Aircraft flight path planning method, device and medium based on large language model

Abstract

The present invention discloses a method, device and medium for aircraft flight path planning based on a large language model, belonging to the technical field of aviation path planning. The present invention utilizes a pre-constructed large language area prediction model to obtain an initial flight path for the target aircraft that can avoid dangerous weather areas based on real-time weather information of cities along the target aircraft's flight mission. The initial flight path is then optimized to determine the optimal path planning strategy that prevents the target aircraft from colliding with other aircraft. The present invention can reasonably plan the aircraft's flight path based on the weather and the conditions of the airspace surrounding the aircraft, thereby improving flight safety and solving the current problem that aircraft path changes can only be guided by ground command.

Description

Technical Field

[0001] The present invention relates to a method, device and medium for aircraft flight path planning based on a large language model, and belongs to the technical field of aviation path planning. Background Art

[0002] When an aircraft encounters an emergency such as weather, the original flight route cannot be maintained and it is necessary to adjust the flight path in real time to implement evasive action to leave the danger zone. This dynamic circumvention behavior not only breaks the execution chain of the existing control and command plan, but also significantly increases the decision-making difficulty and workload of the air traffic control department. It is very easy to induce "missing and forgetting" risk events caused by command errors, which in turn causes economic losses to expand and even lead to secondary safety accidents.

[0003] In air traffic control operations, ground-to-air voice communication serves as the core means of interaction between air traffic controllers and flight crews, enabling real-time command transmission via a very high frequency (VHF) radio system. This interaction process incorporates a triple-confirmation mechanism: the controller issues commands via voice, the pilot repeats the command in full, and the controller provides final confirmation of the repeat, thus establishing a closed-loop verification system for command transmission. Given the dual requirements of safety margins and operational efficiency in civil aviation operations, existing voice recognition systems cannot be directly applied to real-time command scenarios, and their functional positioning is still limited to post-event analysis. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide an aircraft flight path planning method, device and medium based on a large language model, which can reasonably plan the aircraft's flight path according to the weather and the airspace conditions around the aircraft, thereby improving flight safety and solving the current problem that aircraft can only be guided to change their paths through ground commands.

[0005] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions:

[0006] A first aspect of the present invention provides an aircraft flight path planning method based on a large language model, comprising:

[0007] Obtain real-time weather information for cities along the flight route of the target aircraft;

[0008] Input real-time weather information into a pre-built large language area prediction model to obtain the target aircraft's initial flight path that can avoid hazardous weather areas;

[0009] Optimize the initial flight path to determine the optimal path planning strategy that will prevent the target aircraft from colliding with other aircraft in the surrounding airspace;

[0010] According to the optimal path planning strategy, the target aircraft is controlled to adjust the flight path.

[0011] Furthermore, inputting the real-time weather information into a pre-built large language area prediction model to obtain an initial flight path for the target aircraft that can avoid the dangerous weather area includes:

[0012] According to the real-time weather information, the Graham scanning method is used to obtain the initial boundary point set of the dangerous weather area;

[0013] Based on the initial boundary point set, the grey prediction model is used to predict the future boundary of the dangerous weather area and obtain the flight restricted area;

[0014] According to the flight restricted area, the initial flight path is obtained.

[0015] Furthermore, the path optimization of the initial flight path to determine the optimal path planning strategy that prevents the target aircraft from colliding with other aircraft in the surrounding airspace includes:

[0016] The SARSA algorithm is used to map the target aircraft and its initial flight path into the observable virtual environment space;

[0017] Construct a reward function corresponding to the target aircraft's actions in the virtual environment space;

[0018] The target aircraft is made to move in the virtual environment space, and the optimal path planning strategy is determined so that the target aircraft does not collide with other aircraft, can reach the destination and obtain the highest reward value.

[0019] Furthermore, the SARSA algorithm is used to map the target aircraft and its initial flight path to an observable virtual environment space, including:

[0020] The observable virtual environment space is represented by a tuple:

[0021] ;

[0022] in:

[0023] represents the state space, It consists of the state of the target aircraft at each moment;

[0024] represents the action space, It consists of the actions performed by the target aircraft at each moment;

[0025] represents the reward function, It consists of the reward given after the target aircraft performs an action at each moment.

[0026] Furthermore, the reward function includes:

[0027] Distance reward function, the distance reward function is expressed by the following formula:

[0028] ;

[0029] ;

[0030] ;

[0031] express The moment invader reward function, Indicates the distance threshold between the target aircraft and other aircraft in the surrounding airspace. Indicates the minimum distance between the target aircraft and other aircraft in the surrounding airspace;

[0032] express The remaining distance reward function at the moment, Represents a function for calculating the distance between the target aircraft and the destination;

[0033] express Reward function for reaching the destination at any moment;

[0034] Weather reward function, the weather reward function is expressed as follows:

[0035] ;

[0036] ;

[0037] ;

[0038] express The moment rainfall reward function, Indicates the amount of rainfall in the environment where the target aircraft is located;

[0039] express Moment visibility reward function, Indicates the visibility of the target aircraft's environment;

[0040] express The moment cloud cover reward function, Indicates the cloud cover in the target aircraft's environment;

[0041] The flight restricted area reward function is expressed as follows:

[0042] ;

[0043] represents the flight restricted area reward function, represents the maximum reward, Indicates the distance between the aircraft and the center of the minimum enclosing circle corresponding to the flight restricted area. Indicates the radius of the minimum enclosing circle corresponding to the flight restricted area.

[0044] Furthermore, the step of causing the target aircraft to move in the virtual environment space and determining an optimal path planning strategy in which the target aircraft does not collide with other aircraft, can reach the destination, and obtains the highest reward value includes:

[0045] The state-action value function is used to update the current state-action pair of the target aircraft so that the target aircraft can select the optimal path planning strategy for action. The updated current state-action pair of the target aircraft is expressed as:

[0046] ;

[0047] in:

[0048] Represents the current state-action pair estimated value of;

[0049] Represents the current state-action pair Update the estimated value of ;

[0050] Represents the learning rate, which is used to control the update step size;

[0051] represents the time series error, ;

[0052] Indicates the current state of the target aircraft. Indicates the action performed by the target aircraft at the current moment;

[0053] Indicates that the target aircraft performs an action The reward at the next moment, Indicates the state of the target aircraft at the next moment. Indicates the action to be performed by the target aircraft at the next moment;

[0054] Represents the next moment state-action pair estimated value of;

[0055] represents the discount factor;

[0056] The following formula is used to calculate the reward value corresponding to the target aircraft's action in the virtual environment space:

[0057] ;

[0058] express 、 as well as The minimum value in .

[0059] Furthermore, based on the initial boundary point set, the future boundary of the dangerous weather area is predicted using a grey prediction model to obtain the flight restricted area, including:

[0060] Observe the initial boundary point set at fixed time intervals to obtain the original observation sequence set representing the change of the initial boundary over time;

[0061] Calculate the gray parameters of each initial boundary point corresponding to the original observation sequence;

[0062] Input the gray parameters into the gray prediction model to obtain the coordinates of the boundary points at the future time;

[0063] The flight restricted area is obtained according to the coordinates of the boundary points at each future moment.

[0064] Furthermore, the calculation of the gray parameter of each initial boundary point corresponding to the original observation sequence includes:

[0065] The gray parameter of each initial boundary point corresponding to the original observation sequence is calculated by the following formula:

[0066] ;

[0067] ;

[0068] in:

[0069] Represents the gray parameter of the original observation sequence corresponding to the horizontal coordinate of the initial boundary point;

[0070] Represents the accumulated data of the horizontal coordinates of the initial boundary points, represents the accumulation matrix of the horizontal coordinates, represents the abscissa constant vector;

[0071] Represents the gray parameter of the original observation sequence corresponding to the ordinate of the initial boundary point;

[0072] Indicates the accumulated data of the initial boundary point ordinates, The cumulative matrix representing the vertical coordinate; Represents the vertical coordinate constant vector.

[0073] A second aspect of the present invention provides an aircraft flight path planning device based on a large language model, comprising:

[0074] An acquisition module is used to obtain real-time weather information of cities along the flight route of the target aircraft;

[0075] A path planning module is used to input real-time weather information into a pre-built large language area prediction model to obtain an initial flight path for the target aircraft that can avoid hazardous weather areas;

[0076] The path optimization module is used to optimize the initial flight path and determine the optimal path planning strategy to prevent the target aircraft from colliding with other aircraft;

[0077] The processing module is used to control the target aircraft to adjust the flight path according to the optimal path planning strategy.

[0078] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for aircraft flight path planning based on a large language model as described above is implemented.

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] 1. The present invention utilizes a pre-built large language area prediction model and, based on real-time weather information from cities along the target aircraft's flight route, obtains an initial flight path for the target aircraft that avoids hazardous weather areas. This initial flight path is then optimized to determine the optimal path planning strategy that prevents the target aircraft from colliding with other aircraft. This invention rationally plans the aircraft's flight path based on weather conditions and surrounding airspace conditions, improving flight safety and resolving the current problem of aircraft path changes requiring only ground-based command and guidance.

[0081] 2. The present invention uses the SARSA algorithm to map the target aircraft and its initial flight path to an observable virtual environment space, allowing the target aircraft to move in the virtual environment space and combining the reward function to evaluate the state of the target aircraft after performing the corresponding action, which can quickly and effectively optimize the initial flight path. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 This is a flow chart of a method for aircraft flight path planning based on a large language model provided by an embodiment of the present invention;

[0083] Figure 2Schematic diagram of weather reward values at different times in four cities under hazy weather conditions provided by an embodiment of the present invention;

[0084] Figure 3 Schematic diagram of weather reward values at different times in four cities on sunny days provided by an embodiment of the present invention;

[0085] Figure 4 Schematic diagram of weather reward values at different times in four cities under light snow conditions provided by an embodiment of the present invention;

[0086] Figure 5 Schematic diagram of average aircraft reward values under three typical weather scenarios provided by an embodiment of the present invention;

[0087] Figure 6 This is a schematic diagram of the trend of the number of aircraft heading angle modifications provided by an embodiment of the present invention;

[0088] Figure 7 This is a schematic diagram of the trend of the number of aircraft speed modifications provided by an embodiment of the present invention;

[0089] Figure 8 1 is a schematic diagram of the trend of the number of aircraft altitude modifications provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0090] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0091] Example 1

[0092] like Figure 1 As shown, a method for aircraft flight path planning based on a large language model includes:

[0093] Obtain real-time weather information for cities along the flight route of the target aircraft;

[0094] Input real-time weather information into a pre-built large language area prediction model to obtain the initial flight path of the target aircraft that can avoid dangerous weather areas. Specifically:

[0095] In this embodiment, the real-time weather information is a satellite cloud map;

[0096] De-noise and merge areas with hazardous weather conditions in the satellite cloud image. Ignore areas smaller than 7 miles and merge areas with a minimum distance of less than 20 km. Select a reference point near the area as the coordinate origin O and construct a plane rectangular coordinate system with the north direction as the positive Y-axis and the direction 90° east of the north as the positive X-axis.

[0097] The Graham scanning method is used to obtain the convex shape of the dangerous weather area at time The initial boundary point set , ,in Indicates that dangerous weather areas are at No. The coordinates of the initial boundary points, , Indicates Moment The horizontal coordinates of the initial boundary points, Indicates Moment The ordinate of the initial boundary point;

[0098] , Indicates shared A moment, , Indicates shared Initial boundary points;

[0099] Based on the initial boundary point set, the grey prediction model is used to predict the future boundary of the dangerous weather area, and the flight restricted area is obtained, including:

[0100] Observe the initial boundary point set at fixed time intervals to obtain the original observation sequence set representing the change of the initial boundary over time;

[0101] Calculating the gray parameter of each initial boundary point corresponding to the original observation sequence includes: calculating the gray parameter of each initial boundary point corresponding to the original observation sequence by the following formula:

[0102] ;

[0103] ;

[0104] Indicates the gray parameter of the original observation sequence corresponding to the horizontal coordinate of the initial boundary point. The original observation sequence of the horizontal coordinates of the initial boundary points is expressed as ; ;

[0105] Represents the accumulated data of the horizontal coordinates of the initial boundary points, ;

[0106] ; Indicates that The first of the moments The moment The horizontal coordinates of the initial boundary points;

[0107] represents the accumulation matrix of the horizontal coordinate;

[0108] ;

[0109] represents the abscissa constant vector;

[0110] ;

[0111] Indicates the gray parameter of the original observation sequence of the ordinate corresponding to the initial boundary point, The original observation sequence of the ordinate of the initial boundary points is expressed as ; ;

[0112] Indicates the accumulated data of the initial boundary point ordinates, ;

[0113] ; Indicates that The first of the moments The moment The ordinate of the initial boundary point;

[0114] The cumulative matrix representing the vertical coordinate;

[0115] ;

[0116] represents the vertical coordinate constant vector;

[0117] ;

[0118] Input the gray parameters into the gray prediction model to obtain the boundary point coordinate set at the future time ;

[0119]

[0120] Indicates that dangerous weather areas will occur in the future Moment The coordinates of the predicted boundary points, ; In the future Moment The horizontal coordinates of the predicted boundary points, In the future Moment The vertical coordinate of the predicted boundary point;

[0121] Calculate the future by The midpoint position of the convex polygon at the moment :

[0122] ;

[0123] The midpoint of the convex polygon and the predicted boundary point are calculated by the following formula The slope of the line:

[0124] ;

[0125] In this embodiment, the reserve buffer zone is widened by 25 km outside the convex polygon. When the target aircraft enters the reserve buffer zone, it is considered to have entered the dangerous weather area. The following formula is used to calculate the future after the widening: Boundary point coordinates at time :

[0126] ;

[0127] ;

[0128] The initial flight path is obtained according to the convex polygonal flight restricted area surrounded by the coordinates of the boundary points at the widened future time.

[0129] Optimize the initial flight path to determine the optimal path planning strategy that prevents the target aircraft from colliding with other aircraft in the surrounding airspace. Specifically:

[0130] The SARSA algorithm is used to map the target aircraft and its initial flight path into an observable virtual environment space, including:

[0131] The observable virtual environment space is represented by a tuple:

[0132] ;

[0133] represents the state space, The state corresponding to each moment composition, , Indicates the position coordinates of the target aircraft, represents the speed of the target aircraft, Indicates the heading angle of the target aircraft;

[0134] represents the action space, Actions performed by the target aircraft at each moment composition, , Indicates the heading angle change of the target aircraft, Indicates the speed change of the target aircraft, Indicates the altitude change of the target aircraft; the aircraft mainly changes its heading angle, speed and altitude through the actions shown in Table 1;

[0135] Table 1 Specific action description

[0136]

[0137] represents the reward function, The reward given by the target aircraft after performing an action at each moment composition;

[0138] Construct a reward function corresponding to the target aircraft's actions in the virtual environment space, where the reward function includes:

[0139] Distance reward function, the distance reward function is expressed by the following formula:

[0140] ;

[0141] ;

[0142] ;

[0143] express The moment invader reward function, Indicates the distance threshold between the target aircraft and other aircraft in the surrounding airspace. Indicates the minimum distance between the target aircraft and other aircraft in the surrounding airspace;

[0144] express The remaining distance reward function at the moment, Represents a function for calculating the distance between the target aircraft and the destination;

[0145] express Reward function for reaching the destination at any moment;

[0146] Weather reward function, the weather reward function is expressed as follows:

[0147] ;

[0148] ;

[0149] ;

[0150] express The reward function for rainfall at each moment, Indicates the amount of rainfall in the environment where the target aircraft is located;

[0151] express Moment visibility reward function, Indicates the visibility of the target aircraft's environment;

[0152] express The moment cloud cover reward function, Indicates the cloud cover in the target aircraft's environment;

[0153] The reward function also includes the flight restricted area reward function, specifically:

[0154] Based on the convex polygon flight restricted area, the minimum enclosing circle radius that can enclose the entire flight restricted area is calculated to construct the flight restricted area reward function:

[0155] ;

[0156] represents the flight restricted area reward function, represents the maximum reward, ;

[0157] Indicates the distance between the aircraft and the center of the minimum enclosing circle corresponding to the flight restricted area. Indicates the radius of the minimum enclosing circle corresponding to the flight restricted area.

[0158] The target aircraft is made to move in the virtual environment space, and the optimal path planning strategy is determined so that the target aircraft does not collide with other aircraft, can reach the destination, and obtain the highest reward value, including:

[0159] The state-action value function is used to update the current state-action pair of the target aircraft so that the target aircraft can select the optimal path planning strategy for action. The updated current state-action pair of the target aircraft is expressed as:

[0160] ;

[0161] Represents the current state-action pair estimated value of; Represents the current state-action pair Update the estimated value of ; Represents the learning rate, which is used to control the update step size; represents the time series error, ; Indicates the current state of the target aircraft. Indicates that the target aircraft is currently Actions performed at all times; Indicates the reward at the next moment after executing the current action, Indicates the state of the target aircraft at the next moment. Indicates the action to be performed by the target aircraft at the next moment;

[0162] Represents the next moment state-action pair estimated value of;

[0163] represents the discount factor;

[0164] The following formula is used to calculate the reward value corresponding to the target aircraft's action in the virtual environment space:

[0165] ;

[0166] express 、 as well as The minimum value in .

[0167] The target aircraft is controlled by actions corresponding to the optimal path planning strategy.

[0168] In order to verify the method proposed in this example, the following data are used for simulation experiments:

[0169] Software: Python, select the map of China for spatial mapping;

[0170] like Figure 2 、 Figure 3 and Figure 4 As shown in the figure, April 28, 2024, July 31, 2024, and December 10, 2024 were selected as three typical weather scenes: foggy, sunny, and snowy. The two-way passenger route between ZGGG (Guangzhou Baiyun International Airport) and ZWWW (Urumqi Tianshan International Airport) was used for example analysis. In the actual weather system experiment, the route mainly passes through two node cities, Lanzhou and Changsha. Grey prediction was used to construct real-time flight restricted areas in the two node cities.

[0171] like Figure 5As shown in the figure, at the beginning of training, as the number of rounds increases, the average reward value of the aircraft increases rapidly and reaches a peak. Subsequently, due to the influence of weather penalties, the average reward value slowly decreases. After about 5000 training rounds, it fluctuates slightly around 1510, 1264, and 1179, showing a stable trend. This shows that after a large number of iterative training, the aircraft has adapted to the negative impact of the environment.

[0172] like Figure 6 、 Figure 7 and Figure 8 As shown, the number of aircraft heading angle modifications is significantly greater than the number of speed and altitude modifications. This indicates that when facing a potential conflict, the aircraft prioritizes adjusting its heading angle to avoid collision. This is highly consistent with the command operations of traditional controllers. The method proposed in this embodiment is consistent with human experience in decision-making logic.

[0173] Example 2

[0174] An aircraft flight path planning device based on a large language model, comprising:

[0175] An acquisition module is used to obtain real-time weather information of cities along the flight route of the target aircraft;

[0176] A path planning module is used to input real-time weather information into a pre-built large language area prediction model to obtain an initial flight path for the target aircraft that can avoid hazardous weather areas;

[0177] The path optimization module is used to optimize the initial flight path and determine the optimal path planning strategy to prevent the target aircraft from colliding with other aircraft;

[0178] The processing module is used to control the target aircraft to adjust the flight path according to the optimal path planning strategy.

[0179] Example 3

[0180] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the following method:

[0181] Obtain real-time weather information for cities along the flight route of the target aircraft;

[0182] Input real-time weather information into a pre-built large language area prediction model to obtain the target aircraft's initial flight path that can avoid hazardous weather areas;

[0183] Optimize the initial flight path to determine the optimal path planning strategy that will prevent the target aircraft from colliding with other aircraft in the surrounding airspace;

[0184] According to the optimal path planning strategy, the target aircraft is controlled to adjust the flight path.

[0185] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, apparatuses, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0186] The present application is described with reference to the flowcharts of the methods, apparatus (systems), and computer program products according to the embodiments of the present application. It should be understood that each process in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 a process or multiple processes or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0187] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A function specified in a process or multiple processes.

[0188] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 The steps of a specified function in a process or multiple processes.

[0189] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which are all protected by the present invention.

Claims

1. An aircraft flight path planning method based on a large language model, characterized in that: include: Obtain real-time weather information for cities along the flight route of the target aircraft; Input real-time weather information into a pre-built large language area prediction model to obtain an initial flight path for the target aircraft that can avoid hazardous weather areas, including: According to the real-time weather information, the Graham scanning method is used to obtain the initial boundary point set of the dangerous weather area; Based on the initial boundary point set, the grey prediction model is used to predict the future boundary of the dangerous weather area and obtain the flight restricted area; Obtain the initial flight path based on the flight restricted area; Optimize the initial flight path to determine the optimal path planning strategy that prevents the target aircraft from colliding with other aircraft in the surrounding airspace, including: The SARSA algorithm is used to map the target aircraft and its initial flight path into an observable virtual environment space, including: The observable virtual environment space is represented by a tuple: ； in: represents the state space, It consists of the state of the target aircraft at each moment; represents the action space, It consists of the actions performed by the target aircraft at each moment; represents the reward function, It consists of the reward given after the target aircraft performs an action at each moment; Construct a reward function corresponding to the target aircraft's actions in the virtual environment space; Make the target aircraft move in the virtual environment space and determine the optimal path planning strategy that allows the target aircraft to avoid collision with other aircraft, reach the destination, and obtain the highest reward value; According to the optimal path planning strategy, the target aircraft is controlled to adjust the flight path.

2. The aircraft flight path planning method based on a large language model according to claim 1, characterized in that: The reward function includes: Distance reward function, the distance reward function is expressed by the following formula: ； ； ； express The moment invader reward function, Indicates the distance threshold between the target aircraft and other aircraft in the surrounding airspace. Indicates the minimum distance between the target aircraft and other aircraft in the surrounding airspace; express The remaining distance reward function at the moment, Represents a function for calculating the distance between the target aircraft and the destination; express Reward function for reaching the destination at any moment; Weather reward function, the weather reward function is expressed as follows: ； ； ； express The moment rainfall reward function, Indicates the amount of rainfall in the environment where the target aircraft is located; express Moment visibility reward function, Indicates the visibility of the target aircraft's environment; express The moment cloud cover reward function, Indicates the cloud cover in the target aircraft's environment; The flight restricted area reward function is expressed as follows: ； represents the flight restricted area reward function, represents the maximum reward, Indicates the distance between the aircraft and the center of the minimum enclosing circle corresponding to the flight restricted area. Indicates the radius of the minimum enclosing circle corresponding to the flight restricted area.

3. The aircraft flight path planning method based on a large language model according to claim 2, characterized in that: The method of causing the target aircraft to move in the virtual environment space and determining an optimal path planning strategy in which the target aircraft does not collide with other aircraft, can reach the destination, and obtains the highest reward value includes: The state-action value function is used to update the current state-action pair of the target aircraft so that the target aircraft can select the optimal path planning strategy for action. The updated current state-action pair of the target aircraft is expressed as: ； in: Represents the current state-action pair estimated value of; Represents the current state-action pair Update the estimated value of ; Represents the learning rate, which is used to control the update step size; represents the time series error, ; Indicates the current state of the target aircraft. Indicates the action performed by the target aircraft at the current moment; Indicates that the target aircraft performs an action The reward at the next moment, Indicates the state of the target aircraft at the next moment. Indicates the action to be performed by the target aircraft at the next moment; Represents the next moment state-action pair estimated value of; represents the discount factor; The following formula is used to calculate the reward value corresponding to the target aircraft's action in the virtual environment space: ； express 、 as well as The minimum value in .

4. The aircraft flight path planning method based on a large language model according to claim 1, characterized in that: Based on the initial boundary point set, the future boundary of the dangerous weather area is predicted using a grey prediction model to obtain the flight restricted area, including: Observe the initial boundary point set at fixed time intervals to obtain the original observation sequence set representing the change of the initial boundary over time; Calculate the gray parameters of each initial boundary point corresponding to the original observation sequence; Input the gray parameters into the gray prediction model to obtain the coordinates of the boundary points at the future time; The flight restricted area is obtained according to the coordinates of the boundary points at each future moment.

5. The aircraft flight path planning method based on a large language model according to claim 4, characterized in that: The calculation of the gray parameters of the original observation sequence corresponding to each initial boundary point includes: The gray parameter of each initial boundary point corresponding to the original observation sequence is calculated by the following formula: ； ； in: Represents the gray parameter of the original observation sequence corresponding to the horizontal coordinate of the initial boundary point; Represents the accumulated data of the horizontal coordinates of the initial boundary points, represents the accumulation matrix of the horizontal coordinates, represents the abscissa constant vector; Represents the gray parameter of the original observation sequence corresponding to the ordinate of the initial boundary point; Indicates the accumulated data of the initial boundary point ordinates, The cumulative matrix representing the vertical coordinate; Represents the vertical coordinate constant vector.

6. An aircraft flight path planning device based on a large language model, characterized in that: include: An acquisition module is used to obtain real-time weather information of cities along the flight route of the target aircraft; The path planning module is used to input real-time weather information into a pre-built large language area prediction model to obtain an initial flight path for the target aircraft that can avoid hazardous weather areas, including: According to the real-time weather information, the Graham scanning method is used to obtain the initial boundary point set of the dangerous weather area; Based on the initial boundary point set, the grey prediction model is used to predict the future boundary of the dangerous weather area and obtain the flight restricted area; Obtain the initial flight path based on the flight restricted area; The path optimization module is used to optimize the initial flight path and determine the optimal path planning strategy to prevent the target aircraft from colliding with other aircraft. It includes: The SARSA algorithm is used to map the target aircraft and its initial flight path into an observable virtual environment space, including: The observable virtual environment space is represented by a tuple: ； in: represents the state space, It consists of the state of the target aircraft at each moment; represents the action space, It consists of the actions performed by the target aircraft at each moment; represents the reward function, It consists of the reward given after the target aircraft performs an action at each moment; Construct a reward function corresponding to the target aircraft's actions in the virtual environment space; Make the target aircraft move in the virtual environment space and determine the optimal path planning strategy that allows the target aircraft to avoid collision with other aircraft, reach the destination, and obtain the highest reward value; The processing module is used to control the target aircraft to adjust the flight path according to the optimal path planning strategy.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the aircraft flight path planning method based on a large language model as claimed in any one of claims 1 to 5 is implemented.

Images