Hydrogen Network Optimization Across Reformers
Challenges
- A steam methane reformer produces hydrogen by reacting methane from natural gas with steam at high temperatures.
- Frequent changes in hydrogen demand result in suboptimal manual adjustments to operating parameters, and unplanned shutdowns have to be handled at short notice by operators.
- At network level, load is distributed without accounting for cost factors or for each unit's optimal load.
Approach
At unit level, a set of ML models predicts fuel gas consumption, steam net export and constraint limits from operating parameters, and a genetic algorithm searches the feasible operating space for the lowest-cost parameter set that respects every predicted constraint.
At network level, an optimization algorithm sets production levels across the hydrogen producers to meet demand, taking network architecture and operational and capacity constraints into account.
Key features
- ML prediction of fuel gas consumption, steam net export and constraint limits
- Genetic algorithm search for the lowest-cost feasible parameter set
- All predicted constraints respected inside the search
- Network-level production levels set against demand
- Network architecture and capacity constraints accounted for
Impact
5–10%
Uplift in individual unit efficiency
87% → 100%
Increase in utilization of the cost-effective units
Network load distributed by cost, not by habit
As stated in the company deck