From Digital Twin to AI-Integrated Control: A Review and Research Agenda for Large-Scale PEM Electrolyzer Plant Management

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Provides a structured research agenda and identifies the latency gap in digital twin control, offering a foundation for future experimental validation.

🏢実務担当者:Highlights the need for AI-integrated control systems in large-scale electrolyzer plants to improve operational efficiency and reduce downtime.

🏛政策担当者:Offers insights into technology readiness levels and the potential of AI to support national hydrogen production targets.

The pressing need for advanced control strategies that surpass the limitations of traditional distributed control systems is essential for enhancing green hydrogen production using electrolyser technologies to gigawatt capacities. Digital twin technology has surfaced as an innovative framework for enhancing predictive maintenance and optimising operations within hydrogen production processes, ranging from traditional grey hydrogen production methods, such as steam methane reforming, to advanced green ammonia facilities and next-generation proton exchange membrane electrolysers. Recent studies clearly show that the accuracy of degradation predictions ranges from 85% to 95%, and real-time monitoring frameworks have been confirmed across both pilot and industrial scales. However, a significant and often overlooked limitation exists in all these contexts: the inherent architectural constraint of passive DT systems that prevents them from autonomously closing the control loop. The gap between prediction and action introduces delays of 30 to 120 minutes at the critical moment when the dynamic renewable energy load requires responses in sub-seconds. This review presents three original contributions: it examines published experimental evidence from SMR, green ammonia, and PEM electrolyser DT deployments, all based on a structured literature search conducted in Scopus and Web of Science. It systematically defines the latency gap between prediction and action and introduces a three-tier hierarchical structure that includes traditional DCS, digital twin, and an AI decision layer to address this issue. Next, it outlines five specifically designed AI algorithm modules and introduces several challenges related to research agendas, along with evaluations of technology readiness levels. All quantitative projections for AI-integrated DCS are derived from published studies in similar fields and should be regarded as hypotheses that need experimental validation rather than definitive engineering specifications. The 500-MW PEM plant configuration utilised as a reference throughout is conceptually scaled and aligns with near-term national hydrogen strategy targets; however, it does not pertain to any description of an existing facility.

• Research Square https://doi.org/10.20944/preprints202607.2105.v1first seen 2026-08-06 04:34:36