🛰️🗺️Enhanced Solar Energy Intelligence Suite🌤️

This repository bundles multiple deep-learning and classical ML pipelines for satellite-aware solar analytics. It focuses on turning real-time weather feeds, NASA/NOAA irradiance data, and Google Earth Engine (GEE) imagery into accurate solar irradiance and energy predictions, plus cloud detection tooling to improve production estimates.

Highlights

• Full solar workflow: irradiance prediction, energy output forecasting, cloud detection/forecasting, and Google Earth Engine integrations.
• Hybrid data: live APIs (OpenWeatherMap, NREL, NASA POWER), synthetic data generators, and Sentinel/Landsat imagery through GEE.
• Production-oriented: configurable CLI prompts, realistic defaults for South/Southeast Asia, automatic fallbacks when APIs are unavailable.
• Explainability and reporting: charts, confidence distributions, daily breakdowns, and contextual recommendations per location.

Repository Structure

File	Purpose
solar_energy_output_prediction.py	Flagship LSTM + attention model with interactive CLI for end-to-end forecasting.
solar_irradiance_prediction.py	Gradient boosted tree models for irradiance estimation.
irradiance_prediction.py	Weather-driven baselines for irradiance forecasting.
cloud_detection.py	U-Net segmentation pipeline for satellite cloud masks.
cloud_forecasting.py	ConvLSTM-based short-term cloud motion forecaster.
gee_config.py / gee_config_simple.py	Centralized Google Earth Engine initialization utilities.
test_*.py	Smoke tests and integration checks for the main modules.
requirements.txt	Locked dependency list for local development.

Requirements

• Python 3.10+ (tested on Windows 10)
• Optional CUDA-enabled GPU for faster training
• GEE-enabled Google Cloud project (default: instant-text-459407-v4)
• API keys if you want real weather/irradiance data:
  • OpenWeatherMap (One Call 3.0)
  • NREL NSRDB
  • NASA POWER (no key required, but rate limiting applies)

Environment Setup

python -m venv .venv
.venv\Scripts\activate  # On PowerShell
pip install --upgrade pip
pip install -r requirements.txt

Google Earth Engine Authentication

# Service account (recommended for automation)
set GOOGLE_APPLICATION_CREDENTIALS=path\to\service-account.json

# User account (interactive)
earthengine authenticate

Running the Solar Energy Output Predictor

python solar_energy_output_prediction.py

Interactive prompts guide you through:

1. Custom location – decimal-degree coordinates and optional labels, with climate zone detection.
2. Time configuration – training window, forecast horizon (1–168 h), and specific-date target.
3. System parameters – capacity (kW), panel area (m²), training epochs, learning rate.
4. API keys – supply OpenWeatherMap/NREL keys or rely on realistic synthetic generation.

The script then:

• Retrieves or synthesizes weather + irradiance histories.
• Engineers irradiance, weather, temporal, and panel metadata features.
• Trains an attention-enhanced LSTM on sliding time-window datasets.
• Generates forward irradiance projections and energy forecasts.
• Prints human-readable summaries (daily totals, confidence histograms, weather context, revenue estimates).

Key Internals

• SolarPanelSpecs, EnergyPrediction, SystemPerformance, and TrainingDataPoint dataclasses keep metadata consistent.
• FeatureEngineer normalizes datasets and derives panel-adjusted features (temperature coefficients, soiling factors, tilt/azimuth effects).
• LSTMPredictor uses attention pooling to emphasize the most informative time steps.
• Real-data integration happens through RealDataAPIs, abstracting OpenWeatherMap, NREL NSRDB, and NASA POWER calls with synthetic fallbacks.

Other Notable Pipelines

• solar_irradiance_prediction.py: Offers regression models (XGBoost/Random Forest) trained on API + satellite-derived irradiance features.
• cloud_detection.py: U-Net segmentation for satellite imagery; supports training with GEE-sourced chips plus local augmentations.
• cloud_forecasting.py: ConvLSTM sequences for short-term cloud movement, enabling irradiance nowcasting.
• cloud_irradiance utilities (in cloud_* / cloud_segmentation.py if present) feed cloud-cover percentages back into energy predictors for dynamic derates.

Configuration & Customization

• gee_config.py: set your Google Cloud project ID and dataset defaults.
• setup_gee.py: helper for bootstrapping Earth Engine in fresh environments.
• requirements.txt: includes PyTorch, scikit-learn, pandas, structlog, geemap, Google API clients, etc.
• Dark-theme UI preference is already respected in interactive prompts for main/login/register screens (per project conventions).

Testing

Basic test entry points:

python -m pytest test_all_models.py
python test_cloud_detection.py
python test_gee_integration.py

Tests validate data loaders, model initialization, and GEE configuration fallbacks. Expand with scenario-specific cases before deployment (e.g., long-horizon inference, missing sensor data).

Cloud Detection Visualization Preview

The U-Net pipeline exposes composite views that combine raw Sentinel imagery, mask predictions, and overlay diagnostics. Below is an example output generated by cloud_detection.py:

Troubleshooting

• Authentication errors: run earthengine authenticate --clear, then re-auth.
• API quota exceedance: reduce polling frequency or schedule jobs off-peak.
• Slow training: lower sequence_length, disable attention, or leverage GPU acceleration.
• Unstable forecasts: ensure input datasets include required columns (irradiance, temperature, humidity, wind_speed). The scripts auto-fill defaults and log warnings when they patch missing features.

Roadmap Ideas

• Export-ready JSON/CSV summaries from CLI runs.
• RESTful microservice wrapper for SolarEnergyPredictor.
• Integration with weather forecast providers (ECMWF/GFS) for longer horizons.
• Automated model retraining with new GEE imagery batches.

License & Support

Distributed under the MIT License. For GEE usage questions, see the official docs and community forum. For model-specific issues, open an issue or consult the individual scripts for implementation notes.

---

Feel free to adapt the CLI prompts, model architectures, or data feeds to fit your geography, plant design, or deployment environment.

---

Feel free to adapt CLI prompts, model architectures, or data feeds to fit your geography, plant design, or deployment environment.
🌞 # GIS-ML-Intelligence