Multi-objective weather route automatic optimization method based on large language model
By using a multi-objective automatic meteorological flight route optimization method based on a large language model, crossover and mutation operators are generated. Combined with performance metrics and selection mechanisms, the problem of existing meteorological flight route optimization algorithms relying on human experience and having poor adaptability is solved, achieving efficient and autonomous flight route optimization. The quality and efficiency of the generated Pareto optimal solution set are significantly improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DALIAN MARITIME UNIVERSITY
- Filing Date
- 2026-04-02
- Publication Date
- 2026-07-10
AI Technical Summary
Existing meteorological flight route optimization algorithms rely on human experience, are prone to getting stuck in local optima, have poor adaptability, and cannot meet the real-time requirements of autonomous flight route and speed planning. Furthermore, existing evaluation indicators fail to comprehensively consider computational efficiency, resulting in insufficient practicality and flexibility.
A multi-objective automatic optimization method for meteorological flight routes based on a large language model is adopted. By generating crossover and mutation operators, combined with performance metrics and selection mechanisms, a Pareto optimal solution set is output, and a convergence time efficiency index is introduced to achieve autonomous optimization.
The generated algorithm significantly outperforms traditional methods in terms of the diversity and quality of Pareto optimal solution sets. It is highly adaptable to different scenarios, has high optimization efficiency, and can autonomously generate optimization algorithms suitable for different sea conditions, meeting real-time requirements.
Smart Images
Figure CN122369293A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ship meteorological route optimization technology, and in particular to a multi-objective automatic optimization method for meteorological routes based on a large language model. Background Technology
[0002] Maritime transport occupies a central position in global trade, accounting for approximately 80% of global trade volume. This dominant position makes meteorological route optimization in the shipping industry crucial for ensuring navigational safety and improving economic efficiency. Meteorological route optimization is a navigation optimization process based on ship characteristics and environmental conditions. Its core objective is to generate an optimal navigation plan that includes the route and corresponding speed to achieve the expected navigation goals. With changes in shipping market conditions, meteorological route optimization has evolved from a single-objective optimization task to a complex task encompassing multiple objectives. A typical requirement is to simultaneously optimize navigation time and fuel consumption. Such multi-objective optimization often produces a Pareto front. Multi-objective meteorological route optimization mainly consists of three interrelated and synergistic important parts: environmental forecasting, ship performance modeling, and optimization algorithms. Environmental conditions are a key factor in assessing a ship's sailing performance (such as speed loss and fuel consumption) and are also related to the ship's safety level (such as maximum wind resistance level). Typically, predictive data on environmental factors such as waves, wind, and ocean currents are required. The environmental impact is quantified through a ship performance model, and sailing costs are calculated by combining the ship's own parameters (such as static parameters, load status, and engine status). The optimization algorithm generates an optimal plan set that meets preset conditions by jointly optimizing the ship's speed and route.
[0003] In the early stages of technological development, precise algorithms such as dynamic programming were often used to solve optimization problems. These algorithms could guarantee mathematically optimal solutions. However, as modern maritime optimization problems have become increasingly complex, exhibiting characteristics such as conflicting objectives and high-dimensional decision spaces, the research paradigm has gradually shifted from precise methods to heuristic methods. Evolutionary algorithms, among heuristic algorithms, have gained widespread attention due to their high computational efficiency. These algorithms, by simulating natural evolution or swarm intelligence behavior, can efficiently select high-quality and feasible navigation plans in complex environments. Among them, genetic algorithms have become a research hotspot due to their simplicity and practicality. Their core process includes five key steps: population initialization, population update, selection, crossover, and mutation.
[0004] However, existing meteorological route optimization technologies still suffer from several prominent problems: First, the design of crossover and mutation operators, the core components of genetic algorithms, heavily relies on human experience. Limited by the designer's knowledge and thinking, it struggles to break free from established patterns, leading to algorithms prone to getting trapped in local optima and failing to achieve globally optimal planning. Second, significant differences exist in navigation waters, seasonal conditions, and ship status across different voyages. Manually designed general operators have extremely poor adaptability, requiring customized designs for specific voyages. However, manual customization is not only time-consuming and labor-intensive but also lacks a clear design direction, making it difficult to meet the real-time requirements of autonomous route and speed planning. Finally, existing evaluation metrics only focus on the convergence and distribution of Pareto solutions, without comprehensively considering computational efficiency, thus failing to fully measure the algorithm's practical application value. In recent years, machine learning technology has driven the development of automated algorithm design. However, while existing learning-assisted algorithms attempt to optimize operators, they require extensive manual parameter tuning and sample training, resulting in extremely high computational costs and difficulty adapting to dynamically changing navigation scenarios. These intertwined problems severely limit the practicality and flexibility of existing meteorological route optimization algorithms, making them unable to efficiently address the complex and ever-changing demands of maritime navigation. Therefore, there is an urgent need for an automatic algorithm design method that requires no human intervention, is highly adaptive, and balances optimization effectiveness and efficiency. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a multi-objective automatic optimization method for meteorological routes based on a large language model. This invention collects ship static parameters, navigation parameters, and marine environmental prediction data to clarify the dual objective requirements and constraints of minimizing navigation time and fuel consumption. Then, it guides the large language model to autonomously generate operators through exploration and correction strategies. Performance metrics and selection mechanisms are set to evaluate and optimize the operators, obtaining the optimal operator combination. Combining population initialization, updating, and roulette wheel selection mechanisms, a Pareto optimal solution set is output. Simultaneously, a convergence time efficiency index is introduced to comprehensively measure the algorithm's convergence accuracy and computational efficiency.
[0006] The technical means employed in this invention are as follows:
[0007] A multi-objective automatic optimization method for meteorological flight routes based on a large language model includes: acquiring ship parameters and navigation-related parameters, setting optimization objectives and constraints; generating crossover and mutation operators based on the large language model, exploration strategies, and correction strategies; evaluating the crossover and mutation operators using performance metrics, and optimizing the operators using a selection mechanism to obtain the optimal operator combination; using the optimal operator combination in conjunction with population initialization and population update, employing roulette wheel selection to select high-quality individuals, and outputting the Pareto optimal solution set for the flight route; introducing a convergence time efficiency index to evaluate the quality of the Pareto optimal solution set, thereby achieving automatic optimization of the flight route.
[0008] Furthermore, the ship parameters include ship tonnage, braking power, fuel consumption rate, service speed, and speed adjustment range. The navigation-related parameters include: latitude and longitude of the navigation start and end points, navigation season, and corresponding marine environment prediction data. The marine environment prediction data includes: significant wave height, wave period, wind speed, wind direction, ocean current speed, and ocean current direction. The optimization objective is set to minimize navigation time and fuel consumption. The constraints include an unobstructed navigation area and maintaining the speed within the controllable range of the main engine.
[0009] Furthermore, the crossover and mutation operators include route crossover operators, speed crossover operators, route mutation operators, and speed mutation operators; the exploration strategies include differentiation exploration and similarity exploration; the correction strategies include structure optimization and parameter tuning; the structure optimization is used to improve the operator architecture and enhance search capabilities; and the parameter tuning is used to optimize key operator parameters and improve performance.
[0010] Furthermore, the performance metric adopts the scheme execution deviation index, which is defined as the reciprocal of the average distance between the algorithm-generated scheme and the initial scheme in two dimensions: flight time and fuel consumption. The performance metric is expressed as:
[0011] in, This indicates the number of points included in the Pareto front. Indicates the generated first The difference in fuel dimension between each front point and the corresponding point of the initial reference front. Indicates the generated first The difference in time between each front point and the corresponding point of the initial reference front, a constant. To smooth the term and prevent the denominator from approaching zero when the distance between corresponding points at the leading edge is zero.
[0012] The selection mechanism employs a ranking-based random selection strategy, ranking operators according to the values of the performance metrics, and assigning selection probabilities to the selected operators. Represented as:
[0013] in, The total number of operators, For the first Ranking of the algorithms.
[0014] Furthermore, the population initialization involves randomly generating feasible navigation scheme individuals based on the great circle route and constraints. Each individual contains a sequence of route waypoints and corresponding speed parameters. The population update divides all candidate solutions into multiple non-dominated levels based on the Pareto dominance relation, calculates the crowding distance of each individual, and prioritizes retaining high-level, sparsely distributed individuals to maintain population diversity. The roulette wheel selection method is used to screen high-quality individuals, and the optimal operator combination is applied to generate new individuals. This process is repeated iteratively until a Pareto optimal solution set is output.
[0015] Furthermore, the convergence time efficiency index is defined as the ratio of the generation distance of the operator to the total computation time, used to comprehensively measure the convergence accuracy and computational efficiency of the algorithm. The convergence time efficiency index is expressed as:
[0016] in, This represents the convergence time efficiency index, used to comprehensively measure the convergence accuracy and computational efficiency of an algorithm. The generation distance is used to quantify the average distance between the solution set and the theoretical optimal frontier, reflecting the convergence accuracy. This represents the total computation time to complete one full optimization.
[0017] The convergence time efficiency index, combined with the hypervolume index and the generation distance index, evaluates the quality of the Pareto optimal solution set. The hypervolume index is used to quantify the target space volume dominated by the solution set, and the generation distance index is used to measure the average distance between the solution set and the theoretical optimal frontier.
[0018] Compared with the prior art, the present invention has the following advantages: The multi-objective automatic optimization method for meteorological flight routes based on a large language model provided by this invention, through exploration and correction of a dual-hint strategy, combined with the powerful generation capability of a large language model, can autonomously generate customized and efficient genetic algorithm crossover and mutation operators according to the specific conditions of each flight, without the need for manual intervention throughout the process. At the same time, it adopts an alternating optimization strategy, first fixing the mutation operator to optimize the crossover operator, then fixing the crossover operator to optimize the mutation operator, and finally obtaining a high-performance combination of crossover and mutation operators through iterative optimization, effectively solving the problems of traditional genetic algorithm operators relying on experience and having poor adaptability due to manual design.
[0019] The automated algorithm design framework (LLM-AAD) proposed in this invention has been verified through real-world case experiments. The generated algorithms significantly outperform traditional genetic algorithms (including non-dominated sorting genetic algorithm-II, intensive Pareto evolution algorithm, knee-driven evolution algorithm, etc.) in terms of the diversity and quality of Pareto optimal solution sets. Compared with these existing methods, the algorithms generated by this framework improve the core evaluation indicators of hypervolume (HV) and generational distance (GD) by at least 6.24% and 7.41%, respectively. Moreover, it can autonomously generate specific optimization algorithms applicable to different sea state data, demonstrating stronger scenario adaptability.
[0020] The convergence time efficiency (CTE) index proposed in this invention comprehensively considers both the optimization speed and the progress of Pareto front optimization. The comparison results based on this index show that the algorithm generated by the framework of this invention has higher optimization efficiency. Its performance not only surpasses that of traditional genetic algorithms, but also outperforms the novel learning-based genetic algorithm (L-MOEA), providing a reliable basis for the comprehensive evaluation of the optimization effect and efficiency of the algorithm. Attached Figure Description
[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0022] Figure 1 This is a flowchart of the multi-objective automatic optimization method for meteorological flight routes based on a large language model in this invention.
[0023] Figure 2 This is a flowchart of the large language model-driven operator optimization process in this invention.
[0024] Figure 3 This is a flowchart illustrating the optimization of flight routes using the method of the present invention in an embodiment of the present invention. Detailed Implementation
[0025] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0027] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0028] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values of the components and steps described in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.
[0029] like Figure 1 As shown, this invention provides a multi-objective automatic optimization method for meteorological routes based on a large language model, including: acquiring ship parameters and navigation-related parameters, and setting optimization objectives and constraints. In a preferred embodiment of this invention, the ship parameters include ship tonnage, braking power, fuel consumption rate, service speed, and speed adjustment range. The navigation-related parameters include: latitude and longitude of the navigation start and end points, navigation season, and corresponding marine environmental prediction data. The marine environmental prediction data includes: significant wave height, wave period, wind speed, wind direction, ocean current speed, and ocean current direction.
[0030] The optimization objective is to minimize travel time and fuel consumption, with constraints including an unobstructed navigation area and maintaining speed within the controllable range of the main engine. Obstacles include coastlines, islands, reefs, and shoals.
[0031] Crossover and mutation operators are generated based on a large language model, an exploration strategy, and a correction strategy. In a preferred embodiment of this invention, the crossover and mutation operators include route crossover operators, speed crossover operators, route mutation operators, and speed mutation operators. Each type of operator must satisfy constraints such as maintaining consistent start and end points, speed meeting host performance requirements, and a clear path free of obstacles. The operator generation process is executed according to prompts. The large language model follows a description-then-encode workflow, first outputting a detailed algorithm description, then generating executable code. Each type of operator initially generates N individuals, forming an algorithm population.
[0032] The exploration strategies include differentiation exploration and similarity exploration. Differentiation exploration guides the large language model to generate heuristic algorithms that are significantly different from existing operators, thereby increasing the diversity of the operator population. Similarity exploration guides the large language model to retain the core features of existing high-quality operators, ensuring the continuity of the algorithm. The correction strategies include structural optimization and parameter tuning. Structural optimization is used to improve the operator architecture and enhance search capabilities, while parameter tuning is used to optimize key operator parameters and improve performance.
[0033] The crossover and mutation operators are evaluated using performance metrics, and the operators are optimized using a selection mechanism to obtain the optimal operator combination. In a preferred embodiment of this invention, the performance metric used is the scheme execution deviation index. This index is defined as the reciprocal of the average distance between the algorithm-generated scheme and the initial scheme in both flight time and fuel consumption dimensions. A lower value indicates better operator performance. The performance metric is expressed as:
[0034] in, This indicates the number of points included in the Pareto front. Indicates the generated first The difference in fuel dimension between each front point and the corresponding point of the initial reference front. Indicates the generated first The difference in time between each front point and the corresponding point of the initial reference front, a constant. To smooth the term and prevent the denominator from approaching zero when the distance between corresponding points at the leading edge is zero; the constant 10 is used as a normalization coefficient to scale the output value to a reasonable range to avoid loss of precision.
[0035] The selection mechanism employs a ranking-based random selection strategy, ranking operators according to their performance metrics. Operators ranked higher have a higher selection probability. Represented as:
[0036] in, The total number of operators, For the first Ranking of the algorithms.
[0037] Finally, an alternating iterative optimization is implemented, fixing three types of operators and performing a cyclical optimization of selection, evolution, evaluation, and elite retention on the fourth type of operator until the termination condition is met (performance metrics fluctuate for X consecutive generations or the maximum number of iterations K is reached). The optimization of the four types of operators is completed in sequence to obtain the optimal operator combination.
[0038] By combining optimal operator combinations with population initialization and population update, a roulette wheel selection method is used to select high-quality individuals and output the Pareto optimal solution set for the route. In a preferred implementation, population initialization involves randomly generating feasible navigation scheme individuals based on the great circle route and constraints. Each individual contains a sequence of route waypoints and corresponding speed parameters. Population update divides all candidate solutions into multiple non-dominated levels based on Pareto dominance, calculates the congestion distance (sum of Manhattan distances) for each individual, and prioritizes retaining high-level, sparsely distributed individuals to maintain population diversity. The roulette wheel selection method is used to select high-quality individuals, and optimal operator combinations are applied to generate new individuals. This process is repeated iteratively until the Pareto optimal solution set is output.
[0039] A convergence time efficiency index is introduced to evaluate the quality of the Pareto optimal solution set, thereby enabling automatic route optimization.
[0040] In a specific implementation, as a preferred embodiment of the present invention, the convergence time efficiency index is defined as the ratio of the generation distance of the operator to the total computation time. This index is used to comprehensively measure the convergence accuracy and computational efficiency of the algorithm. A higher value indicates better convergence efficiency. The convergence time efficiency index is expressed as:
[0041] in, This represents the convergence time efficiency index, used to comprehensively measure the convergence accuracy and computational efficiency of an algorithm. The generation distance is used to quantify the average distance between the solution set and the theoretical optimal frontier, reflecting the convergence accuracy. This represents the total computation time to complete one full optimization.
[0042] The convergence time efficiency index, combined with the hypervolume index and the generation distance index, is used to evaluate the quality of the Pareto optimal solution set. The hypervolume index is used to quantify the target space volume dominated by the solution set, and the generation distance index is used to measure the average distance between the solution set and the theoretical optimal frontier.
[0043] Example In this embodiment, a 47,203-ton container ship was selected as the experimental vessel. Its braking power is 28,710 kW, fuel consumption rate is 166 g / kWh, service speed is 23.80 knots, and speed adjustment range is 101-125 revolutions per minute. The navigation route was set from Kashima (35.25°N, 141.75°E) to Long Beach (36.75°N, 122.25°W), and marine environmental data from January, April, July, and October 2022 were used as experimental data. Following the steps of this invention, combined with... Figure 2 The operator optimization flowchart shown uses the DeepSeek-v3 model as the large language model. The maximum number of iterations for the algorithm population is set to 8 generations. Each generation generates 4 heuristic algorithms (corresponding to four strategies: differentiation exploration, similarity exploration, structure optimization, and parameter tuning). Each algorithm generates 8 elite candidate algorithms. The termination condition for operator evolution is set as performance metric fluctuation ≤5% for two consecutive generations. Through the above iterative optimization process, the optimal crossover and mutation operators are finally obtained. Combining the optimal operators with other framework structures of multi-objective genetic algorithms, such as... Figure 3 As shown, the total iteration cycle is set to 50 times, and the final output consists of 10 non-dominated solutions forming a Pareto optimal solution set. In the experiment, the existing speed loss formula is used to calculate the navigation time and fuel consumption, and the specific navigation scheme is updated. Finally, a Pareto optimal solution set that meets the dual-objective optimization requirements is output.
[0044] Compared with traditional great circle routes, this invention can avoid high-wave and high-altitude areas, achieving earlier flight times and a 2.2% reduction in fuel consumption while ensuring navigation safety. This invention was compared with traditional great circle routes, Non-Dominated Sorting Genetic Algorithm-II (NSGA-II), Intensity Pareto Evolutionary Algorithm 2 (SPEA2), Knee-Driven Evolutionary Algorithm (KnEA), Improved Genetic Algorithm (MGA), Local Search Multi-Objective Evolutionary Algorithm (MOEA-LS), and Learning-Assisted Multi-Objective Evolutionary Algorithm under the same experimental conditions. The experimental results are shown in Table 1.
[0045] Table 1 Comparison of Experimental Results
[0046] As shown in Table 1, compared with five traditional algorithms—NSGA-II, KnEA, MOEA-LS, SPEA2, and MGA—the Pareto optimal solution set generated by this invention outperforms the above five traditional algorithms in terms of convergence time efficiency under the same dataset. Compared with the learning-assisted multi-objective evolutionary algorithm, the Pareto optimal solution set generated by this invention has similar performance, but the convergence time of this invention is shorter, showing a significant advantage in convergence time efficiency. Furthermore, operators trained in different months show the best performance in the original scene, proving that the algorithm of this invention has good scene adaptability. In the extended experiments involving cross-regional and multi-ship types, this invention can automatically generate operators adapted to new scenes without manual intervention, and the convergence time meets the requirements of practical applications.
[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A multi-objective automatic optimization method for meteorological flight routes based on a large language model, characterized in that, include: Obtain ship parameters and navigation-related parameters, and set optimization objectives and constraints; Generate crossover and mutation operators based on large language models, exploration strategies, and correction strategies; The crossover and mutation operators are evaluated using performance metrics, and the operators are optimized using a selection mechanism to obtain the optimal combination of operators. By combining the optimal operator combination with population initialization and population update, the roulette wheel selection method is used to select high-quality individuals and output the Pareto optimal solution set of the route. A convergence time efficiency index is introduced to evaluate the quality of the Pareto optimal solution set, thereby achieving automatic route optimization.
2. The multi-objective automatic meteorological flight route optimization method based on a large language model according to claim 1, characterized in that, The ship parameters include ship tonnage, braking power, fuel consumption rate, service speed and speed adjustment range. The navigation-related parameters include: latitude and longitude of navigation start and end points, navigation season and corresponding marine environment prediction data. The marine environment prediction data includes: significant wave height, wave period, wind speed, wind direction, ocean current speed and ocean current direction. The optimization objective is set as minimizing flight time and fuel consumption, and the constraints include an unobstructed flight area and maintaining the speed within the controllable range of the main engine.
3. The multi-objective automatic meteorological flight route optimization method based on a large language model according to claim 1, characterized in that, The crossover and mutation operators include route crossover operators, speed crossover operators, route mutation operators, and speed mutation operators; the exploration strategies include differentiation exploration and similarity exploration; the correction strategies include structure optimization and parameter tuning. The structure optimization is used to improve the operator architecture and enhance search capabilities, while the parameter tuning is used to optimize key operator parameters and improve performance.
4. The multi-objective automatic meteorological flight route optimization method based on a large language model according to claim 1, characterized in that, The performance metric adopted is the scheme execution deviation index, which is defined as the reciprocal of the average distance between the algorithm-generated scheme and the initial scheme in two dimensions: flight time and fuel consumption. The performance metric is expressed as: in, This indicates the number of points included in the Pareto front. Indicates the generated first The difference in fuel dimension between each front point and the corresponding point of the initial reference front. Indicates the generated first The difference in time between each front point and the corresponding point of the initial reference front, a constant. To smooth out the terms and prevent the denominator from approaching zero when the distance between corresponding points at the leading edge is zero; The selection mechanism employs a ranking-based random selection strategy, ranking operators according to the values of the performance metrics, and assigning selection probabilities to the selected operators. Represented as: in, The total number of operators, For the first Ranking of the algorithms.
5. The multi-objective automatic meteorological flight route optimization method based on a large language model according to claim 1, characterized in that, The population initialization process involves randomly generating feasible navigation schemes individuals based on the great circle route and constraints. Each individual contains a sequence of route waypoints and corresponding speed parameters. The population update process divides all candidate solutions into multiple non-dominated levels based on the Pareto dominance relation, calculates the crowding distance for each individual, and prioritizes retaining high-level, sparsely distributed individuals to maintain population diversity. The roulette wheel selection method is used to select high-quality individuals, and the optimal operator combination is applied to generate new individuals. This process is repeated iteratively until a set of Pareto optimal solutions is output.
6. The multi-objective automatic meteorological flight route optimization method based on a large language model according to claim 1, characterized in that, The convergence time efficiency index is defined as the ratio of the generation distance of the operator to the total computation time, used to comprehensively measure the convergence accuracy and computational efficiency of the algorithm. The convergence time efficiency index is expressed as: in, This represents the convergence time efficiency index, used to comprehensively measure the convergence accuracy and computational efficiency of an algorithm. The generation distance is used to quantify the average distance between the solution set and the theoretical optimal frontier, reflecting the convergence accuracy. This represents the total computation time to complete one full optimization. The convergence time efficiency index, combined with the hypervolume index and the generation distance index, evaluates the quality of the Pareto optimal solution set. The hypervolume index is used to quantify the target space volume dominated by the solution set, and the generation distance index is used to measure the average distance between the solution set and the theoretical optimal frontier.