How ruohan-li/solarcube works
Main Components
a. Data Layer:
• SolarCube Dataset (containing satellite imagery and in-situ data)
• Data Loading/Generation Components
• Data Generator
• DataLoader
• Time Features Processing
b. Model Layer (Multiple Architectures):
• ConvLSTM Model
• Cuboid Transformer
• Informer Transformer
• LSTM Model
c. Utils Layer:
• Metrics and Evaluation
• Configuration Management
• Checkpoint Handling
• Tools and Utilities
[Data Layer] → [Model Layer] → [Evaluation Layer]
[Satellite Images] →\
• [Data Generator] → [DataLoader] → [Time Features]
[In-situ Data] →/
• /→ [ConvLSTM Branch]
• /→ [Cuboid Transformer Branch]
[Input] → → [Informer Transformer Branch]
• → [LSTM Branch]
Specific Implementation Details
For each model architecture, show:
• Encoder-Decoder structure (where applicable)
• Attention mechanisms (especially in transformer-based models)
• Configuration files connection
• Training flow
Key Relationships to Highlight
• Data preprocessing pipeline
• Model training workflows
• Evaluation metrics flow
• Configuration management system
Technical Details
• File formats and data types
• Model architecture specifications
• Training parameters
• Evaluation metrics
The configuration management system
Remember to maintain a balance between detail and clarity, focusing on the architectural decisions that make this system unique, particularly the multiple model approaches for solar energy forecasting.