🏡⚡SynaptiGrid.AI - Home Energy Management System (HEMS)

A sophisticated controller powered by Spiking Neural Networks (SNN) and Reinforcement Learning (RL) to optimize home energy usage, comfort, and environmental sustainability.

---

🧠 Overview

This intelligent HEMS system:

✅ Optimizes HVAC operations ✅ Manages battery storage intelligently ✅ Minimizes energy costs while maintaining comfort ✅ Reduces carbon emissions ✅ Adapts to weather & electricity prices ✅ Provides an interactive dashboard for real-time control

---

🏗️ System Architecture

🔧 Core Components

| Component | Description | | ------------------- | -------------------------------------------------------------- | | rl_environment.py | Simulates home energy with thermal, grid, and battery dynamics | | rl_agent.py | TD3 agent that learns optimal control policies | | snn_model.py | Spiking Neural Network for energy signal processing | | data_processor.py | Handles preprocessing, normalization, and feature encoding | | hems_dashboard.py | Streamlit dashboard for visualization and control |

🧬 Neural Architecture

🎯 Reinforcement Learning (TD3)

• Actor: 256-256 neurons
• Critic: 256-256 neurons
• Feature extractor with layer normalization

⚡ Spiking Neural Network (SNN)

• Input: 6 neurons
• Hidden: 128 LIF neurons
• Output: 3 neurons
• Learning: STDP (Spike-Timing-Dependent Plasticity)

---

🚀 Features

🧠 Intelligence & Optimization

• 🎯 Multi-objective Optimization: Cost, comfort, carbon footprint
• 🔄 Adaptive Control: Dynamic strategy adjustment
• 📈 Prediction: Anticipates energy needs & price trends
• 🧩 Context-Aware Decisions: Uses weather, time, user settings
• ✅ Constraint Enforcement: Ensures valid actions via optimization layer

🖥️ User Interface Highlights

• 📊 Interactive Dashboard: Live control and insights
• 🌞 Energy Flow Analysis: Source & consumption breakdown
• 🌡️ Temperature Monitoring: Indoor vs. outdoor
• 🔋 Battery Visualization: Charge & discharge cycles
• 🧪 Controller Analysis: RL + SNN behavior tracking
• 🔬 Neural Model Viewer: SNN spikes and learning visualization

---

📈 Models & Performance

| 🧪 Model Type | 📂 Path | 🎯 Reward | 📝 Description | | ------------------ | -------------------------------------------- | --------- | ------------------------- | | 🔁 RL + SNN Hybrid | logs/run_20250531_135517/td3_hems_best.zip | 10.7 | Latest hybrid integration | | 🏆 Best RL Model | logs/run_20250531_131322/td3_hems_best.zip | 12.80 | Use of only RL |

---

⚙️ Installation

# 📥 Clone the repository
git clone https://github.com/yourusername/hems-research.git
cd hems-research

# 🧪 Create and activate virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# 📦 Install dependencies
pip install -r requirements.txt

---

▶️ Usage

🖥️ Run Dashboard

./run_dashboard.sh
# or
streamlit run hems_dashboard.py

🎓 Train a Model

./train_model.sh
# or
python train_optimized.py --timesteps 50000 --episode_length 48 --snn_epochs 15

🧪 Test a Model

python test_model.py --model logs/run_20250531_135517/td3_hems_best.zip

---

🧰 Key Improvements

🏠 Environment

• 🔄 Observation normalization
• 🧠 Smarter reward shaping
• ⚡ Battery model refinement
• 🌡️ Thermal inertia modeling
• 🌍 Realistic environmental variation

🤖 RL Agent

• 🧱 Custom feature extractor
• 📉 Learning rate scheduling
• 🧪 Gradient clipping
• 🧠 Ornstein-Uhlenbeck noise for exploration
• 📊 Curricular learning with progressive complexity

🔌 SNN Integration

• 🧬 SNN pre-processing of time-series signals
• 🔁 Efficient spike encoding
• 🔧 Tuned hyperparameters for energy domains

---

🏆 Results Summary

| ✅ Feature | 🔍 Outcome | | ---------------- | -------------------------------------- | | Battery Usage | Context-sensitive charging/discharging | | HVAC Control | Temperature kept within ±0.5°C | | Grid Interaction | Reduced peak grid consumption | | Solar Use | Maximized self-consumption | | Cost Savings | 15–25% lower than rule-based control |

---

📂 Project Structure

.
├── hems/                      # Main package
│   ├── __init__.py            # Package initialization
│   ├── rl_environment.py      # HEMS environment with dynamics modeling
│   ├── rl_agent.py            # TD3 agent implementation
│   ├── snn_model.py           # Spiking Neural Network model
│   ├── data_processor.py      # Data preprocessing utilities
│   └── utils/                 # Utility functions
├── tests/                     # Test directory
│   ├── __init__.py            # Test package initialization
│   ├── test_rl_env.py         # Tests for RL environment
│   ├── test_rl_agent.py       # Tests for RL agent
│   ├── test_snn_model.py      # Tests for SNN model
│   └── test_data_processor.py # Tests for data processor
├── scripts/                   # Utility scripts
│   ├── __init__.py            # Scripts package initialization
│   ├── train_model.py         # Training script
│   ├── train.sh               # Training shell script
│   └── run.sh                 # Dashboard runner script
├── hems_dashboard.py          # Streamlit dashboard
├── hems_data_final.csv        # Dataset for training and evaluation
├── setup.py                   # Package installation
├── setup.cfg                  # Package configuration
├── pyproject.toml             # Build system configuration
├── logs/                      # Training logs and saved models
└── results/                   # Evaluation results and visualizations

Installation as a Package

You can install the HEMS package for development:

# Clone the repository
git clone https://github.com/Rajarshi012003/SynaptiGridAI.git
cd SynaptiGridAI

# Install in development mode
pip install -e .

Then import the package in your code:

# Import components
from hems import HEMSEnvironment, SNN_Model
from hems import train_rl_agent, evaluate_rl_agent

# Use the components
env = HEMSEnvironment(...)
model = train_rl_agent(env, ...)

---

🧭 Future Work

• 🏠 Real-world smart home deployment
• 🌐 Federated learning for multi-home systems
• 🧑‍💼 Preference learning over time
• 🔌 Integration with smart appliances
• 🧠 Explainable AI for control decisions

---

📜 License

Licensed under the MIT License. See the LICENSE file for details.

---

🙏 Acknowledgments

1. This system is built on the HEMS PLAN REPORT, which proposes a unified mathematical framework for energy optimization using SNN and RL techniques.
2. Open Data Providers: We gratefully acknowledge the creators and maintainers of open datasets such as NREL ResStock, IDEAL, and others, whose data enabled robust modeling and validation.
3. Open Source Community: Appreciation to the developers of snnTorch, PyTorch, stable-baselines3, and other open-source libraries that formed the backbone of this project.