Integrated Design and Scheduling of Hydrogen Processes under Uncertainty: A Quantile Neural Network Approach.

Area of Science:

• Chemical Engineering
• Optimization
• Machine Learning

Background:

• Electrolysis-based hydrogen production plant design and scheduling face uncertainty from electricity price predictions.
• Traditional two-stage stochastic programming methods for this problem are computationally intensive.

Purpose of the Study:

• To develop a computationally efficient surrogate model for the second stage of stochastic programming in hydrogen production.
• To enable risk-aware optimization using quantile neural networks for integrated design and scheduling.

Main Methods:

• Utilized quantile neural networks as surrogate models for the second-stage value function in a two-stage stochastic program.
• Embedded the neural network surrogate into the optimization framework to avoid sampling-based approximations.
• Incorporated conditional value at risk (CVaR) alongside expectation for joint optimization.

Main Results:

• The surrogate-based approach significantly reduces computational requirements compared to sample average approximation.
• Optimization incorporating risk measures (CVaR) led to increased investment in electrolyzers and storage.
• The method provides high-quality decisions for integrated hydrogen plant design and scheduling.

Conclusions:

• Quantile neural network surrogates offer an efficient and effective solution for optimizing hydrogen production under price uncertainty.
• Integrating risk measures enhances plant robustness against volatile electricity costs.
• This surrogate-based method advances the field of process systems engineering for renewable energy applications.