Technology topic mining and evolution path analysis based on dynamic topic modeling: a case study of hydrogen energy vehicle technology

Technology topic evolution analysis and convergence trend prediction constitute important research directions in technological intelligence. Addressing the identity drift and threshold dependency issues inherent in conventional topic models for cross-period topic matching, as well as the limited capacity of existing technology evolution studies to quantitatively predict technology convergence trends, this paper proposes an integrated framework for technology evolution analysis and convergence prediction that combines the Dynamic Topic Model (DTM), technology lifecycle theory, and Graph Convolutional Networks (GCN). The framework first employs a Logistic growth model to delineate technology lifecycle stages, subsequently applies DTM to dynamically model time-series patent texts for continuous topic tracking and multi-dimensional evolution identification, and further constructs a topic co-occurrence network to perform link prediction and evaluation of potential technology convergence relationships using GCN. Based on patent data from the Derwent Innovation Index covering hydrogen energy vehicles from 2001 to 2024 (11,872 valid patents) as a case study, a dynamic evolution map of 50 technology topics was constructed and technology convergence trends were predicted. The findings reveal that: (1) hydrogen energy vehicle technology exhibits a three-stage evolutionary pattern characterized by “single-point breakthrough—system expansion—intelligent convergence,” with core technology topics progressively shifting from fuel cell stack R&D toward membrane electrode integration, from high-pressure hydrogen storage toward diversified storage and transportation, and from power control toward intelligent energy management; (2) topic persistence analysis indicates that technology differentiation and reorganization intensified significantly during 2017–2024, with newly emerging topics predominantly oriented toward vehicle-level integration and thermal management, reflecting a shift in innovation focus from the “component level” to the “system integration level”; (3) the GCN-based technology convergence prediction model outperforms all four baseline models across four of the five evaluation metrics (AUC = 0.8925), and the prediction results reveal two differentiated convergence pathways—“emerging–emerging” and “mature–emerging”—providing forward-looking references for technology deployment. The proposed integrated “evolution analysis–convergence prediction” framework offers a systematic and quantitative methodology for technology evolution research, supporting technology planning and strategic decision-making in the hydrogen energy industry.