Wind farm maintenance planning: an intelligent solution based on LLM optimization

The efficient operation of wind farms presents numerous challenges, with maintenance task planning representing a particularly complex aspect. Consequently, the development of efficient and adaptable maintenance strategies is regarded as essential. This study introduces an intelligent maintenance planning method that utilizes Large Language Models (LLMs). The proposed approach integrates heterogeneous data sources, including information on maintenance teams and fault locations, through the use of structured prompts to fully exploit the problem-solving capabilities of LLMs. Experimental evaluations reveal that, in comparison to direct outputs from LLMs, the method reduces maintenance costs by 8.6%, with an average deviation rate of 8.2% from the optimal cost. A comprehensive framework is developed to improve the performance of LLMs in optimization tasks. This framework comprises structured prompting, memory of historical strategies, adaptive search refinement, and a specialized mathematical module. It ensures solution robustness, eliminates the risk of omissions or redundant repairs, and maintains consistency in output format. Additional validation using smaller LLMs yields an average deviation rate of 12.4%, confirming the practicality of the method in environments with limited computational resources. The findings offer new perspectives on industrial maintenance scheduling and highlight the potential of LLMs for addressing complex planning problems.