Towards automated optimization algorithm design with LLM: An exploratory study in multi-objective weather routing

Weather routing plays a crucial role in enhancing navigation safety and economic efficiency. As a key component in weather routing systems, optimization algorithms have attracted considerable research attention. Currently, the design of such algorithms relies heavily on manual effort, with the primary goal of developing widely applicable and effective approaches. However, since factors such as sailing waters, seasonal conditions and ship status vary across voyages, optimization algorithms must be tailored to specific conditions of each voyage. Designing a dedicated algorithm for individual voyages is time-consuming, which is unsuitable for supporting autonomous route and speed planning. To address this limitation, this study introduces a novel framework for the automated design of optimization algorithms, with a focus on multi-objective weather routing problems. Specially, leveraging a large language model (LLM), the framework adaptively configures genetic algorithms (GAs) according to voyage-specific conditions. It automatically generates effective crossover and mutation operators without manual intervention, thereby producing high-quality Pareto solution sets. The proposed framework is evaluated using real-world case experiments. Results demonstrate that, compared to traditional manually designed GAs, the proposed framework can autonomously generate high-performance optimization algorithms tailored to specific voyages. The algorithms generated by the proposed framework achieve more diverse and higher-quality Pareto fronts (PFs) with improved computational efficiency. Executable codes are available at https://github.com/Ldiper/LLM_AAD.