Motivation
Combined-cycle power plants (CCPP) are among the most efficient fossil-fuel power generation technologies, and their integration with hydrogen production via Liquid Organic Hydrogen Carriers (LOHC) represents a pathway to lower-carbon operation. Optimizing such a system is non-trivial:
- Many interacting design variables (turbine parameters, LOHC flow rates, heat integration points)
- Each Aspen Plus simulation takes minutes — too slow for iterative optimization
- Multiple competing objectives (power output, hydrogen production, efficiency) require multi-objective treatment
The solution: replace expensive Aspen simulations with fast ML surrogate models, enabling thousands of optimization evaluations per minute.
Approach
- Process Modeling — Built the CCPP + LOHC system in Aspen Plus, validated against literature data
- Automation — Python automation framework (COM interface) for large-scale DOE-based dataset generation
- Surrogate Modeling — Trained four ML models (RF, GB, NN, GP) on the simulation dataset
- Optimization — Applied Bayesian (single-objective) and NSGA-II (multi-objective) on the surrogate
Phase 1 — Grid Sweep (Complete)
4-line heat recovery comparison across the design grid (SR × MCH) per line:
| Line | Location | Temp | Success Rate | Penalty | E1 Impact |
|---|---|---|---|---|---|
| Line 6 | Post-SH gas | *** °C | *** | ~*** kW | None |
| Line 5 | Post-GT gas | *** °C | *** | ~*** kW | None |
| Line 9 | Post-SH steam | *** °C | *** | ~*** kW | None |
| Line 4 | Post-combustion | *** °C | *** | ~*** kW | Linear decrease |
Quantitative values withheld pending publication. Penalty rankings and qualitative findings are summarized below.
Key findings (qualitative)
- Zero-penalty heat recovery zone discovered at Line 6: LOHC integration causes no measurable reduction in CCPP power output
- 4-line penalty ranking: Line 6 < Line 5 < Line 9 < Line 4 (lowest to highest penalty)
- LOHC competes with steam turbine efficiency, not full CCPP efficiency — penalty scope is narrower than initially expected
- LHHW kinetic model validated at base case with physically meaningful heat-limited behavior
- Maximum H₂ output achieved at Line 6 (high MCH, near-maximum SR)
Phase 2 — LHS + Surrogate + NSGA-II (Pending)
Planned approach:
- LHS with 4 variables (SR, MCH, P, U), 500–1,000 samples per line
- Surrogate model: ANN / RF / GPR comparison
- NSGA-II: 4-line independent Pareto fronts
- Overlay comparison with base case (no LOHC integration)
Engineering Implications
- Line 6 (zero-penalty): hydrogen production from CCPP waste heat with no measurable power output cost
- Line 4: avoid — ~50,000 kW penalty makes integration economically unattractive
- Line 5: attractive — small penalty + higher source temperature extends MCH conversion range
- Line 9: moderate penalty + additional steam-side error boundary constraint
Limitations
- Phase 1 grid sweep used 2 variables (SR, MCH); full 4-variable LHS pending
- Steady-state Aspen Plus model; dynamic transients not captured
- Economic analysis (LOHC system cost, electricity price, H₂ market value) not yet included
- LHHW kinetic parameters from a single literature reference (Usman 2012)
→ Built with: Aspen Automation Framework