Combination of Site-Wide and Real-Time Optimization for the Control of Systems of Electrolyzers (2404.06748v2)

Abstract: The rapid expansion of renewable energy sources has introduced significant volatility and unpredictability in the energy supply chain, necessitating advanced control strategies to ensure grid stability and reliability. Green hydrogen production via electrolysis offers a viable solution for converting and storing this volatile renewable energy. However, the inherent fluctuations of renewable energy sources present challenges for consistent utilization and integration of green hydrogen. This work proposes a two-stage optimization approach, combining site-wide optimization and real-time optimization for managing systems of electrolyzers. By adapting an existing static optimization model, dual use is achieved in both site-wide optimization and real-time optimization. The hierarchical optimization structure, characterized by distinct temporal resolutions, enables effective responses to both dynamic changes and long-term trends. The side-wide optimization layer generates long-term plans based on forecast data, while the real-time optimization layer refines these plans in real-time, accommodating immediate fluctuations and ensuring efficient operation. The results from the case study on a system of electrolyzers demonstrate the method's effectiveness in aligning electrolyzer operation with actual availability of renewable energy. This approach offers a robust framework for optimizing the operation of electrolyzers but also other types of flexible energy resources, contributing to sustainable and economically viable energy management.

• The paper introduces a two-stage optimization method combining site-wide and real-time optimization to manage forecast uncertainties in renewable energy integration.
• It adapts an existing static model for dual use, ensuring consistent control across long-term planning and rapid real-time adjustments.
• A case study shows that the approach improves hydrogen production efficiency and grid stability with manageable computational times.

This paper (2404.06748) introduces a practical two-stage optimization method for controlling systems of flexible energy resources, specifically focusing on electrolyzers integrated with renewable energy sources. The core problem addressed is the challenge of managing the inherent volatility and unpredictability introduced by renewable energy (RE) in green hydrogen production. Traditional static optimization methods often struggle with real-time fluctuations and forecast uncertainties.

The proposed solution is a hierarchical optimization approach combining Site-Wide Optimization (SWO) and Real-Time Optimization (RTO). This structure leverages distinct temporal resolutions to handle both long-term trends and immediate changes:

1. Site-Wide Optimization (SWO): Operates with a longer temporal resolution (e.g., hours to days). It uses long-term forecast data (e.g., day-ahead RE generation, market prices) to generate a comprehensive plan for the entire optimization horizon ($\mathcal{T}$). This plan defines targets and initial conditions for subsequent RTO steps.
2. Real-Time Optimization (RTO): Operates at a finer temporal resolution (e.g., minutes). Its optimization horizon ($T$) is typically equivalent to one SWO time step ($\Delta\tau$). RTO refines the SWO plan in near real-time, using updated short-term forecasts and measured system states to react to immediate fluctuations and deviations from the initial plan.

A key innovation of this work is demonstrating how an existing static optimization model can be adapted for dual use in both SWO and RTO. This avoids potential inconsistencies that could arise from using different models for different control layers. The necessary adaptations for the static model primarily involve:

• Fixing historical variables: In the RTO stage, variables representing past system states (e.g., resource states, setpoints) are fixed to their realized values or outcomes from the previous optimization steps.
• Setting RTO targets based on SWO results: A crucial constraint is introduced (Equation 1 in the paper) ensuring that the total energy procured or consumed over the RTO horizon matches the target defined by the SWO for that specific time step. This maintains alignment with the long-term plan while allowing RTO flexibility within that target.
• Potential objective modification: While the SWO might minimize cost (e.g., electricity procurement), the RTO objective might be adjusted or the constraints used to prioritize other factors like maximizing RE integration or hydrogen output within the bounds set by the SWO.

The paper provides an algorithm (flowchart in Figure 1) for the continuous execution of this two-stage process:

1. Initialize the system state.
2. Import long-term forecasts.
3. Solve the SWO model for the entire horizon $\mathcal{T}$.
4. Store the SWO results (the plan).
5. Use the SWO results for the current time step ($\tau$) as initial values for the RTO.
6. Import high-resolution short-term forecasts for the RTO horizon ($T$).
7. Solve the RTO model for the horizon $T$.
8. If the end of the RTO horizon is not reached: transmit optimized setpoints to the real system/simulation.
9. Receive measurement values from the system.
10. Incorporate measured values and proceed to the next RTO time step within the same SWO interval $\tau$, using current setpoints as new starting values.
11. If the end of the RTO horizon ($T$) is reached: check if the end of the SWO horizon ($\mathcal{T}$) is reached.
12. If not at the end of $\mathcal{T}$, obtain new starting values for the next SWO interval ($\tau+1$) from the SWO results and return to step 5.
13. The process is typically continuous, restarting for a new SWO horizon as needed.

The method was evaluated using a case paper involving a system of electrolyzers connected to both the electricity grid (using European intra-day market prices) and a wind farm (using real wind data with added uncertainty). The optimization models were built upon a validated, modular structure [WRF23b] parameterized using a systematic methodology [WaFa24]. The implementation used Java, IBM ILOG CPLEX for solving the optimization problems, and OPC-UA for communication with a simulation model [MOCK20241885].

Results demonstrated the method's effectiveness:

• SWO provided a solid initial plan based on available forecasts.
• RTO successfully adapted to short-term fluctuations in RE availability that deviated from the initial SWO forecasts.
• The RTO adjustments led to increased hydrogen production by capitalizing on higher-than-forecasted RE output, showing improved efficiency compared to strictly following the initial static plan.
• Computational times (around 10 seconds per SWO interval including all its RTO steps) were deemed manageable for practical application.
• The dual use of the model allowed RTO to react effectively within the framework set by SWO, preventing major deviations or the need for unplanned adjustments.

The practical implications are significant for the widespread adoption of RE and green hydrogen. This two-stage approach provides a robust and adaptable framework for optimizing flexible resources under uncertainty, contributing to grid stability and minimizing energy costs by allowing better integration of volatile RE. The ability to adapt existing static models reduces implementation barriers. While forecast accuracy remains a limitation, the RTO mechanism effectively mitigates the impact of short-term forecast errors. The method is also presented as generalizable to other types of flexible energy resources.