AI-based aerospace electromagnetic environment big data processing system

Abstract

The invention relates to the technical field of electromagnetic spectrum management, and discloses an AI-based aerospace electromagnetic environment big data processing system, which comprises an electromagnetic information collection module used for collecting electromagnetic information; the first order information construction module is used for constructing first order information of two structural dimensions based on the electromagnetic information; the link quality judgment module inputs the first order information into a quality model, and the quality model outputs the quality result of each link; the spectrum allocation module inputs the first order information into an allocation model, and the allocation model outputs a spectrum allocation result of each communication node; according to the invention, intelligent dynamic allocation of spectrum resources in the space-air-ground communication network is realized by analyzing electromagnetic information, atmospheric environment information is further considered, and compared with only allocation of spectrum resources, the system provides a spectrum allocation scheme with high allocation quality, low cost and high utilization rate under the condition of considering atmospheric influence.

Description

Technical Field

[0001] The present invention relates to the technical field of electromagnetic spectrum management, and more specifically, it relates to an AI-based big data processing system for the aerospace electromagnetic environment. Background Art

[0002] Low-Earth orbit satellite constellations, ground communication stations, and aerial platforms constitute a complex three-dimensional communication system. However, with the increasing complexity of the aerospace electromagnetic environment and the growing tension of electromagnetic spectrum resources, how to achieve intelligent and dynamic allocation of electromagnetic spectrum resources has become the focus and difficulty of current research. By constructing a hierarchical decision-making model and a reward mechanism, efficient allocation and dynamic adjustment of spectrum resources have been achieved. However, atmospheric refraction and scattering cause drastic fluctuations in link quality, affecting the decision-making accuracy of spectrum resource allocation, resulting in frequent adjustments of allocation schemes and a decline in system performance. Summary of the Invention

[0003] The purpose of the present invention is to provide an AI-based big data processing system for the aerospace electromagnetic environment to solve the above problems.

[0004] The present invention provides an AI-based big data processing system for the aerospace electromagnetic environment, including: An electromagnetic information collection module, which is used to collect electromagnetic information, and the electromagnetic information includes communication node information and atmospheric environment information; the communication nodes include satellites, ground stations, and aerial platforms; A first-order orderliness information construction module, which constructs first-order orderliness information in two structural dimensions based on the electromagnetic information. In the first structural dimension, objects are used as basic structural elements, and the objects include: satellites, ground stations, aerial platforms, and atmospheric environment; in the second structural dimension, first-order time series units are used as basic structural elements; A link quality determination module, which inputs the first-order orderliness information into a quality model, and the quality model outputs the results of the quality of each link; A spectrum allocation module, which inputs the first-order orderliness information into an allocation model, and the allocation model outputs the results of spectrum allocation for each communication node.

[0005] Further, the method for processing to obtain the first-order orderliness information includes: Extracting the objects of the first-order orderliness information; Extracting specific features associated with the objects, and dividing the extracted specific features into multiple parts according to the data acquisition period. One part is mapped to one first-order time series unit, and the order of the first-order time series units is consistent with the time order of the acquisition period of the specific features; Establishing a mapping between the extracted objects and their associated specific features; For any two objects, an object connection is established between the two objects if there is a relevance related to the electromagnetic spectrum management task between the associated specific features or there is a relevance related to the electromagnetic spectrum management task between the objects.

[0006] Furthermore, the specific features associated with the satellite: the position of the satellite, the currently allocated frequency, the allocated bandwidth, the transmission power; The specific features associated with the ground station: the position of the ground station, the currently allocated frequency, the allocated bandwidth, the transmission power; The specific features associated with the airborne platform: the position of the airborne platform, the currently allocated frequency, the allocated bandwidth, the transmission power; The specific features associated with the atmospheric environment: the atmospheric range, the refractive index, the rainfall rate, the atmospheric pressure, the temperature distribution, the water vapor pressure.

[0007] Furthermore, the relevance related to the risk detection task includes: There is a frequency overlap between two communication nodes and they are within their respective communication coverage ranges; If there is a relevance between two communication nodes and the signal propagation path passes through the corresponding atmospheric environment, then the two communication nodes are respectively associated with the atmospheric environment; The signal propagation path is the spatial trajectory of the electromagnetic wave propagation between two communication nodes.

[0008] Furthermore, the quality model includes a first backbone network, and the first backbone network includes a first time series layer, a second graph network layer, and a first output layer. The first time series layer inputs the first-ordered information and outputs the first hidden feature to the second graph network layer. The second graph network layer outputs the second hidden feature to the first output layer, and the first output layer outputs the results of the quality of each link.

[0009] Furthermore, a link indicates that there is an association between two communication nodes.

[0010] Furthermore, the allocation model includes a second backbone network, and the second backbone network includes a third time series layer, a fourth time series layer, a fifth graph network layer, a feature fusion layer, and a second output layer. The third time series layer inputs the first-ordered information and outputs the third hidden feature to the feature fusion layer. The fourth time series layer inputs the second-ordered information and outputs the fourth hidden feature to the feature fusion layer. The feature fusion layer outputs the fused feature to the fifth graph network layer. The fifth graph network layer outputs the fifth hidden feature to the second output layer, and the second output layer outputs the results of the spectrum allocation of each communication node.

[0011] Furthermore, based on the link quality data output by the quality model, a second structured data including one structural dimension is constructed. The basic structural elements included in this structural dimension are second-order time series units. The structuring method includes: Extract the association information of the link quality determination task, and divide the extracted specific features into multiple parts according to the data collection period. One part is mapped to a second time series unit, and the order of the second time series units is consistent with the time order of the collection period of the association information of the link quality determination task.

[0012] Furthermore, train the link quality determination module through supervised learning. The calculation method of the loss function of the link quality determination module is: ; Wherein, represents the loss value of the link quality determination module, represents the total number of link samples, represents the th true quality value of the link sample, represents the quality value of the th link sample output by the link quality determination module; Train the spectrum allocation module through Q-learning. The calculation method of the immediate reward is: ; Wherein, represents the total number of allocation samples, and represent the quality values of the th and th allocation samples output by the link quality determination module for the th time, represents the switching cost when executing , represents the spectrum utilization rate when executing , , , are the first, second, and third hyperparameters; The calculation method of the spectrum utilization rate is: ; Wherein, represents the number of communication nodes allocated under the policy , represents the bandwidth allocated to the th communication node, represents the transmit power allocated to the th communication node, represents the total available bandwidth, represents the maximum available transmit power; The calculation method of the switching cost is: ; Among them, represents the frequency assigned to the th communication node under the policy , represents the frequency assigned to the th communication node at the previous moment , represents the total available frequency range, represents the total number of communication nodes.

[0013] The present invention provides a computer storage medium for storing computer-readable instructions, which can execute the aforementioned AI-based space-air electromagnetic environment big data processing system when the computer-readable instructions are read.

[0014] The beneficial effects of the present invention are as follows: By analyzing electromagnetic information, the present invention realizes the intelligent dynamic allocation of spectrum resources in the space-air-ground communication network, and further considers the atmospheric environment information. Compared with only allocating spectrum resources, the system provides a spectrum allocation scheme with high allocation quality, low cost, and high utilization rate under the consideration of atmospheric influence. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is the system module diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described relative to some examples can also be combined in other examples.

[0017] In at least one embodiment of the present invention, an AI-based space-air electromagnetic environment big data processing system is disclosed, as Figure 1 shown, including: An electromagnetic information collection module 101 for collecting electromagnetic information, where the electromagnetic information includes communication node information and atmospheric environment information; In one embodiment of the present invention, the communication nodes include satellites, ground stations, and aerial platforms; A first-order order information construction module 102 for constructing first-order order information in two structural dimensions based on the electromagnetic information. In the first structural dimension, the object is used as the basic structural element, and the objects include: satellites, ground stations, aerial platforms, and atmospheric environment; in the second structural dimension, the first-order time sequence unit is used as the basic structural element; In some embodiments of the present invention, the method for processing to obtain the first-order order information includes: Extracting the object of the first-order order information; Extracting the specific features associated with the object, and dividing the extracted specific features into multiple parts according to the data collection period. One part is mapped to one first-order time sequence unit, and the order of the first-order time sequence units is consistent with the time sequence of the collection period of the specific features; Establishing a mapping between the extracted object and the specific features associated therewith; For any two objects, if there is a relevance related to the electromagnetic spectrum management task between the associated specific features or there is a relevance related to the electromagnetic spectrum management task between the objects, an object connection is constructed between the two objects.

[0018] In some embodiments of the present invention, the specific features associated with the satellite: the position of the satellite, the currently allocated frequency, the allocated bandwidth, the transmit power; The specific features associated with the ground station: the position of the ground station, the currently allocated frequency, the allocated bandwidth, the transmit power; The specific features associated with the airborne platform: the position of the airborne platform, the currently allocated frequency, the allocated bandwidth, the transmit power; The specific features associated with the atmospheric environment: the atmospheric range, the refractive index, the rainfall rate, the atmospheric pressure, the temperature distribution, the water vapor pressure; In some embodiments of the present invention, the relevance related to the risk detection task includes: There is a frequency overlap between two communication nodes and they are within their respective communication coverage ranges; There is a relevance between two communication nodes, and their signal propagation paths pass through the corresponding atmospheric environment, then the two communication nodes are respectively associated with the atmospheric environment; In one embodiment of the present invention, the signal propagation path is the spatial trajectory of the electromagnetic wave propagation between two communication nodes.

[0019] In one embodiment of the present invention, the refractive index is calculated through the refractive index model:

[0020] where The refractive index at the spatial position and time , , represent the spatial coordinates on the horizontal plane, is the height, is the time, is the rainfall rate (mm / h), is the rainfall influence coefficient, is the height attenuation coefficient, is the surface air pressure is the pressure scale height is the surface temperature is the lapse rate, is the surface water vapor pressure is the water vapor pressure scale height, represents the variation of the rainfall rate with height in P 0 is the surface air pressure.

[0021] The link quality determination module 103 inputs the first-order order information into the quality model. The quality model includes a first main network, and the first main network includes a first time series layer, a second graph network layer, and a first output layer. Among them, the first time series layer inputs the first-order order information and outputs the first hidden feature to the second graph network layer. The second graph network layer outputs the second hidden feature to the first output layer, and the first output layer outputs the results of the link quality of each link; In an embodiment of the present invention, a link indicates that there is an association between two communication nodes; The spectrum allocation module 104 inputs the first-order order information into the allocation model. The allocation model includes a second main network, and the second main network includes a third time series layer, a fourth time series layer, a fifth graph network layer, a feature fusion layer, and a second output layer. Among them, the third time series layer inputs the first-order order information and outputs the third hidden feature to the feature fusion layer. The fourth time series layer inputs the second-order order information and outputs the fourth hidden feature to the feature fusion layer. The feature fusion layer outputs the fusion feature to the fifth graph network layer. The fifth graph network layer outputs the fifth hidden feature to the second output layer, and the second output layer outputs the results of the spectrum allocation of each communication node; In an embodiment of the present invention, based on the link quality data output by the quality model, a second structured data including a structural dimension is constructed. The basic structural elements included in this structural dimension are second-order time series units. The structuring method includes: Extract the association information of the link quality determination task, divide the extracted specific features into multiple parts according to the data collection period, and map one part to one second-order time series unit. The order of the second-order time series units is consistent with the time order of the collection period of the association information of the link quality determination task.

[0022] In an embodiment of the present invention, both the first time series layer and the third time series layer adopt time series neural networks, such as RNN or LSTM or GRU.

[0023] An expression form of the first time series layer, the third time series layer, and the fourth time series layer is as follows:

[0024]

[0025]

[0026]

[0027] Wherein: : The reset gate vector at the -th time step, which is used to control the degree of reset of historical information; : The update gate vector at the -th time step, which is used to control the degree of update of new information; : The candidate hidden state vector at the -th time step; : The hidden state vector at the -th time step, that is, the first, third, or fourth hidden feature (the first time series layer outputs the first hidden feature, the third time series layer outputs the third hidden feature, and the fourth time series layer outputs the fourth hidden feature); : The hidden state vector at the -th time step : The input vector at the -th time step, corresponding to the first or second time series unit (the first and third time series layers are the first time series unit, and the fourth time series layer is the second time series unit); , , : The weight matrices of the reset gate, update gate, and candidate state, which are trainable parameters; , , : The input weight matrices of the reset gate, update gate, and candidate state, which are trainable parameters; , , : The bias vectors of the reset gate, update gate, and candidate state, which are trainable parameters; : The Hadamard product (element-wise multiplication) of vectors; : The Sigmoid activation function; : The hyperbolic tangent activation function; : The time step index, whose value range is , is the sequence length. When holds, As the initial state Both the second graph network layer and the fifth graph network layer adopt a multi-layer structure. The calculation formula for the layer is as follows:

[0028] Where: represents the hidden state vector of the object in the layer; represents the hidden state vector of the object in the layer; represents the set of neighbor nodes directly connected to the object ; represents the set of neighbor nodes directly connected to the object ; represents the set of objects having an object connection with the object ; represents the set of objects having an object connection with the object ; represents the learnable weight matrix of the layer; represents the network layer index, and the value range is , where is the total number of layers. When , , where, for the second graph network layer, , represents the last hidden feature output when the specific features of all the first-time sequence units associated with the th object are input to the first-time sequence layer. is the total number of the first-time sequence units. For the fifth graph network layer, , represents the fused feature output by the feature fusion layer corresponding to the object . When , is the finally output node feature (the second graph network layer outputs the second hidden feature, and the fifth graph network layer outputs the fifth hidden feature, but the trainable parameters of the second graph network layer and the fifth graph network layer are different); The expression of the feature fusion layer is as follows:

[0029] Among them, represents the fusion feature associated with the th object, represents a feature combination function (concatenation function or summation function), represents the th object. When all the specific feature inputs of the first-time sequence units associated with the object enter the first-time sequence layer, the last hidden feature is output, is the total number of the first-time sequence units, represents that when the second-time sequence unit enters the fourth-time sequence layer, the last fourth hidden feature is output, is the total number of the second-time sequence units.

[0030] The expression of the first output layer is as follows:

[0031] Among them represents a link quality vector. The c-th component value of the link quality vector represents the quality value of the c-th link, represents the th layer's th object's second hidden feature, represents the set of all objects, and FC represents a fully connected layer.

[0032] The expression of the second output layer is as follows:

[0033] Among them represents an allocation vector. The d-th component value of the allocation vector represents the probability value of the d-th policy. The policy with the largest selection probability value is selected as the output. The policy group contains all executable policies. A policy represents a spectrum allocation scheme of a communication node, including frequency allocation, bandwidth allocation, and power allocation, represents the th layer's th object's fifth hidden feature, represents the set of all objects, represents the total number of layers, and FC represents a fully connected layer.

[0034] In an embodiment of the present invention, the link quality determination module is trained through supervised learning. The calculation method of the loss function of the link quality determination module is:

[0035] Among them, represents the loss value of the link quality determination module, represents the total number of link samples, represents the true quality value of the th link sample, represents the quality value of the th link sample output by the link quality determination module.

[0036] In an embodiment of the present invention, the spectrum allocation module is trained by Q-learning, where the immediate reward is calculated as follows:

[0037] where, represents the total number of allocation samples, and represent the quality values of the th and th allocation samples output by the link quality determination module for the th link sample, represents the switching cost when executing , represents the spectrum utilization rate when executing , , , are the first, second, and third hyperparameters; In an embodiment of the present invention, the spectrum utilization rate is calculated as:

[0038] where, represents the number of communication nodes allocated under the policy , represents the bandwidth allocated to the th communication node, represents the transmit power allocated to the th communication node, represents the total available bandwidth, represents the maximum available transmit power; The switching cost is calculated as:

[0039] where, represents the frequency allocated to the th communication node under the policy , represents the frequency allocated to the th communication node at the previous moment , represents the total available frequency range, represents the total number of communication nodes.

[0040] In at least one embodiment of the present invention, a computer storage medium is provided for storing computer-readable instructions that, when read, are capable of executing the aforementioned AI-based space electromagnetic environment big data processing system.

[0041] The embodiments of the present invention have been described above, but these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative rather than restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. An AI-based air and space electromagnetic environment big data processing system, characterized in that: include: An electromagnetic information collection module is used to collect electromagnetic information, including communication node information and atmospheric environment information; Communications nodes include satellites, ground stations, and aerial platforms; The No. 1 ordered information construction module constructs No. 1 ordered information in two structural dimensions based on electromagnetic information. In the first structural dimension, objects are used as basic structural elements, including satellites, ground stations, air platforms, and atmospheric environment; in the second structural dimension, No. 1 time sequence units are used as basic structural elements; A link quality determination module, which inputs the first order information into the quality model, and the quality model outputs the result of each link quality; The spectrum allocation module inputs the No. 1 ordered information into the allocation model, and the allocation model outputs the spectrum allocation result of each communication node.

2. The AI-based space electromagnetic environment big data processing system according to claim 1 is characterized in that: The method for processing and obtaining the number one sequential information includes: Extract the object with order information of number one; Extract specific features associated with the object, and divide the extracted specific features into multiple parts according to the data collection cycle, and map each part to a first time sequence unit, and the order of the first time sequence unit is consistent with the time order of the collection cycle of the specific features; Establishing a mapping between the extracted objects and their associated specific features; For any two objects, if there is a correlation between the associated specific features related to the electromagnetic spectrum management task or there is a correlation between the objects related to the electromagnetic spectrum management task, an object relationship is established between the two objects.

3. The AI-based air-space electromagnetic environment big data processing system according to claim 1 is characterized in that: Satellite-associated specific characteristics: satellite location, current allocated frequency, allocated bandwidth, transmit power; Specific characteristics associated with ground sites: location of ground sites, current allocated frequency, allocated bandwidth, and transmit power; Specific characteristics associated with the aerial platform: location of the aerial platform, current allocated frequency, allocated bandwidth, and transmit power; Specific characteristics associated with the atmospheric environment: atmospheric extent, refractive index, rainfall rate, atmospheric pressure, temperature distribution, water vapor pressure.

4. The AI-based space electromagnetic environment big data processing system according to claim 1 is characterized in that: Relevance to the risk detection task includes: There is frequency overlap between the two communication nodes and they are within their respective communication coverage areas; If there is a correlation between two communication nodes and their signal propagation paths pass through the corresponding atmospheric environments, then the two communication nodes are respectively associated with the atmospheric environments; The signal propagation path is the spatial trajectory of electromagnetic wave propagation between two communication nodes.

5. The AI-based air-space electromagnetic environment big data processing system according to claim 1 is characterized in that: The quality model includes a No. 1 backbone network, which includes a first timing layer, a second graph network layer, and a first output layer. The first timing layer inputs No. 1 sequential information and outputs a first hidden feature to the second graph network layer. The second graph network layer outputs a second hidden feature to the first output layer. The first output layer outputs the results of the quality of each link.

6. The AI-based space electromagnetic environment big data processing system according to claim 5 is characterized in that: A link indicates that there is an association between two communicating nodes.

7. The AI-based space electromagnetic environment big data processing system according to claim 1 is characterized in that: The allocation model includes a No. 2 backbone network, which includes a third timing layer, a fourth timing layer, a fifth graph network layer, a feature fusion layer and a second output layer, wherein the third timing layer inputs the No. 1 sequential information and outputs the third hidden feature to the feature fusion layer, wherein the fourth timing layer inputs the No. 2 sequential information and outputs the fourth hidden feature to the feature fusion layer, the feature fusion layer outputs the fused feature to the fifth graph network layer, the fifth graph network layer outputs the fifth hidden feature to the second output layer, and the second output layer outputs the results of spectrum allocation of each communication node.

8. The AI-based air-space electromagnetic environment big data processing system according to claim 7 is characterized in that: Based on the link quality data output by the quality model, a second structured data including a structural dimension is constructed. The basic structural element included in this structural dimension is a second timing unit. The structuring method includes: Extract the associated information of the link quality determination task, divide the extracted specific features into multiple parts according to the data collection cycle, map one part to a No. 2 timing unit, and the order of the No. 2 timing unit is consistent with the time order of the collection cycle of the associated information of the link quality determination task.

9. The AI-based space electromagnetic environment big data processing system according to claim 1 is characterized in that: The link quality determination module is trained through supervised learning, and the loss function calculation method of the link quality determination module is: ； in, Represents the loss value of the link quality determination module, represents the total number of link samples, Indicates The actual quality value of the link samples, The link quality determination module outputs The quality value of each link sample; The spectrum allocation module is trained by Q-learning, with instant rewards The calculation method is: ； in, represents the total number of assigned samples, and Indicates Second and The output of the secondary link quality determination module is The quality value of the assigned samples, Indicates execution The switching cost when Indicates execution The spectrum utilization rate is , , are the first, second, and third hyperparameters; Spectrum Utilization The calculation method is: ； in, Indicated in strategy The number of communication nodes allocated under Indicates The bandwidth allocated to each communication node is Indicates The transmission power allocated to each communication node is Represents the total available bandwidth, Indicates the maximum available transmit power; Switching Cost The calculation method is: ； in, Indicated in strategy Next The frequencies assigned to the communication nodes are Indicates that at the last moment Time The frequencies assigned to the communication nodes are Represents the total available frequency range, Indicates the total number of communication nodes.

10. A computer storage medium, characterized in that: It is used to store computer-readable instructions, which, when read, can execute an AI-based aerospace electromagnetic environment big data processing system as described in any one of claims 1-9.

Images