Identifying and analyzing technological ideas in hydrogen fuel cells using topic models and large language models

Hydrogen fuel cells are a crucial clean energy technology. This study provides a comprehensive identification and analysis of the technological ideas within the hydrogen fuel cell domain, systematically tracing their full life-cycle from origination and evolution to application. Here, technological ideas are defined as creative problems and solutions generated through human or human–machine collaboration during research and development processes. Using global patent data collected from 1978 to 2025, this study identifies 111 fine-grained technological ideas in the hydrogen fuel cell domain based on Latent Dirichlet Allocation and further synthesizes them into 19 coarse-grained clusters with the aid of GPT-5. The findings reveal that the original technological ideas in hydrogen fuel cells emerged through three stages, demonstrating an evolutionary path from foundational materials toward system integration and end-use applications. Based on a proposed multidimensional dynamic measurement framework for emerging technological ideas, hydrogen fuel cell technological idea clusters exhibit four S-curve patterns, and the following recent emerging directions are identified: membrane electrode assembly and polymer membranes, secondary battery materials, energy conversion and storage systems, vehicles, and wireless communication and information transmission. Furthermore, 12 strongly associated technological idea combinations are uncovered using the Apriori algorithm, and idea mining conducted with large language models suggests that future innovation will increasingly emphasize systematic, cross-scale integrated design paradigms. Integrating scientometric methods with large language models, this study expands the boundaries of idea mining, offering a micro perspective for research on technological idea development in hydrogen fuel cells and inspiring future innovation.