Thermodynamic Liquid Manifold Networks (TLMN)

🌍 Physics-Bounded Deep Learning for Autonomous Off-Grid Microgrids 🌍

---

"Zero-magnitude nocturnal error across all 1,826 testing days."

Important

Implementation Note: This repository contains the core architecture and settings as described in the associated research paper. However, some code structures and experimental configurations have been slightly adjusted to facilitate educational study, modification, and independent testing. The codebase will be fully synchronized with the exact methodology presented in the manuscript upon the paper's final formal publication.

---

🎯 The Problem We Solved

Contemporary solar forecasting models fail in two catastrophic ways:

Failure Mode	Impact	TLMN Solution
☁️ Temporal Phase Lag	Delayed cloud response	Dilated 1D-CNN (zero lag)
🌙 Phantom Nocturnal Generation	Impossible predictions	Thermodynamic Alpha-Gate

🏆 Results (5-Year Horizon, Semi-Arid Climate)

Metric	Value
📉 RMSE	18.31 Wh/m²
📊 Pearson Correlation	0.988
🌙 Nocturnal Error	Zero (all 1,826 days)
⚡ Phase Response	< 30 min during rapid transients
🧮 Parameters	63,458 (ultra-lightweight)

---

🏗️ Architecture (v3)

┌─────────────────────────────────────────────────────┐
│                                                     │
│   ☀️  NASA POWER Input (22 Variables, 24h Window)   │
│       Meteorological + Geometric + Derived          │
│                        │                            │
│              ┌─────────▼──────────┐                 │
│              │  Hankel Embedding  │  Koopman         │
│              │  + Tanh Projection │  linearization   │
│              │  + Positional Enc. │                  │
│              └─────────┬──────────┘                 │
│                        │                            │
│       ┌────────────────▼────────────────┐           │
│       │  🔥 Dilated 1D-CNN Encoder      │           │
│       │  3 layers × dilation [1,2,4]    │  Zero-lag │
│       │  Receptive field = 13 steps     │  temporal │
│       │  (replaces LiquidNeuralODE)     │  encoding │
│       └────────────────┬────────────────┘           │
│                        │                            │
│       ┌────────────────▼────────────────┐           │
│       │  🎯 Symplectic Cross-Attention  │  Physics  │
│       │  Q = meteorological features    │  guided   │
│       │  K/V = ClearSky + SZA           │  (after   │
│       │  γ-weighted residual            │  encoder) │
│       └────────────────┬────────────────┘           │
│                        │                            │
│    ┌───────────────────▼───────────────────┐        │
│    │  🌡️ Thermodynamic Alpha-Gate          │        │
│    │  pred = σ(KAN(h)) × ClearSky_norm    │  Night  │
│    │  Structural zero guarantee           │  = 0    │
│    └───────────────────┬───────────────────┘        │
│                        │                            │
│             📊 GHI Prediction (Wh/m²)               │
│             Physically bounded, always              │
│                                                     │
└─────────────────────────────────────────────────────┘

🔬 v2 → v3 Evolution

Component	v2	v3 (Current)
Temporal Encoder	LiquidNeuralODE (slow)	Dilated 1D-CNN (fast, no lag)
Loss Function	MSE + penalty terms	Log-Cosh (peak-aggressive)
Physics Enforcement	Loss penalties	Structural gate (100% guarantee)
Sliding Window	Fixed	3-step stride
Input Features	15	22 variables

---

📂 Repository Structure

📦 TLMN-Thermodynamic-Solar-Microgrids/
│
├── 📁 training_code/
│   └── 🧠 TLMN_Model.py                 # Full TLMN v3 architecture
│
├── 📁 evaluation_code/
│   └── 📊 TLMN_Test.py                   # Test evaluation pipeline
│
├── 📁 training_data/
│   ├── 📊 Hourly_2010_2015.csv           # NASA POWER hourly data
│   └── 📊 Hourly_2020_2025.csv
│
├── 📄 TLMN_Paper.pdf                     # Published paper
├── 📄 TLMN_Paper.docx
├── 📋 requirements.txt
└── 📖 README.md

---

🚀 Quick Start

# Clone & setup
git clone https://github.com/Marco9249/TLMN-Thermodynamic-Solar-Microgrids.git
cd TLMN-Thermodynamic-Solar-Microgrids
pip install -r requirements.txt

# Train TLMN v3
python training_code/TLMN_Model.py

# Evaluate
python evaluation_code/TLMN_Test.py

---

📚 Related Research Papers

#	Paper	Repository	arXiv
1	Physics-Guided CNN-BiLSTM Solar Forecast
2	Physics-Informed State Space Model (PI-SSM)
3	TLMN (this repo) 🌟
4	Asymmetric-Loss Industrial RUL Prediction
🎮	Interactive 3D Architecture Visualization		—

---

📖 Citation

@misc{abdullah2026tlmn,
  title   = {Thermodynamic Liquid Manifold Networks: Physics-Bounded
             Deep Learning for Solar Forecasting in Autonomous
             Off-Grid Microgrids},
  author  = {Mohammed Ezzeldin Babiker Abdullah},
  year    = {2026},
  eprint  = {2604.11909},
  archivePrefix = {arXiv},
  primaryClass  = {cs.LG},
  url     = {https://arxiv.org/abs/2604.11909}
}

APA 7th Edition: Abdullah, M. E. B. (2026). Thermodynamic Liquid Manifold Networks: Physics-Bounded Deep Learning for Solar Forecasting in Autonomous Off-Grid Microgrids. arXiv. https://arxiv.org/abs/2604.11909

---

👤 Author

Mohammed Ezzeldin Babiker Abdullah

---

© 2026 Mohammed Ezzeldin Babiker Abdullah — All rights reserved.