Multi-Period Optimization of Multi-Timescale Energy Systems: Application to Solid-Oxide Electrolysis Cells

Nishant Giridhar, Douglas Allan, Steve Zitney, Eric Liese, Lorenz T. Biegler, Debangsu Bhattacharyya2024 · DOI

TL;DR

This paper introduces an adaptive coupling framework for optimizing energy systems that operate across vastly different timescales, such as Solid-Oxide Electrolysis Cells (SOECs). By dynamically triggering interactions between fast-timescale operations (hours) and slow-timescale degradation (years) only when specific error thresholds are met, the researchers achieved a 50–90% reduction in computational time compared to traditional trial-and-error methods.

What Problem Does This Paper Address?

Many energy systems are "intrinsically multi-timescale." For an SOEC-based hydrogen production plant, engineers must manage fast dynamics (seconds to hours) for controls and market fluctuations while simultaneously accounting for slow dynamics (thousands of hours) like chemical degradation of the cell microstructure.

The gap in current research lies in tractability. Standard approaches often use a "quasi-steady state" assumption for the fast timescale to make the long-term optimization solvable. However, this oversimplification can lead to inaccuracies in predicting the cumulative impact of degradation on lifetime metrics like Net Present Value (NPV) or Levelized Cost of Hydrogen (LCOH). The challenge is to couple these timescales efficiently without the prohibitive computational cost of a brute-force simulation over a 30-year plant life.

Methods

The authors propose a Multi-Period Dynamic Optimization approach featuring Adaptive Coupling. Key technical components include:

  • Coupling Trigger Conditions: Instead of fixed intervals, the model uses mathematical triggers to decide when the fast and slow timescales should interact. These triggers are based on:
  • Slow-timescale dynamics: Triggered if the degradation state ($\theta$) changes beyond a tolerance ($\epsilon_\theta$).
  • Fast-timescale dynamics: Triggered if external parameters (like market prices or hydrogen demand) change significantly.
  • Prespecified events: Scheduled maintenance or equipment replacement.
  • Variable Period Lengths: The slow dynamics are reformulated using normalized time ($\xi \in [0,1]$). The actual duration of these periods ($d_{i,j}$) becomes an optimization variable. This allows the solver to automatically "stretch" or "shrink" time periods to satisfy error constraints.
  • Nested Multi-Period Model: The framework chains multiple major periods together to reflect seasonal market variations, supporting nested dynamics that are amenable to mathematical decomposition.

Datasets

The study utilizes a physics-based SOEC System Flowsheet Model coupled with Microstructure Degradation Models.

  • The degradation model accounts for factors like Ni-agglomeration in the fuel electrode and oxygen electrode degradation.
  • The system was evaluated using two primary cases:
  • Constant $H_2$ production rate: A baseline scenario.
  • Diurnal production profiles: A more complex scenario where production fluctuates daily, requiring nested fast-timescale dynamics.

Key Insights and Ideas

  • Efficiency through Sparsity: You don't need to update the slow-timescale model every hour if the system state hasn't changed significantly. By setting a tolerance ($\epsilon$), the model only performs expensive coupling calculations when necessary.
  • Optimization-Driven Discretization: By making period boundaries free to move during optimization, the model finds the mathematically optimal points to check for degradation, rather than relying on human intuition or fixed grids.
  • Broad Applicability: While demonstrated on SOECs, the authors emphasize that this framework is not specific to fuel cells. It is applicable to any system where fast operational decisions impact long-term structural health, such as chemical reactors or structural health monitoring in civil engineering.

Key Results

  • Computational Speedup: The adaptive coupling method showed a ~90% decrease in computational time compared to a trial-and-error grid refinement approach.
  • Reduced Complexity: To achieve a 1.5% error threshold, the adaptive method required only 4 coupling points, whereas a standard fixed-grid approach required 9 points and took nearly twice as long to solve.
  • Large-Scale Performance: For the complex diurnal production case involving approximately 150,000 variables, the model reached a solution in roughly 18 minutes (including initialization).
  • Trade-off Analysis: The results clearly show how different market conditions favor different strategies: low energy costs favor conservative operation to preserve stack life, while high energy costs favor aggressive operation despite faster degradation.

Limitations and Future Work

The authors identify several areas for further development:

  • Integration Methods: The current work uses Implicit Euler finite elements for discretization; future iterations could allow for a wider choice of integration methods in the fast timescale.
  • Period Decomposition: To handle even more rigorous fast-dynamics models, the authors suggest implementing period decomposition to further speed up solution times.
  • External Inducement: While the framework handles hourly/seasonal changes well, more work could be done on how external market volatility (LMP) specifically induces multi-timescale behavior in different geographies.

Summary updated

Abstract

A presentation detailing the recent progress in the optimization of Solid-Oxide Cells under Chemical Degradation. The existing methods typically involve a quasi-steady state assumption to make this large-scale problem tractable. Here, we introduce an extension to this method that allows for adaptive coupling of degradation with the fast-timescale process based on a variety of error thresholds. This reduces the error accumulated during long-term optimization of the SOC under degradation and allows for decision making at multiple timescales. The results presented include long-term operating profiles of the SOC under degradation with both steady-state and fluctuating day to day operation.
BibTeX
@misc{giridhar2024multi,
  author = {Nishant Giridhar and Douglas Allan and Steve Zitney and Eric Liese and Lorenz T. Biegler and Debangsu Bhattacharyya},
  title = {{Multi-Period Optimization of Multi-Timescale Energy Systems: Application to Solid-Oxide Electrolysis Cells}},
  year = {2024},
  doi = {10.2172/2478222},
}
Figure