Exploring Time-Series Forecasting with Two Powerful Deep Learning Models โSpatial-Temporal Transformer (SL Transformer) and LSTM (Long Short-Term Memory)โto predict solar energy generation using historical and weather data. The SL Transformer is designed to handle both short-term (next 6 hours) and long-term (next few months) predictions, while LSTM is used as a baseline model for comparison.
This project builds a dual-scale solar energy forecasting system using cutting-edge deep learning models:
๐น A Spatial-Temporal Transformer (SL Transformer) that captures both local fluctuations and long-term seasonal patterns
๐ธ An LSTM model that serves as a traditional baseline for long-range forecasting
Whether it's predicting the next 6 hours or the next 6 months, this system translates solar trends into actionable insights โ ideal for smart grids, green energy systems, and future-ready infrastructure.
- Built for multi-horizon forecasting.
- Captures spatial and temporal dependencies using attention mechanisms.
- Predicts:
- โ Next 6 hours (real-time applications)
- โ Next few months (seasonal trend analysis)
- Handles sequential dependencies over long ranges.
- Used for long-term solar trend prediction.
- Serves as a baseline for benchmarking the Transformer model.
- ๐ Location:
/data
- โ
Features:
- Solar irradiance components
- Meteorological variables (temperature, humidity, wind speed)
Model | MSE (kWh) | MAE (kWh) |
---|---|---|
LSTM | 0.0015 | 0.0157 |
SL Transformer | 0.0067 | 0.0469 |
While LSTM demonstrates stronger performance on long-term forecasting with lower error metrics, the SL Transformer stands out for its ability to generalize across both short-term (next 6 hours) and long-term (next few months) predictions within a single architecture. It captures temporal dependencies and spatial dynamics, making it more scalable and adaptive for real-world solar energy systems where conditions change rapidly. Its ability to learn complex spatial-temporal patterns makes it highly adaptable for real-time energy management, grid optimization, and scalable deployment in smart energy systems.
- ๐ Python 3.12
- ๐ฆ TensorFlow / PyTorch
- ๐ Pandas, NumPy, Scikit-learn
- ๐ Matplotlib, Seaborn
- ๐ Google Colab (for training + inference)
Indu Sree.N
๐ง indusreen78@gmail.com
๐ GitHub
If you have ideas, suggestions, or just want to connectโfeel free to reach out!