Challenges, trends, and the role of LSTM in AI-based predictive maintenance of electrolyzers for solar hydrogen systems: a review

Abstract

The transition toward low-carbon energy systems has increased interest in hydrogen as a clean energy carrier, with solar-driven water electrolysis emerging as a promising technology due to its high efficiency and compatibility with renewable energy sources. However, dynamic operating conditions and intermittent renewable input accelerate electrolyzer degradation, reducing reliability and system lifespan. Predictive maintenance (PdM), supported by artificial intelligence (AI), offers a data-driven approach to anticipate failures and improve operational durability. This review systematically investigates AI-based PdM approaches for electrolyzers, with an emphasis on long short-term memory (LSTM) networks and Internet of things (IoT) integration. Following PRISMA 2020 guidelines, 35 peer-reviewed studies (2018–2026) were selected and analyzed based on AI methodology, dataset characteristics, validation strategy, and deployment capability. Results show that LSTM models effectively capture temporal degradation behavior, while alternative approaches such as random forests, convolutional neural networks, and hybrid models also achieve competitive performance, depending on the data structure and application context. However, most of the existing studies remain offline, with limited real-time validation and partial IoT integration, hence leaving a significant gap between model development and industrial deployment. This review identifies key challenges in data availability, validation, and system integration, emphasizing the need for edge–cloud architectures and real-time PdM frameworks for practical implementation in hydrogen energy systems.