license: mit
tags:
- reinforcement-learning
- multi-agent
- time-series
- diffusion-model
- energy-management
- smart-grid
☀️ SolarSys: Scalable Hierarchical Coordination for Distributed Solar Energy
[Source: The SolarSys paper (e.g., your PDF) is the primary source for all claims below.]
[cite_start]SolarSys is a novel Hierarchical Multi-Agent Reinforcement Learning (HRL) system designed to manage energy storage and peer-to-peer (P2P) trading across large communities of solar-equipped residences[cite: 10]. This repository contains the full source code for the SolarSys system, including the trained policies, the custom Gym environment, and the hierarchical diffusion model used for data augmentation.
🚀 Key Features and Performance
[cite_start]SolarSys addresses the scalability limitations of traditional Multi-Agent RL (MARL) methods (like MAPPO and MADDPG) in large Virtual Power Plants (VPPs)[cite: 9, 145].
| Metric | SolarSys Performance (1000 Agents) | Key Mechanism |
|---|---|---|
| Grid Import Reduction | [cite_start]$27.48 \pm 0.42%$ [cite: 18] | [cite_start]Two-tier control scheme [cite: 12, 69] |
| Daytime Solar Utilization | [cite_start]$82.76 \pm 5.11%$ [cite: 18] | [cite_start]Intra-cluster MAPPO optimization [cite: 13] |
| Fairness (Jain's Index) | [cite_start]0.773 [cite: 18] | [cite_start]Fairness term in reward function [cite: 391, 511] |
| Scalability | [cite_start]Stable convergence at 1000+ agents [cite: 504] | [cite_start]Mean-Field Coordination at the Inter-Cluster layer [cite: 14] |
🧠 System Architecture
The core of SolarSys is a two-level decision hierarchy:
- [cite_start]Low-Level (Intra-Cluster): Individual households use a MAPPO agent to make instantaneous decisions (charge, discharge, local P2P trade, grid trade) based on local meter readings and price signals[cite: 13, 313].
- [cite_start]High-Level (Inter-Cluster): Cluster Managers use a Mean-Field policy to coordinate bulk energy transfers between clusters, ensuring the overall system remains balanced against grid constraints[cite: 14, 314].
📊 Data Generation Framework
[cite_start]To enable large-scale simulation with realistic temporal dynamics, SolarSys includes a Hierarchical Diffusion Model for generating synthetic, long-duration energy profiles that maintain both long-term (seasonal/monthly) and short-term (daily/hourly) characteristics[cite: 254, 255].
- [cite_start]Model: Hierarchical Diffusion U-Net [cite: 254, 255]
- [cite_start]Input: Household ID and Day-of-Year conditioning [cite: 256]
- Output: High-resolution time series for Grid Usage and Solar Generation (kWh).
📁 Repository Structure
The project is organized into core modules and data folders.
SolarSys/
├── data/
│ ├── per_house/ # Raw CSVs for diffusion model training
│ ├── training/ # Cleaned RL training datasets
│ └── testing/ # Cleaned RL evaluation datasets
├── models/
│ ├── diffusion_models/ # Trained Hierarchical Diffusion Model checkpoints
│ ├── mappo_models/ # Trained MAPPO baselines and low-level agents
│ └── inter_agent_models/ # Trained MeanField high-level coordinator
├── Environment/
│ ├── __init__.py
│ └── solar_sys_environment.py # Custom Gym environment for flat RL
├── cluster/
│ ├── __init__.py
│ └── inter_cluster_coordinator.py # Logic for high-level trade matching
└── trainers/
├── __init__.py
├── hierarchical_train.py # Main SolarSys HRL training script
└── evaluation_scripts/ # Scripts for baselines (PG, MADDPG, MAPPO, MFAC)