonstrate how thoughtful architecture and experimentation can lead to more reliable renewable energy forecasting.