SolarCycle Analysis & Prediction

Sunspot Cycle Insights + LSTM Time-Series Forecasting

This project performs an end-to-end analysis of solar sunspot cycles and builds an LSTM-based forecasting model to predict future Solar Sunspot Numbers (SSN).
It combines scientific visualization, time-series pattern discovery, and deep learning forecasting to better understand solar activity and its impact on space weather.

---

⭐ Why This Project Matters

Solar cycles are approximately 11-year cycles where the Sun’s magnetic activity fluctuates, influencing:

• Satellite operations
• Communication systems
• Navigation signals
• Power grid stability (space weather impact)

This repository demonstrates both interpretation (analysis) and prediction (modeling) using real historical SSN trends.

---

🎯 Objectives

• Visualize long-term sunspot activity patterns
• Compare SSN with correlated solar indicators
• Study cyclic behavior, lag patterns, and seasonality
• Forecast SSN values using an LSTM neural network
• Evaluate forecasting quality using MAE and visualization

---

📥 Dataset

This project uses historical Sunspot Number data from the Kaggle dataset:

🔗 Sunspots Dataset:
https://www.kaggle.com/datasets/robervalt/sunspots

📊 About the Dataset

The dataset contains:

• Monthly sunspot numbers from 1700s to present
• Used for studying solar magnetic activity cycles
• Ideal for time-series analysis and forecasting

---

📊 Sunspot Cycle Analysis (Exploratory Insights)

1) Historical Sunspot Cycle (1749–2021)

• Cyclical pattern repeats roughly every 11 years
• A full cycle can range from 9 to 14 years
• Each cycle transitions through:
  • Solar Minimum → Solar Maximum → Solar Minimum

---

2) Comparing Radio Flux with Sunspot Number

• Sunspot Number: direct indicator of solar magnetic activity
• 10.7 cm Radio Flux: indirect measure via radio energy emission

Both rise and fall together, showing a strong correlation.

---

3) Comparing Polar Field with Sunspot Number

This plot highlights an inverse relationship and supports the concept of magnetic polarity reversal approximately every 11 years.

---

4) Seasonality and Lag Analysis

Lag plots show the series is not random, meaning it contains strong predictive structure.
Key historical behavior includes:

• Dip around 1790–1820 → Dalton Minimum
• Major peak around 1960, followed by fluctuations

---

5) Comparing Total Solar Irradiance (TSI) with Sunspot Numbers

A positive correlation is observed:

• Higher sunspot activity → higher solar irradiance

---

6) Asymmetry and Long-Term Seasonality

The heatmap suggests:

1. Short-term ~11-year solar cycle
2. Long-term ~100-year brightness cycle

---

🔮 Solar Sunspot Number Forecasting with LSTM

This module uses a Long Short-Term Memory (LSTM) neural network to forecast SSN values from historical time-series data.

Key Steps

• MinMax Scaling for stable neural training
• Lag Feature Engineering using 16 time steps
• 3-layer LSTM architecture with 32 units per layer
• Callbacks
  • Learning rate reduction on plateau
  • Early stopping to reduce overfitting
• Evaluation using Mean Absolute Error (MAE)

---

🧠 Model Summary

• Model Type: LSTM (Time-Series Forecasting)
• Lag Window: 16 steps
• Loss Function: Mean Squared Error (MSE)
• Optimizer: Adam
• Output: Next-step SSN prediction

---

📈 Results

• Baseline MAE: (update with your value)
• LSTM MAE: 13.29 (inverse transformed)

---

🗂️ Repository Contents

SolarCycle-analysis_and_prediction/
│
├── solorcycle_analysis.ipynb          # Full solar cycle analysis notebook
├── Lstm_sunspot_predict_model.ipynb   # LSTM forecasting notebook
├── image1.png                         # Historical SSN cycle plot
├── image2.png                         # SSN vs Radio Flux
├── image3.png                         # Polar Field vs SSN
├── image4.png                         # Lag & Seasonality plots
├── image5.png                         # TSI vs SSN
├── image6.png                         # Asymmetry heatmap
├── prediction_plot.png                # LSTM prediction vs actual
└── README.md

---

⚙️ Requirements

• Python 3.x
• Libraries:
  • NumPy
  • Pandas
  • Matplotlib
  • scikit-learn
  • statsmodels
  • TensorFlow / Keras

Install dependencies:

pip install numpy pandas matplotlib scikit-learn statsmodels tensorflow

---

▶️ How to Run

Option 1: Run in Jupyter / Colab

1. Open solorcycle_analysis.ipynb and run all cells
2. Open Lstm_sunspot_predict_model.ipynb and run all cells

Option 2: Local Execution

jupyter notebook

---

🔍 Conclusion

This project shows that sunspot data contains strong cyclic patterns, meaningful relationships with solar indicators, and clear forecasting potential.
By combining analysis + LSTM prediction, it provides a strong foundation for extending solar forecasting using more advanced deep learning and hybrid time-series methods.

---

🚀 Future Improvements

• Experiment with different lag lengths
• Try alternative architectures (GRU, ConvLSTM)
• Add exogenous features (radio flux, polar field, TSI)
• Perform hyperparameter tuning (Optuna / KerasTuner)
• Multi-step forecasting (predict multiple months ahead)

---

👨‍💻 Author

Nithish Gowda H N - BTech(Hons.) CSE, AI & ML Major

Prajna - BTech(Hons.) CSE, Cloud & Full Stack Major