A robust MCDM framework with LLM for offshore wind power-seawater hydrogen production-marine ranch integrated system investment decision

Abstract

The Offshore Wind Power-Seawater Hydrogen Production-Marine Ranch Integrated System (I-OWHR) addresses the dual challenges of offshore wind power integration and intensive marine space utilization through vertical layered development, demonstrating considerable development potential. However, the project is still at an early stage of development, with limited historical data and practical experience. Moreover, the involvement of multiple interconnected subsystems makes investment decision-making for I-OWHR highly complex. To address these challenges, this study integrates a knowledge-driven Large Language Model (LLM) with a Multi-Criteria Decision-Making (MCDM) framework and proposes a two-stage intelligent robust decision-making framework, termed Large-Language-Model-driven Literature-Frequency-Driven Weighting (LLM-LFDW)-Weighted Perturbation-based Stochastic TOPSIS (WP-STOPSIS). It used the LLM to identify evaluation criteria and further incorporated the risk of fluctuations in criteria importance into the MCDM through WP-STOPSIS. The results indicated that hydrogen blending and co-transportation via existing natural gas pipelines, along with the reutilization of decommissioned offshore platforms, consistently emerged as the preferred investment and construction options for current I-OWHR projects. Comparative experimental analysis confirmed the rationality of the criteria identification and weighting method based on LLM-LFDW, with a correlation coefficient of 0.7 relative to mainstream subjective weighting approaches, indicating strong consensus representativeness. Meanwhile, sensitivity comparison analysis demonstrated that WP-STOPSIS exhibits high robustness under weight perturbations, achieving a robustness index of 0.9903. Compared with conventional MCDM methods, the sensitivity is reduced by 73.18%. Furthermore, the proposed two-stage intelligent robust MCDM approach can be independently applied to support decision-making in other complex energy projects at an early stage of development.