A Large-Scale Language Model Based System for Automated Generation of Offshore Wind Power Feasibility Study Reports

Driven by global energy transition goals, the large-scale development of offshore wind power imposes rigid requirements for professionalism, standardization, and timeliness on feasibility study reports (FSR). Traditional manual compilation and existing automated methods fail to meet these requirements due to interdisciplinary complexity, poor process controllability, and insufficient domain adaptation. To address these challenges, this paper proposes a configurable and interpretable offshore wind FSR generation system built on a three-tier framework that encompasses “data support, process orchestration, and quality assurance”. The system integrates a YAML-based workflow architecture, multi-level prompt engineering, and a comprehensive evaluation system. Notably, the introduced “Cyclic Aggregation Mode” enables the iterative generation and logical summarization of multi-subproject data, effectively distinguishing this system from traditional linear text generation models. Experimental results demonstrate that the proposed “Retrieval-Augmented Generation (RAG) + Large-scale Language Model (LLM) + Workflow” system outperforms baseline models with key metrics including semantic consistency (0.6592), information coverage (0.3908), structural compliance (0.5123), and an overall score (0.5965). Ablation studies validate the independent contributions of the RAG and Workflow components, thereby establishing the “RAG + LLM + Workflow” paradigm for intelligent professional document generation. This work addresses core challenges related to controllability, accuracy, and interpretability in high-stakes decision-making scenarios while providing a reusable technical pathway for the automated feasibility demonstration of offshore wind power projects.