# Workpackages

**WP1: Literature Review & System Design:** [https://www.sciencedirect.com/science/article/pii/S0360319924031628](https://www.sciencedirect.com/science/article/pii/S0360319924031628)[](https://www.sciencedirect.com/science/article/pii/S0360319924031628)
*   Define system architecture
*   Define Research questions
*   Extract and Analyse Data
*   Report findings
*   Research gap: Limited AI research in a real-world hydrogen system, and more case studies are required for realistic implementations [https://www.sciencedirect.com/science/article/pii/S1364032125007944](https://www.sciencedirect.com/science/article/pii/S1364032125007944)[](https://www.sciencedirect.com/science/article/pii/S1364032125007944)
**WP2: Forecasting Models**
*   Use **pvlib** to generate PV features (irradiance, temperature, angles).
*   Train **LSTM or Transformer** models in **PyTorch Lightning** for both PV and hospital demand forecasting.
*   Evaluate against **scikit-learn baselines**. [https://www.youtube.com/watch?v=SIEaLBXr0rk](https://www.youtube.com/watch?v=SIEaLBXr0rk)[](https://www.youtube.com/watch?v=SIEaLBXr0rk)
**WP3: Simulation Environment**
*   Implement a simulation framework (e.g., Python/Simulink/GridLAB-D).
*   Model physical components (efficiency curves, storage losses, demand profiles).
**WP4: RL Control Development**
*   Build **hospital microgrid environment** in **Gymnasium or SimPy**.
*   Train RL agents (PPO or SAC) using **Stable-Baselines3**.
*   Integrate forecasting models from the ML stage to provide predictive states.
*   Benchmark RL vs. rule-based control (as in H2B2–PowiDian).
**WP5: Multi-Objective Evaluation**
*   Evaluate system performance across multiple KPIs: cost, emissions, renewable fraction, hydrogen efficiency, crop-supporting reliability.
*   Conduct sensitivity analysis on different grid pricing and weather conditions.
**WP6: Innovation Extensions**
*   Integrate a basic crop growth/CO₂ enrichment module to test closed-loop optimization.
*   Simulate greenhouse-to-grid interactions (selling surplus electricity/hydrogen).
*   Implement explainability metrics for RL decisions.
*   Conduct scalability tests for multiple greenhouses connected as a distributed resource.
**WP7: Results & Reporting**
*   Summarize findings in terms of **innovation compared to existing hydrogen optimizers**.