Research Topic Evolution Mapping in Smart Grid AI Applications Using BERT

The rapid development of smart grid systems has been accompanied by an exponential increase in artificial intelligence (AI) applications, ranging from demand forecasting to fault detection and renewable energy optimization. Tracing the evolution of topics in the interdisciplinary domain under consideration is critical in ascertaining novel avenues needed for guiding proactive inquiry and supporting strategic planning in the academia and the industry. Although the literature in the domain is burgeoning, the existing methodologies for topic evolution analysis focus on traditional keyword-centric or statistical approaches, which fail to capture the semantic context, domain-specific intricacies, and the fluid mobility across disparate research domains. These issues pose formidable challenges to mapping the evolution of research accurately and delineating new research avenues. This paper proposes the Research Topic Evolution Mapping Enhanced by BERT (RTEM-BERT) as the first to integrate BERT-based contextualized linguistic frameworks for analyzing and visualizing the evolution of topics in smart grid artificial intelligence. RTEM-BERT applies BERT embeddings in analyzing massive scholarly corpora to enable semantic finegrained clustering and longitudinal research theme tracking. Experimental investigations carried out on research articles published during the period 2010-2024 exceeding 15,000 units demonstrate that the RTEM-BERT model surpasses baseline models, including LDA and Word2Vec, on coherence, spatiotemporal consistency, and interpretability. The model successfully traces the evolution of topics from the early periods such as load forecasting, to the contemporary focus of edge intelligence, cyber security, and integration of renewable energy. RTEM-BERT enhances methodologies that quantifies the transformation of artificial intelligence smart grid research and supports knowledge discovery and research planning activities.

Перевод пока недоступен