Satellite orbit resource allocation method and system based on AI large model

Abstract

The invention discloses a satellite orbit resource allocation method and system based on an AI large model, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining the position information and task information of each communication satellite in a communication satellite system when an orbit resource allocation request is received; establishing a resource allocation basic algorithm according to the track resource allocation request; optimizing the resource allocation basic algorithm through a big data optimization algorithm; and constructing a target AI resource allocation large model according to the target resource allocation algorithm, and dynamically allocating orbit resources to each communication satellite according to the position information and the task information based on the target AI resource allocation large model. Through the above mode, the resource allocation basic algorithm is optimized by using the big data optimization algorithm, and the orbit resources are dynamically allocated to the communication satellites according to the model constructed by the target resource allocation algorithm, so that the accuracy of allocation of the orbit resources can be effectively improved, dynamic allocation of the orbit resources is realized, and a system scheme is provided.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to a satellite orbit resource allocation method and system based on an AI large model. Background Art

[0002] With the rapid development of satellite communication and Earth observation technologies, the scale and complexity of satellite networks have been continuously increasing. While this increase brings many opportunities, it is also accompanied by a series of challenges. For example, the allocation of satellite orbit resources. Currently, the common methods for allocating orbit resources rely on inherent setting rules and limited optimization algorithms, resulting in the resources finally allocated to satellites being unable to meet the dynamically changing requirements. Based on this, the accuracy of allocating orbit resources by the above methods is relatively low.

[0003] The above content is only used to assist in understanding the technical solution of this application, and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide a satellite orbit resource allocation method and system based on an AI large model, aiming to solve the technical problem of relatively low accuracy in allocating orbit resources in the prior art.

[0005] To achieve the above purpose, this application proposes a satellite orbit resource allocation method and system based on an AI large model. The method includes:

[0006] When receiving an orbit resource allocation request, obtain the position information and task information of each communication satellite in the communication satellite system;

[0007] Establish a basic resource allocation algorithm according to the orbit resource allocation request;

[0008] Establish a big data optimization algorithm, and optimize the basic resource allocation algorithm through the big data optimization algorithm to obtain a target resource allocation algorithm;

[0009] Construct a target AI resource allocation large model according to the target resource allocation algorithm, and based on the target AI resource allocation large model, dynamically allocate orbit resources for each communication satellite according to the position information and the task information.

[0010] In one embodiment, the step of constructing a target AI resource allocation large model according to the target resource allocation algorithm includes:

[0011] Obtain a multi-dimensional learning framework in the field of communication satellites;

[0012] Obtain the historical position information and historical task information of each communication satellite in the communication satellite system;

[0013] Based on the target resource allocation algorithm, determine the target historical orbital resources corresponding to the historical location information and the historical task information;

[0014] Based on the target historical orbital resources, train the multi-dimensional learning framework to obtain a target AI resource allocation large model.

[0015] In one embodiment, after the step of determining the target historical orbital resources corresponding to the historical location information and the historical task information based on the target resource allocation algorithm, the method further includes:

[0016] Dynamically allocate the historical orbital resources to each communication satellite in the communication satellite system;

[0017] Perform performance detection on each communication satellite in the communication satellite system to obtain communication performance characteristic values and coverage performance characteristic values;

[0018] Calculate the target performance characteristic value according to the communication performance characteristic value, communication weight, the coverage performance characteristic value, and the coverage weight. Specifically:

[0019] Total xn =tx xn *Q1+fg xn *Q2;

[0020] Where, Total xn represents the target performance characteristic value, tx xn represents the communication performance characteristic value, Q1 represents the communication weight, fg xn represents the coverage performance characteristic value, and Q2 represents the coverage weight;

[0021] When the target performance characteristic value is greater than a preset threshold, based on the target resource allocation algorithm, determine the target historical orbital resources corresponding to the historical location information and the historical task information.

[0022] In one embodiment, the step of training the multi-dimensional learning framework based on the target historical orbital resources to obtain a target AI resource allocation large model includes:

[0023] Generate initial training samples according to the target historical orbital resources, historical location information, and historical task information;

[0024] Perform data cleaning on the initial training samples, and perform normalization processing on the cleaned initial training samples;

[0025] Rotate the normalized initial training samples according to a preset rotation angle based on a target interpolation strategy, and flip the rotated initial training samples based on a target flip axis;

[0026] Normalize the initial training original after flipping to obtain target training samples; wherein, the target training samples include model training samples, model validation samples, and model test samples;

[0027] Train the multi-dimensional learning framework based on the model training samples to obtain an initial AI resource allocation large model;

[0028] Calculate the loss value between the predicted value and the true value of the initial AI resource allocation large model by using the MLP head with the cross-entropy loss function, specifically:

[0029]

[0030] Among them, loss represents the loss value between the predicted value and the true value of the initial AI resource allocation large model, P represents the number of samples, C represents the total number of categories for resource allocation, y i,c represents the true value, and p i,c represents the predicted value;

[0031] Perform validation training on the initial AI resource allocation large model based on the model validation samples until the loss value converges to a preset value.

[0032] In one embodiment, after the step of performing validation training on the initial AI resource allocation large model based on the model validation samples until the loss value converges to a preset value, the following steps are further included:

[0033] Test the initial AI resource allocation large model based on the model test samples to obtain the number of correctly allocated resource sample instances, the number of correctly allocated non-resource sample instances, the number of wrongly allocated resource sample instances, and the number of wrongly classified non-resource sample instances;

[0034] Calculate the track resource allocation accuracy rate according to the number of correctly allocated resource sample instances, the number of correctly allocated non-resource sample instances, the number of wrongly allocated resource sample instances, and the number of wrongly classified non-resource sample instances, specifically:

[0035]

[0036] Among them, O represents the track resource allocation accuracy rate, TP represents the number of correctly allocated resource sample instances, TN represents the number of correctly allocated non-resource sample instances, FP represents the number of wrongly allocated resource sample instances, and FN represents the number of wrongly allocated non-resource sample instances;

[0037] When the track resource allocation accuracy rate is greater than or equal to the target threshold, use the initial AI resource allocation large model as the target AI resource allocation large model;

[0038] When the correct rate of the orbital resource allocation is less than the target threshold, return the step of validating and training the initial AI resource allocation large model based on the model validation samples until the loss value converges to a preset value.

[0039] In one embodiment, the step of establishing the big data optimization algorithm includes:

[0040] Obtain the application scenarios of the communication satellite system, and determine the resource allocation optimization function and multi-dimensional constraint conditions according to the application scenarios, where the resource allocation optimization function is:

[0041]

[0042] where D represents the total cost of minimizing resource allocation, N represents the number of resources, M represents the number of tasks, c ab represents the cost of allocating resource a to communication satellite b, and x ab represents the decision variable that minimizes the single cost;

[0043] where the multi-dimensional constraint conditions are:

[0044]

[0045] where the first line of the formula represents the constraint condition that the usage amount of resource a cannot exceed the capacity A a and a ab represents resource a allocated to communication satellite b. Since the cost c of allocating resource a to communication satellite b ab is relatively high, if it exceeds the capacity A a , it will cause the total cost D of resource allocation to be extremely high and other communication satellites cannot be allocated sufficient resources; the second line of the formula represents the resources that must be satisfied when the communication satellite performs different tasks, and b ab represents the contribution amount of resource a to communication satellite b, D b represents the demand of communication satellite b. The ultimate optimization goal is to make the contribution amount of resource a to communication satellite b at least meet the demand of the communication satellite and minimize the total cost D of resource allocation; the third line of the formula represents the time window constraint condition, and t b represents the time when communication satellite b performs the task; the fourth line of the formula represents the mutual exclusion constraint condition. When X bta = 1, it indicates that communication satellite b occupies resource a at time t, and at this time resource a cannot be occupied by other communication satellites. When X bta = 0, it indicates that communication satellite b does not occupy resource a at time t, and at this time resource a can be occupied by other communication satellites, that is, the same resource can only be occupied by a single communication satellite at the same moment;

[0046] Establish a big data optimization algorithm based on the resource allocation optimization function and the multi-dimensional constraint conditions.

[0047] In one embodiment, the step of optimizing the basic resource allocation algorithm through the big data optimization algorithm to obtain the target resource allocation algorithm includes:

[0048] Determine the optimal solution of resource allocation through the big data optimization algorithm;

[0049] Optimize each allocation parameter in the basic resource allocation algorithm according to the optimal solution of resource allocation;

[0050] Generate a target resource allocation algorithm according to the optimized allocation parameters.

[0051] In one embodiment, after the step of dynamically allocating orbital resources for each communication satellite according to the position information and the task information, the method further includes:

[0052] Obtain frequency interference data and communication quality data generated by each satellite at the next moment;

[0053] Determine the current frequency interference change value and the current communication quality change value according to the frequency interference data and communication quality data generated by each satellite at the next moment and the frequency interference data and communication quality data generated by each satellite at the current moment;

[0054] Determine the frequency interference weight and the communication quality weight; calculate the frequency interference reference value according to the current frequency interference change value and the frequency interference weight, and calculate the communication quality reference value according to the current communication quality change value and the communication quality weight;

[0055] When the frequency interference reference value is less than the first threshold and the communication quality reference value is greater than the second threshold, perform dynamic allocation of orbital resources for other communication satellite systems.

[0056] In addition, to achieve the above object, the present application also proposes a satellite orbital resource allocation system based on an AI large model, the system includes:

[0057] An acquisition module, configured to acquire the position information and task information of each communication satellite in the communication satellite system when receiving an orbital resource allocation request;

[0058] A establishment module, configured to establish a basic resource allocation algorithm according to the orbital resource allocation request;

[0059] An optimization module, configured to establish a big data optimization algorithm and optimize the basic resource allocation algorithm through the big data optimization algorithm to obtain a target resource allocation algorithm;

[0060] An allocation module is configured to construct a target AI resource allocation large model according to the target resource allocation algorithm, and based on the target AI resource allocation large model, dynamically allocate orbital resources for each communication satellite according to the position information and the task information.

[0061] One or more technical solutions proposed in this application have at least the following technical effects: when receiving an orbital resource allocation request, obtaining the position information and task information of each communication satellite in the communication satellite system; establishing a basic resource allocation algorithm according to the orbital resource allocation request; establishing a big data optimization algorithm, and optimizing the basic resource allocation algorithm through the big data optimization algorithm to obtain a target resource allocation algorithm; constructing a target AI resource allocation large model according to the target resource allocation algorithm, and based on the target AI resource allocation large model, dynamically allocate orbital resources for each communication satellite according to the position information and the task information. By the above method, after establishing the big data optimization algorithm, the basic resource allocation algorithm is optimized through the big data optimization algorithm. At this time, the optimal target resource allocation algorithm is used to establish a target AI resource allocation large model for dynamically allocating orbital resources, and then combined with the position information and task information of each communication satellite to dynamically allocate orbital resources for each communication satellite, so as to effectively improve the accuracy of allocating orbital resources and realize the dynamic allocation of orbital resources. Brief Description of the Drawings

[0062] The drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with this application, and are used together with the specification to explain the principles of this application.

[0063] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0064] Figure 1 It is a schematic flowchart provided for Embodiment 1 of the satellite orbital resource allocation method and system based on the AI large model of this application;

[0065] Figure 2 It is a schematic flowchart provided for Embodiment 2 of the satellite orbital resource allocation method and system based on the AI large model of this application;

[0066] Figure 3 It is a schematic diagram of the module structure of the satellite orbital resource allocation system based on the AI large model for the embodiments of this application.

[0067] The implementation, functional features, and advantages of the purpose of this application will be further described in conjunction with the embodiments and with reference to the drawings. Detailed implementation manners

[0068] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication, and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device, an orbital resource allocation device, etc. that can implement the above functions. Hereinafter, taking the orbital resource allocation device as an example, this embodiment and the following embodiments will be described.

[0069] Based on this, the embodiments of the present application provide a satellite orbital resource allocation method and system based on an AI large model. Refer to Figure 1 , Figure 1 which is a schematic flowchart of the first embodiment of the satellite orbital resource allocation method and system based on the AI large model of the present application.

[0070] In this embodiment, the satellite orbital resource allocation method and system based on the AI large model include steps S10 to S40:

[0071] Step S10, when receiving an orbital resource allocation request, obtain the position information and task information of each communication satellite in the communication satellite system.

[0072] It should be noted that the orbital resource allocation request refers to a request initiated by a user terminal when orbital resources need to be allocated. The position information represents the positions of each communication satellite in the communication satellite system at the current moment. For example, communication satellite A is located in the first orbit with coordinates (x1, y1, z1), communication satellite B is located in the second orbit with coordinates (x2, y2, z2), communication satellite C is located in the third orbit with coordinates (x3, y3, z3), etc. The task information represents the tasks to be performed by the communication satellites. For example, the task information of communication satellite A is to achieve fast transmission of information such as voice, data, and video globally, and the task information of communication satellite B is to establish a temporary communication link to ensure the smooth progress of rescue work.

[0073] Step S20, establish a basic resource allocation algorithm according to the orbital resource allocation request.

[0074] It can be understood that the basic resource allocation algorithm refers to an algorithm commonly used for resource allocation. This basic resource allocation algorithm is an algorithm that allocates resources in a fixed form. For example, no matter when an orbital resource allocation request is received, communication satellite A is always allocated fixed resource a, communication satellite B is allocated fixed resource b, and communication satellite C is allocated fixed resource c.

[0075] Step S30, establish a big data optimization algorithm, and optimize the basic resource allocation algorithm through the big data optimization algorithm to obtain a target resource allocation algorithm.

[0076] It should be understood that the big data optimization algorithm refers to an algorithm that optimizes the basic resource allocation algorithm. The big data optimization algorithm can be established based on the resource allocation optimization function and multi-dimensional constraint conditions. The target resource allocation algorithm refers to the optimal algorithm for dynamically allocating orbital resources. The target resource allocation algorithm can be obtained by optimizing the basic resource allocation algorithm. The dimensions of optimization include but are not limited to parameters, allocation strategies, etc.

[0077] Furthermore, the steps of establishing the big data optimization algorithm include: obtaining the application scenario of the communication satellite system, and determining the resource allocation optimization function and multi-dimensional constraint conditions according to the application scenario. Among them, the resource allocation optimization function is:

[0078]

[0079] where D represents the total cost of minimizing resource allocation, N represents the number of resources, M represents the number of tasks, c ab represents the cost of allocating resource a to communication satellite b, and x ab represents the decision variable that minimizes the single cost;

[0080] Among them, the multi-dimensional constraint conditions are:

[0081]

[0082] Among them, the formula in the first line above represents the constraint condition that the usage amount of resource a cannot exceed the capacity A a a ab represents resource a allocated to communication satellite b. Since the cost c ab of allocating resource a to communication satellite b is relatively high, if it exceeds the capacity A a , it will cause the total cost D of resource allocation to be extremely high and other communication satellites cannot be allocated sufficient resources; the formula in the second line represents the resources that must be satisfied when the communication satellite performs different tasks. b ab represents the contribution amount of resource a to communication satellite b, D b represents the demand of communication satellite b. The ultimate optimization goal is to make the contribution amount of resource a to communication satellite b at least meet the demand of the communication satellite and the total cost D of resource allocation is the lowest; the formula in the third line represents the time window constraint condition, t b represents the time when communication satellite b performs the task; the formula in the fourth line represents the mutual exclusion constraint condition. When X bta =1, it indicates that communication satellite b occupies resource a at time t. At this time, resource a cannot be occupied by other communication satellites. When X bta =0, it indicates that communication satellite b does not occupy resource a at time t. At this time, resource a can be occupied by other communication satellites, that is, the same resource can only be occupied by a single communication satellite at the same moment.

[0083] It is understandable that the application scenario refers to the scenario in which the communication satellite system is applied at the current moment. The application scenario of the communication satellite system can be a satellite communication network scenario or an Earth observation scenario. For different application scenarios, the determined resource allocation optimization function and multi-dimensional constraint conditions are different. The resource allocation optimization function can be a resource allocation cost minimization function, and the goal is to minimize the total cost of resource allocation through the above resource allocation optimization function.

[0084] It should be understood that the multi-dimensional constraint conditions include but are not limited to capacity dimension constraint conditions, required resource dimension constraint conditions, time window constraint conditions, and mutual exclusion constraint conditions, and each dimension of the constraint conditions corresponds to a formula. After determining the resource allocation optimization function and multi-dimensional constraint conditions, a big data optimization algorithm is established.

[0085] Furthermore, the steps of optimizing the resource allocation basic algorithm through the big data optimization algorithm to obtain the target resource allocation algorithm include: determining the optimal solution of resource allocation through the big data optimization algorithm; optimizing each allocation parameter in the resource allocation basic algorithm according to the optimal solution of resource allocation; and generating the target resource allocation algorithm according to the optimized allocation parameters.

[0086] It is understandable that the optimal solution of resource allocation refers to the optimal solution that can minimize the resource allocation cost and meet the requirements necessary for each communication satellite to perform tasks. After determining the optimal solution of resource allocation, each allocation parameter in the resource allocation basic algorithm is optimized according to the optimal solution of resource allocation, so that the optimized target resource allocation algorithm, that is, the target resource allocation algorithm, is the optimal algorithm for resource allocation in the current scenario. In addition, in order to meet the optimal resource allocation method, in additional cases, the allocation strategy corresponding to the resource allocation basic algorithm can also be adaptively adjusted.

[0087] Step S40: Construct a target AI resource allocation large model according to the target resource allocation algorithm, and based on the target AI resource allocation large model, dynamically allocate orbital resources to each communication satellite according to the position information and the task information.

[0088] It is understandable that after determining the target AI resource allocation large model, based on the target AI resource allocation large model, predict the best orbital resources of each communication satellite at different times according to the position information and the task information, and allocate the best orbital resources to each communication satellite to achieve the best allocation of orbital resources and ensure the efficient, reliable and flexible operation of the satellite network.

[0089] Further, after step S40, the method further includes: obtaining frequency interference data and communication quality data generated by each satellite at the next moment; determining a current frequency interference change value and a current communication quality change value according to the frequency interference data and communication quality data generated by each satellite at the next moment and the frequency interference data and communication quality data generated by each satellite at the current moment; determining a frequency interference weight and a communication quality weight; calculating a frequency interference reference value according to the current frequency interference change value and the frequency interference weight, and calculating a communication quality reference value according to the current communication quality change value and the communication quality weight; when the frequency interference reference value is less than a first threshold value and the communication quality reference value is greater than a second threshold value, dynamically allocating orbital resources to other communication satellite systems.

[0090] It should be understood that the current frequency interference change value refers to the change value of the frequency interference data generated by each satellite from the current moment to the next moment. When the current frequency interference change value is negative, it indicates that the frequency interference decreases. The current communication quality change value refers to the change value of the communication quality from the current moment to the next moment. When the current communication quality change value is positive, it indicates that the communication quality is enhanced. At this time, the respective reference values, that is, the frequency interference reference value and the communication quality reference value, are calculated in combination with the frequency interference weight and the communication quality weight. When at least one of the conditions that the frequency interference reference value is greater than or equal to the first threshold value and the communication quality reference value is less than or equal to the second threshold value is satisfied, the orbital resource optimization algorithm is optimized to obtain the optimal orbital resource optimization algorithm. On the contrary, when the above conditions are not satisfied, the above method is used to dynamically allocate orbital resources to other communication satellite systems.

[0091] In this embodiment, when receiving an orbital resource allocation request, the position information and task information of each communication satellite in the communication satellite system are obtained; a basic resource allocation algorithm is established according to the orbital resource allocation request; a big data optimization algorithm is established, and the basic resource allocation algorithm is optimized by the big data optimization algorithm to obtain a target resource allocation algorithm; a target AI resource allocation large model is constructed according to the target resource allocation algorithm, and based on the target AI resource allocation large model, the orbital resources are dynamically allocated to each communication satellite according to the position information and the task information. By the above method, after establishing the big data optimization algorithm, the basic resource allocation algorithm is optimized by the big data optimization algorithm. At this time, the optimal target resource allocation algorithm is used to establish a target AI resource allocation large model for dynamically allocating orbital resources, and then the orbital resources are dynamically allocated to each communication satellite in combination with the position information and task information of each communication satellite, so as to effectively improve the accuracy of allocating orbital resources, realize the dynamic allocation of orbital resources, and provide a system solution.

[0092] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar content as that in the above-mentioned first embodiment can be referred to the above introduction and will not be repeated hereinafter. On this basis, please refer to Figure 2 , step S40 includes steps S401 to S404:

[0093] Step S401, obtain a multi-dimensional learning framework in the field of communication satellites.

[0094] It should be understood that the multi-dimensional learning framework refers to a learning framework for training an AI large model. This multi-dimensional learning framework can be determined according to the current application scenario. This multi-dimensional learning framework can be a framework that combines the TensorFlow deep learning framework and the PyTorch deep learning framework, that is, combines the advantages of the TensorFlow deep learning framework and the PyTorch deep learning framework. Compared with a single-dimensional learning framework, the accuracy of resource allocation by the target AI resource allocation large model trained using the multi-dimensional learning framework is higher.

[0095] Step S402, obtain the historical position information and historical task information of each communication satellite in the communication satellite system.

[0096] It should be understood that the historical position information refers to the position information of each communication satellite in the communication satellite system before the current moment. When the current moment is represented as t, the moment before the current moment can be represented as t - 1. The historical task information characterizes the tasks that the communication satellite needs to execute at the historical position.

[0097] Step S403, based on the target resource allocation algorithm, determine the target historical orbit resources corresponding to the historical position information and the historical task information.

[0098] It can be understood that the target historical orbit resources refer to the optimal orbit resources that should be allocated when the communication satellite executes historical tasks at the historical position. The target historical orbit resources are analyzed by the target resource allocation algorithm according to the historical position information and the historical task information.

[0099] Further, after step S403, it further includes: dynamically allocate the historical orbit resources to each communication satellite in the communication satellite system; perform performance detection on each communication satellite in the communication satellite system to obtain communication performance characteristic values and coverage performance characteristic values; calculate the target performance characteristic value according to the communication performance characteristic value, communication weight, the coverage performance characteristic value, and the coverage weight. Specifically:

[0100] Total xn =tx xn *Q1+fg xn *Q2;

[0101] Among them, Total xn represents the target performance characteristic value, tx xn represents the communication performance characteristic value, Q1 represents the communication weight, fg xn represents the coverage performance characteristic value, Q2 represents the coverage weight; when the target performance characteristic value is greater than a preset threshold, based on the target resource allocation algorithm, the target historical orbit resources corresponding to the historical location information and the historical task information are determined.

[0102] It should be understood that before determining the target historical orbit resources using the target resource allocation algorithm, it is necessary to determine whether the performance of each communication satellite in the communication satellite system meets the requirements, that is, to determine whether the target performance characteristic value is greater than the preset threshold. Among them, the target performance characteristic value can be calculated based on the communication performance characteristic value, the communication weight, the coverage performance characteristic value, and the coverage weight. For example, if the communication performance characteristic value is 8.1, the coverage performance characteristic value is 9.6, the set communication weight is 0.6, and the coverage weight is 0.4, then the target performance characteristic value calculated based on the above formula is 8.1 * 0.6 + 9.6 * 0.4 = 5.784. If the set preset threshold is 0.5, it indicates that the target performance characteristic value is greater than the preset threshold. At this time, the target historical orbit resources corresponding to the historical location information and the historical task information can be analyzed and determined based on the target resource allocation algorithm.

[0103] Step S404: Based on the target historical orbit resources, train the multi-dimensional learning framework to obtain a target AI resource allocation large model.

[0104] Furthermore, step S204 includes: generating an initial training sample according to the target historical orbit resources, historical location information, and historical task information; performing data cleaning on the initial training sample, and performing normalization processing on the cleaned initial training sample; rotating the normalized initial training sample based on a target interpolation strategy according to a preset rotation angle, and flipping the rotated initial training sample based on a target flip axis; performing standardization processing on the flipped initial training sample to obtain a target training sample; where the target training sample includes a model training sample, a model verification sample, and a model test sample; training the multi-dimensional learning framework based on the model training sample to obtain an initial AI resource allocation large model; calculating the loss value between the predicted value and the true value of the initial AI resource allocation large model by using the cross-entropy loss function and the MLP head, specifically:

[0105]

[0106] Among them, loss represents the loss value between the predicted value and the true value of the initial AI resource allocation large model, P represents the number of samples, C represents the total number of categories for resource allocation, y i,c represents the true value, and p i,c represents the predicted value; the initial AI resource allocation large model is verified and trained based on the model verification samples until the loss value converges to a preset value.

[0107] It can be understood that after obtaining the target historical orbit resources, initial training samples are generated by combining historical position information and historical task information. To ensure the efficiency and accuracy of the training model, it is necessary to clean the initial training samples to delete abnormal information, delete duplicate information, fill in missing information, and modify the information format, etc. To avoid the influence of information at different scales on the training of the target AI resource allocation large model, it is necessary to normalize the cleaned initial training samples to achieve the purpose of unifying the scale, and the algorithm used can be the min-max normalization algorithm.

[0108] It should be noted that to increase the diversity of data samples, alleviate the problem of data imbalance, and help improve the generalization ability of the model, data augmentation is also required, that is, rotation and flipping. After the above series of processes, target training samples are obtained. At this time, the target training samples can be divided into model training samples, model verification samples, and model test samples according to the ratio of 2:1:1. Then, the multi-dimensional learning framework is trained using the model training samples, and the loss value between the probability value predicted by the initial AI resource allocation large model and the true value is calculated using the above formula. For example, the total number of categories C for resource allocation is 1, the number of samples P is 1, the predicted value of the initial AI resource allocation large model is 0.85, and the true value is 0.9. At this time, loss = 0.00443.

[0109] Further, after the step of verifying and training the initial AI resource allocation large model based on the model verification samples until the loss value converges to a preset value, it further includes: testing the initial AI resource allocation large model based on the model test samples to obtain the number of correct resource allocation sample instances, the number of correct non-resource allocation sample instances, the number of incorrect resource allocation sample instances, and the number of incorrect non-resource classification sample instances; calculating the orbit resource allocation accuracy rate according to the number of correct resource allocation sample instances, the number of correct non-resource allocation sample instances, the number of incorrect resource allocation sample instances, and the number of incorrect non-resource classification sample instances, specifically:

[0110]

[0111] Where, O represents the correct rate of orbit resource allocation, TP represents the number of correctly allocated resource sample instances, TN represents the number of correctly allocated non-resource sample instances, FP represents the number of incorrectly allocated resource sample instances, and FN represents the number of incorrectly allocated non-resource sample instances; when the correct rate of orbit resource allocation is greater than or equal to the target threshold, the initial AI resource allocation large model is used as the target AI resource allocation large model; when the correct rate of orbit resource allocation is less than the target threshold, return to the step of validating and training the initial AI resource allocation large model based on the model validation samples until the loss value converges to a preset value.

[0112] It should be understood that after training the initial AI resource allocation large model, it is also necessary to evaluate the initial AI resource allocation large model through the correct rate of orbit resource allocation, that is, calculate the correct rate of orbit resource allocation using the number of correctly allocated resource sample instances, the number of correctly allocated non-resource sample instances, the number of incorrectly allocated resource sample instances, and the number of incorrectly classified non-resource sample instances. For example, if the number of correctly allocated resource sample instances is 20, the number of correctly allocated non-resource sample instances is 3, the number of incorrectly allocated resource sample instances is 2, and the number of incorrectly classified non-resource sample instances is 2, then the correct rate of orbit resource allocation = (20 + 3) / (20 + 3 + 2 + 2) = 85.19%. When the target threshold is 85%, it indicates that the correct rate of orbit resource allocation is greater than the target threshold. At this time, the initial AI resource allocation large model is used as the target AI resource allocation large model. Conversely, if it is less than the target threshold, the initial AI resource allocation large model needs to be iteratively trained, that is, validated and trained based on the model validation samples until the loss value converges to a preset value, and then the correct rate of orbit resource allocation is calculated again until the correct rate of orbit resource allocation is greater than or equal to the target threshold.

[0113] In this embodiment, a multi-dimensional learning framework in the field of communication satellites is obtained; historical position information and historical task information of each communication satellite in the communication satellite system are obtained; based on the target resource allocation algorithm, the target historical orbit resources corresponding to the historical position information and the historical task information are determined; based on the target historical orbit resources, the multi-dimensional learning framework is trained to obtain a target AI resource allocation large model. Through the above method, after obtaining the historical position information and historical task information of each communication satellite in the communication satellite system, the best orbit resources that should be allocated when the communication satellite performs historical tasks at the historical position are analyzed and determined based on the target resource allocation algorithm, that is, the target historical orbit resources, and then the target AI resource allocation large model is trained in combination with the multi-dimensional learning framework in the field of communication satellites, so as to effectively improve the accuracy of training the target AI resource allocation large model, and further improve the accuracy of allocating orbit resources.

[0114] This application also provides a satellite orbit resource allocation system based on an AI large model. Please refer toFigure 3 , the satellite orbit resource allocation system based on the AI large model includes:

[0115] An acquisition module 10, configured to acquire the position information and task information of each communication satellite in the communication satellite system when receiving an orbit resource allocation request.

[0116] A building module 20, configured to establish a basic resource allocation algorithm according to the orbit resource allocation request.

[0117] An optimization module 30, configured to establish a big data optimization algorithm, and optimize the basic resource allocation algorithm through the big data optimization algorithm to obtain a target resource allocation algorithm.

[0118] An allocation module 40, configured to construct a target AI resource allocation large model according to the target resource allocation algorithm, and based on the target AI resource allocation large model, dynamically allocate orbit resources for each communication satellite according to the position information and the task information.

[0119] In this embodiment, when receiving an orbit resource allocation request, the position information and task information of each communication satellite in the communication satellite system are acquired; a basic resource allocation algorithm is established according to the orbit resource allocation request; a big data optimization algorithm is established, and the basic resource allocation algorithm is optimized through the big data optimization algorithm to obtain a target resource allocation algorithm; a target AI resource allocation large model is constructed according to the target resource allocation algorithm, and based on the target AI resource allocation large model, orbit resources are dynamically allocated for each communication satellite according to the position information and the task information. By the above method, after establishing the big data optimization algorithm, the basic resource allocation algorithm is optimized through the big data optimization algorithm. At this time, the optimal target resource allocation algorithm is used to establish a target AI resource allocation large model for dynamically allocating orbit resources, and then the position information and task information of each communication satellite are combined to dynamically allocate orbit resources for each communication satellite, so as to effectively improve the accuracy of allocating orbit resources, realize the dynamic allocation of orbit resources, and provide a system solution.

[0120] In one embodiment, the optimization module 30 is further configured to acquire the application scenario of the communication satellite system, and determine a resource allocation optimization function and multi-dimensional constraint conditions according to the application scenario, where the resource allocation optimization function is:

[0121]

[0122] where D represents the minimum total cost of resource allocation, N represents the number of resources, M represents the number of tasks, c ab represents the cost of allocating resource a to communication satellite b, and x ab represents the decision variable that minimizes the single cost;

[0123] Among them, the multi-dimensional constraint conditions are as follows:

[0124]

[0125] Among them, the formula in the first line indicates that the usage amount of resource a cannot exceed the capacity A a of the constraint condition, a ab represents the resource a allocated to communication satellite b. Since the cost c ab of allocating resource a to communication satellite b is relatively high, if it exceeds the capacity A a , it will cause the total cost D of resource allocation to be extremely high and other communication satellites cannot be allocated sufficient resources; the formula in the second line indicates the resources that communication satellite must satisfy when performing different tasks, b ab represents the contribution amount of resource a to communication satellite b, D b represents the demand of communication satellite b. The ultimate optimization goal is to make the contribution amount of resource a to communication satellite b at least meet the demand of the communication satellite and minimize the total cost D of resource allocation; the formula in the third line represents the time window constraint condition, t b represents the time when communication satellite b performs the task; the formula in the fourth line represents the mutual exclusion constraint condition. When X bta = 1, it indicates that communication satellite b occupies resource a at time t, and at this time resource a cannot be occupied by other communication satellites. When X bta = 0, it indicates that communication satellite b does not occupy resource a at time t, and at this time resource a can be occupied by other communication satellites, that is, the same resource can only be occupied by a single communication satellite at the same moment;

[0126] Establish a big data optimization algorithm according to the resource allocation optimization function and the multi-dimensional constraint conditions.

[0127] In one embodiment, the optimization module 30 is further configured to determine the optimal solution of resource allocation through the big data optimization algorithm;

[0128] Optimize each allocation parameter in the resource allocation basic algorithm according to the optimal solution of resource allocation;

[0129] Generate a target resource allocation algorithm according to the optimized allocation parameters.

[0130] In one embodiment, the allocation module 40 is further configured to obtain a multi-dimensional learning framework in the field of communication satellites;

[0131] Obtain the historical position information and historical task information of each communication satellite in the communication satellite system;

[0132] Based on the target resource allocation algorithm, determine the target historical orbit resources corresponding to the historical location information and the historical task information;

[0133] Based on the target historical orbit resources, train the multi-dimensional learning framework to obtain a target AI resource allocation large model.

[0134] In one embodiment, after the step of determining, by the allocation module 40, the target historical orbit resources corresponding to the historical location information and the historical task information based on the target resource allocation algorithm, the method further includes:

[0135] Dynamically allocate the historical orbit resources to each communication satellite in the communication satellite system;

[0136] Perform performance detection on each communication satellite in the communication satellite system to obtain communication performance characteristic values and coverage performance characteristic values;

[0137] Calculate the target performance characteristic value according to the communication performance characteristic value, the communication weight, the coverage performance characteristic value, and the coverage weight, specifically:

[0138] Total xn =tx xn *Q1+fg xn *Q2;

[0139] wherein, Total xn represents the target performance characteristic value, tx xn represents the communication performance characteristic value, Q1 represents the communication weight, fg xn represents the coverage performance characteristic value, and Q2 represents the coverage weight;

[0140] When the target performance characteristic value is greater than a preset threshold, based on the target resource allocation algorithm, determine the target historical orbit resources corresponding to the historical location information and the historical task information.

[0141] In one embodiment, the step of training, by the allocation module 40, the multi-dimensional learning framework based on the target historical orbit resources to obtain a target AI resource allocation large model includes:

[0142] Generate initial training samples according to the target historical orbit resources, the historical location information, and the historical task information;

[0143] Perform data cleaning on the initial training samples, and perform normalization processing on the initial training samples after cleaning;

[0144] Rotate the normalized initial training samples according to a preset rotation angle based on a target interpolation strategy, and flip the rotated initial training samples based on a target axis of flipping;

[0145] Perform normalization processing on the flipped initial training samples to obtain target training samples; wherein, the target training samples include model training samples, model validation samples, and model test samples;

[0146] Train a multi-dimensional learning framework based on the model training samples to obtain an initial AI resource allocation large model;

[0147] Calculate the loss value between the predicted value and the true value of the initial AI resource allocation large model by using an MLP head with a cross-entropy loss function, specifically:

[0148]

[0149] where loss represents the loss value between the predicted value and the true value of the initial AI resource allocation large model, P represents the number of samples, C represents the total number of categories for resource allocation, y i,c represents the true value, and p i,c represents the predicted value;

[0150] Perform validation training on the initial AI resource allocation large model based on the model validation samples until the loss value converges to a preset value.

[0151] In one embodiment, after the step of performing validation training on the initial AI resource allocation large model based on the model validation samples until the loss value converges to a preset value, the allocation module 40 is further configured to:

[0152] Test the initial AI resource allocation large model based on model test samples to obtain the number of correctly allocated resource sample instances, the number of correctly allocated non-resource sample instances, the number of incorrectly allocated resource sample instances, and the number of incorrectly classified non-resource sample instances;

[0153] Calculate the orbital resource allocation accuracy rate according to the number of correctly allocated resource sample instances, the number of correctly allocated non-resource sample instances, the number of incorrectly allocated resource sample instances, and the number of incorrectly classified non-resource sample instances, specifically:

[0154]

[0155] where O represents the orbital resource allocation accuracy rate, TP represents the number of correctly allocated resource sample instances, TN represents the number of correctly allocated non-resource sample instances, FP represents the number of incorrectly allocated resource sample instances, and FN represents the number of incorrectly allocated non-resource sample instances;

[0156] When the correct rate of the orbital resource allocation is greater than or equal to the target threshold, the initial AI resource allocation large model is used as the target AI resource allocation large model.

[0157] In one embodiment, the allocation module 40 is further configured to obtain frequency interference data and communication quality data generated by each satellite at the next moment.

[0158] Determine the current frequency interference change value and the current communication quality change value according to the frequency interference data and communication quality data generated by each satellite at the next moment, and the frequency interference data and communication quality data generated by each satellite at the current moment.

[0159] Determine the frequency interference weight and the communication quality weight; calculate the frequency interference reference value according to the current frequency interference change value and the frequency interference weight, and calculate the communication quality reference value according to the current communication quality change value and the communication quality weight.

[0160] When the frequency interference reference value is less than the first threshold value and the communication quality reference value is greater than the second threshold value, perform dynamic allocation of orbital resources to other communication satellite systems.

[0161] The satellite orbital resource allocation system based on the AI large model provided by the present application adopts the satellite orbital resource allocation method based on the AI large model in the above embodiment, and can solve the technical problem of low accuracy of allocating orbital resources in the prior art. Compared with the prior art, the beneficial effects of the satellite orbital resource allocation system based on the AI large model provided by the present application are the same as those of the satellite orbital resource allocation method based on the AI large model provided by the above embodiment, and other technical features in the satellite orbital resource allocation system based on the AI large model are the same as the features disclosed in the above embodiment method, and will not be described in detail here.

[0162] It should be understood that each part disclosed in the present application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0163] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0164] The above are only some embodiments of the present application, and thus do not limit the patent scope of the present application. Any equivalent structural transformation made under the technical concept of the present application by using the content of the specification and drawings of the present application, or any direct / indirect application in other related technical fields is included in the patent protection scope of the present application.

Claims

1. A satellite orbit resource allocation method and system based on AI large model, characterized in that: The method comprises: Upon receiving an orbital resource allocation request, obtaining position information and mission information of each communication satellite in the communication satellite system; Establishing a basic resource allocation algorithm according to the track resource allocation request; Establishing a big data optimization algorithm, and optimizing the basic resource allocation algorithm through the big data optimization algorithm to obtain a target resource allocation algorithm; A target AI resource allocation model is constructed according to the target resource allocation algorithm, and based on the target AI resource allocation model, orbital resources are dynamically allocated to each communication satellite according to the location information and the mission information.

2. The method according to claim 1, characterized in that The step of constructing a target AI resource allocation model according to the target resource allocation algorithm includes: Acquire a multi-dimensional learning framework in the field of communication satellites; Obtain historical position information and historical mission information of each communication satellite in the communication satellite system; Based on the target resource allocation algorithm, determining the target historical track resource corresponding to the historical position information and the historical mission information; Based on the target historical orbital resources, the multi-dimensional learning framework is trained to obtain a target AI resource allocation model.

3. The method according to claim 2, characterized in that After the step of determining the target historical track resource corresponding to the historical position information and the historical mission information based on the target resource allocation algorithm, the method further includes: Dynamically allocating the historical orbital resources to each communication satellite in the communication satellite system; Performing performance testing on each communication satellite in the communication satellite system to obtain a communication performance characteristic value and a coverage performance characteristic value; The target performance characteristic value is calculated according to the communication performance characteristic value, the communication weight, the coverage performance characteristic value and the coverage weight, specifically: Total xn =tx xn *Q1+fg xn *Q2; Among them, Total xn Indicates the target performance characteristic value, tx xn represents the communication performance characteristic value, Q1 represents the communication weight, fg xn represents the coverage performance characteristic value, Q2 represents the coverage weight; When the target performance characteristic value is greater than a preset threshold, a target historical orbit resource corresponding to the historical position information and the historical mission information is determined based on the target resource allocation algorithm.

4. The method according to claim 2, characterized in that The step of training the multi-dimensional learning framework based on the target historical orbital resources to obtain a target AI resource allocation model includes: Generate initial training samples according to the target historical orbit resources, historical position information and historical mission information; Performing data cleaning on the initial training samples, and performing normalization processing on the cleaned initial training samples; Based on the target interpolation strategy, the normalized initial training samples are rotated according to a preset rotation angle, and the rotated initial training samples are flipped based on the target flip axis; The flipped initial training sample is standardized to obtain a target training sample; wherein the target training sample includes a model training sample, a model verification sample, and a model test sample; Based on the model training samples, the multi-dimensional learning framework is trained to obtain an initial AI resource allocation model; The loss value between the predicted value and the true value of the initial AI resource allocation model is calculated using the MLP head by using the cross entropy loss function, specifically: Among them, loss represents the loss value between the predicted value and the true value of the initial AI resource allocation model, P represents the number of samples, C represents the total number of categories for resource allocation, and y i,c represents the true value, p i,c represents the predicted value; The initial AI resource allocation model is verified and trained based on the model verification sample until the loss value converges to a preset value.

5. The method according to claim 4, characterized in that After the step of performing verification training on the initial AI resource allocation large model based on the model verification sample until the loss value converges to a preset value, the step further includes: Based on the model test samples, the initial AI resource allocation model is tested to obtain the number of correctly allocated resource sample instances, the number of correctly allocated non-resource sample instances, the number of incorrectly allocated resource sample instances, and the number of incorrectly classified non-resource sample instances; The track resource allocation accuracy is calculated according to the number of correctly allocated resource sample instances, the number of correctly allocated non-resource sample instances, the number of incorrectly allocated resource sample instances, and the number of incorrectly classified non-resource sample instances, specifically: Among them, O represents the correct rate of track resource allocation, TP represents the number of correctly allocated resource sample instances, TN represents the number of correctly allocated non-resource sample instances, FP represents the number of incorrectly allocated resource sample instances, and FN represents the number of incorrectly allocated non-resource sample instances; When the track resource allocation accuracy is greater than or equal to the target threshold, the initial AI resource allocation model is used as the target AI resource allocation model; When the track resource allocation accuracy is less than the target threshold, return to the step of performing verification training on the initial AI resource allocation large model based on the model verification sample until the loss value converges to a preset value.

6. The method according to claim 1, characterized in that The steps of establishing a big data optimization algorithm include: Obtain an application scenario of the communication satellite system, and determine a resource allocation optimization function and multi-dimensional constraints according to the application scenario, wherein the resource allocation optimization function is: Where D represents the resource allocation minimizing the total cost, N represents the number of resources, M represents the number of tasks, and c ab represents the cost of allocating resource a to communication satellite b, x ab represents the decision variable that minimizes the single cost; Among them, the multi-dimensional constraints are: The formula in the first line above indicates that the usage of resource a cannot exceed capacity A. a The constraint condition, a ab represents the resource a allocated to communication satellite b. The cost c of allocating resource a to communication satellite b is ab Higher, if it exceeds capacity A a , the total cost D of resource allocation will be extremely high and other communication satellites will not be able to allocate enough resources; the second line of formulas represents the resources that must be met when communication satellites perform different tasks, b ab represents the contribution of resource a to communication satellite b, D b represents the demand for communication satellite b. The ultimate optimization goal is to make the contribution of resource a to communication satellite b at least meet the demand of communication satellite and the total resource allocation cost D is the lowest. The third line of formula represents the time window constraint, t b represents the time for communication satellite b to perform its mission; the fourth line of formula represents the mutually exclusive constraint condition, in X bta = 1, indicating that communication satellite b occupies resource a at time t. At this time, resource a cannot be occupied by other communication satellites. bta =0, indicating that communication satellite b does not occupy resource a at time t, and resource a can be occupied by other communication satellites at this time, that is, the same resource can only be occupied by a unique communication satellite at the same time; A big data optimization algorithm is established according to the resource allocation optimization function and the multi-dimensional constraint conditions.

7. The method according to claim 1, characterized in that The step of optimizing the basic resource allocation algorithm by the big data optimization algorithm to obtain a target resource allocation algorithm includes: Determine the optimal solution for resource allocation through big data optimization algorithms; Optimizing each allocation parameter in the basic resource allocation algorithm according to the optimal solution for resource allocation; A target resource allocation algorithm is generated according to the optimized allocation parameters.

8. The method according to claim 1, characterized in that After the step of dynamically allocating orbital resources for each communication satellite according to the position information and the mission information, the method further includes: Obtain frequency interference data and communication quality data generated by each satellite at the next moment; Determine a current frequency interference change value and a current communication quality change value according to the frequency interference data and communication quality data generated by each satellite at the next moment and the frequency interference data and communication quality data generated by each satellite at the current moment; Determine a frequency interference weight and a communication quality weight; calculate a frequency interference reference value according to the current frequency interference change value and the frequency interference weight, and calculate a communication quality reference value according to the current communication quality change value and the communication quality weight; When the frequency interference reference value is less than a first threshold value and the communication quality reference value is greater than a second threshold value, orbital resources are dynamically allocated to other communication satellite systems.

9. A satellite orbit resource allocation system based on AI big model, characterized in that: The system comprises: An acquisition module, used to acquire the position information and mission information of each communication satellite in the communication satellite system upon receiving an orbital resource allocation request; An establishing module, used for establishing a basic resource allocation algorithm according to the track resource allocation request; An optimization module, used to establish a big data optimization algorithm, and optimize the basic resource allocation algorithm through the big data optimization algorithm to obtain a target resource allocation algorithm; An allocation module is used to build a target AI resource allocation model according to the target resource allocation algorithm, and based on the target AI resource allocation model, dynamically allocate orbital resources to each communication satellite according to the location information and the mission information.

Images