Internet of Things low earth orbit satellite communication data interaction generation method and system

Abstract

The invention provides an Internet of Things low earth orbit satellite communication data interaction generation method and system, and relates to the technical field of data processing, and the method comprises the steps: selecting a satellite as a group management node and other satellites as member satellites in each satellite group, and constructing a collaborative deep reinforcement learning framework; the member satellites perform local model training based on the network state information and the dynamic adjustment coefficient, and a local network weight is obtained after a preset training round is reached; performing weighted averaging on the local network weight, and introducing regularization constraint of the difference between the local network weight and the global network weight to be updated in the weighted averaging process to obtain the updated global network weight; and based on the updated global network weight and in combination with the dynamic adjustment coefficient, executing a routing forwarding decision of the inter-satellite data packet, and completing interaction of the Internet of Things data in the low-orbit satellite communication network. According to the method, the state sensing precision, the resource utilization efficiency and the data interaction stability of the low-orbit satellite network are improved.

Description

IoT Low-Earth Orbit Satellite Communication Data Interaction Generation Method and System Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for generating low-orbit satellite communication data interaction for the Internet of Things. Background Technology

[0002] In marine environmental monitoring scenarios, to achieve comprehensive perception of hydrological and meteorological parameters across vast sea areas, a large number of various IoT sensor nodes are typically deployed within the sea area. The monitoring data collected by these sensor nodes, such as temperature, salinity, ocean current velocity, and wind speed, needs to be transmitted back to the ground processing center via communication links built on low-Earth orbit (LEO) satellite constellations to provide data support for marine environmental analysis and disaster early warning. Due to the low orbital altitude and high speed of satellite nodes, the topology of LEO satellite networks exhibits significant dynamic changes. The relative positions between satellite nodes continuously change with satellite movement, leading to fluctuations in key indicators such as the connectivity of inter-satellite links, transmission bandwidth, and transmission latency. For example, during satellite transit switching, existing inter-satellite links may suddenly be interrupted, requiring the re-establishment of new inter-satellite links, which affects the continuity and stability of data transmission. Currently, most routing methods used for LEO satellite networks are based on preset static topology information. Data forwarding configurations based on fixed shortest path principles have limited adaptive adjustment capabilities. When network topology changes dynamically or inter-satellite link status fluctuates, it may be difficult to quickly make corresponding routing adjustments, easily leading to problems such as data transmission interruptions, increased latency, or decreased transmission efficiency. Limited by the hardware resources and communication range of satellite nodes, a single satellite node can usually only obtain local network status information of its directly connected neighboring satellite nodes, and cannot grasp the global network situation of the entire low-Earth orbit satellite constellation. This limitation of local information perception may lead to a lack of global rationality in routing decisions, further affecting the effectiveness of monitoring data interaction. In addition, considering that the scale of low-Earth orbit satellite constellations is usually large, if a centralized routing decision-making method is adopted, the amount of signaling interaction between the ground processing center and each satellite node will increase significantly, which will not only generate high signaling overhead, but also prolong data transmission latency to a certain extent, which may be difficult to meet the real-time transmission requirements of marine environmental monitoring data. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method and system for generating low-Earth orbit satellite communication data interaction for the Internet of Things, which improves the accuracy of state perception, resource utilization efficiency and data interaction stability of low-Earth orbit satellite networks.

[0004] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a method for generating IoT low-Earth orbit satellite communication data interaction, the method comprising: Step 1, collecting network status information of each satellite node in a low-Earth orbit satellite constellation and converting it into a spatially distributed feature point set, and constructing a state feature field based on the feature point set; Step 2, determining a reference region in the state feature field, and dividing the reference region according to the density of feature points and the gradient of link state changes to obtain multiple feature subdomains, mapping the feature point set to the feature subdomains to generate an aggregated information block; Step 3, performing statistical analysis on the feature points within the aggregated information block, extracting feature parameters, and generating a dynamic adjustment coefficient based on the feature parameters; Step 4, correcting the average connection of the satellite nodes according to the dynamic adjustment coefficient. The process involves: 1) ...

[0005] Secondly, the IoT low-Earth orbit satellite communication data interaction generation system includes: a data acquisition module, used to acquire network status information of each satellite node in the low-Earth orbit satellite constellation and convert it into a spatially distributed feature point set, and construct a state feature field based on the feature point set; a partitioning module, used to determine a reference region in the state feature field, and partition the reference region according to the density of feature points and the gradient of link state changes to obtain multiple feature subdomains, and map the feature point set to the feature subdomains to generate aggregated information blocks; an analysis module, used to perform statistical analysis on the feature points in the aggregated information blocks, extract feature parameters, and generate dynamic adjustment coefficients based on the feature parameters; and a processing module, used to correct the average connectivity of satellite nodes according to the dynamic adjustment coefficients and process the low-Earth orbit satellites. The constellation is dynamically divided into groups, generating multiple satellite groups. A training module selects one satellite as the group management node and the remaining satellites as member satellites within each group, constructing a collaborative deep reinforcement learning framework. Member satellites train their local models based on network state information and dynamic adjustment coefficients, obtaining local network weights after a preset number of training rounds. A weighted average is then applied to these local network weights, incorporating a regularization constraint on the difference between the local network weights and the global network weights to be updated, resulting in updated global network weights. A routing module, based on the updated global network weights and the dynamic adjustment coefficients, executes routing and forwarding decisions for inter-satellite data packets, enabling the interaction of IoT data within the low-Earth orbit satellite communication network.

[0006] Thirdly, a computing device includes: one or more processors; and a storage device for storing one or more programs that, when executed by the one or more processors, cause the one or more processors to implement the method.

[0007] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0008] The above-mentioned solution of the present invention includes at least the following beneficial effects: by collecting network status information of each satellite node and constructing a status feature field, dividing feature subdomains and generating aggregated information blocks according to the density of feature points and the gradient of link status changes, and then extracting feature parameters to obtain dynamic adjustment coefficients, the accurate perception and quantitative representation of dynamic topology changes of low-orbit satellite networks are realized; based on the dynamic adjustment coefficients, the average connectivity of satellite nodes is corrected and dynamic grouping is performed, which can quickly adjust the routing configuration according to the real-time fluctuations of key indicators such as inter-satellite link connectivity, transmission bandwidth, and transmission latency; especially in scenarios such as satellite transit handover and sudden interruption of inter-satellite links, communication links can be quickly rebuilt through dynamic group collaboration, effectively avoiding problems such as data transmission interruption, increased latency, or decreased transmission efficiency, and ensuring the continuity and stability of marine environmental monitoring data backhaul. Attached Figure Description

[0009] Figure 1 is a schematic flowchart of the IoT low-orbit satellite communication data interaction generation method provided by an embodiment of the present invention.

[0010] Figure 2 is a schematic diagram of an IoT low-orbit satellite communication data interaction generation system provided in an embodiment of the present invention. Detailed Implementation

[0011] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0012] As shown in Figure 1, an embodiment of the present invention proposes a method for generating IoT low-Earth orbit satellite communication data interaction. The method includes the following steps: Step 1, collecting network state information of each satellite node in the low-Earth orbit satellite constellation and converting it into a spatially distributed feature point set, and constructing a state feature field based on the feature point set; Step 2, determining a reference region in the state feature field, and dividing the reference region according to the density of feature points and the gradient of link state changes to obtain multiple feature subdomains, mapping the feature point set to the feature subdomains to generate aggregated information blocks; Step 3, performing statistical analysis on the feature points within the aggregated information blocks, extracting feature parameters, and generating dynamic adjustment coefficients based on the feature parameters; Step 4, correcting the average connectivity of the satellite nodes based on the dynamic adjustment coefficients. The low-Earth orbit (LEO) satellite constellation is dynamically divided into multiple satellite groups. In step 5, within each satellite group, one satellite is selected as the group management node and the remaining satellites are member satellites, and a collaborative deep reinforcement learning framework is constructed. Member satellites train local models based on network state information and dynamic adjustment coefficients, and obtain local network weights after reaching a preset number of training rounds. The local network weights are then weighted and averaged, and a regularization constraint is introduced during the weighting process to determine the difference between the local network weights and the global network weights to be updated, resulting in updated global network weights. In step 6, based on the updated global network weights and combined with the dynamic adjustment coefficients, routing and forwarding decisions for inter-satellite data packets are executed to complete the interaction of IoT data in the LEO satellite communication network.

[0013] In this embodiment of the invention, by collecting network status information of each satellite node and constructing a status feature field, feature subdomains are divided according to the density of feature points and the gradient of link status changes, and aggregated information blocks are generated. Then, feature parameters are extracted to obtain dynamic adjustment coefficients, thereby achieving accurate perception and quantitative representation of dynamic topology changes in low-orbit satellite networks. Based on the dynamic adjustment coefficients, the average connectivity of satellite nodes is corrected and dynamic grouping is performed. This enables rapid adjustment of routing configurations according to real-time fluctuations in key indicators such as inter-satellite link connectivity, transmission bandwidth, and transmission latency. Especially in scenarios such as satellite transit handover and sudden interruption of inter-satellite links, communication links can be quickly rebuilt through dynamic group collaboration, effectively avoiding problems such as data transmission interruption, increased latency, or decreased transmission efficiency, and ensuring the continuity and stability of marine environmental monitoring data transmission.

[0014] In a preferred embodiment of the present invention, step 1, collecting network status information of each satellite node in the low-Earth orbit satellite constellation and converting it into a spatially distributed feature point set, and constructing a state feature field based on the feature point set, may include: Step 101, the ground station acquires the network status information of each satellite node in the low-Earth orbit satellite constellation. The network status information includes the instantaneous three-dimensional position coordinates of the satellite node, the signal-to-noise ratio of the link between the satellite node and its neighboring satellite nodes, the queue occupancy rate of the satellite node, and the total energy consumption of the satellite node per unit time. Specifically, in the scenario of marine environmental monitoring, the ground processing center establishes a stable communication connection with each satellite node in the low-Earth orbit satellite constellation through the deployed satellite-to-ground communication link, and receives and acquires the complete network status information of each satellite node at the current moment in real time. This network status information is the core foundation for subsequent satellite network status modeling and routing decisions, and specifically includes four key dimensions. The parameters and value ranges of each dimension are as follows: First, the instantaneous three-dimensional position coordinates of the satellite node in three-dimensional space. These coordinates are collected and reported in real time by the positioning equipment carried by the satellite itself, accurately reflecting the satellite's position. The first is the satellite's current position in orbit, with coordinates ranging from 300km to 1500km, corresponding to three-dimensional spatial coordinates. Longitude ranges from 0° to 360°, latitude from -90° to 90°, and orbital altitude from 300km to 1500km. The second is the signal-to-noise ratio (SNR) of the inter-satellite communication links established between the satellite node and all neighboring satellite nodes. The SNR of each inter-satellite link is monitored in real-time by the satellite node's communication equipment, directly reflecting the transmission quality of the link. The SNR value ranges from 10d. The values ​​range from 0% to 40dB; thirdly, the current occupancy rate of the cache queue used to cache IoT monitoring data to be forwarded within the satellite node. This percentage is calculated by the cache management unit of the satellite node and reflects the current data processing load of the satellite. The queue occupancy rate ranges from 0% to 100%; fourthly, the total energy consumption generated by the satellite node during the process of completing inter-satellite communication, equipment operation, and data forwarding within a unit of time. This value is calculated in real time by the energy monitoring unit of the satellite node and reflects the energy consumption status of the satellite. The total energy consumption ranges from 50W to 200W.

[0015] Step 102: Map each satellite node to a feature point in three-dimensional space. The instantaneous three-dimensional position coordinates of the satellite node are used as the spatial coordinates of the feature point, and the signal-to-noise ratio (SNR), queue occupancy rate, and total energy consumption of the link are used as the attribute values ​​of the feature point. All feature points corresponding to all satellite nodes together constitute a spatially distributed feature point set. Specifically, to achieve a spatialized representation of the discrete satellite node network state and facilitate subsequent analysis and management of the global network situation, each satellite node in the low-Earth orbit satellite constellation is uniformly mapped to an independent feature point in three-dimensional space. The mapping process follows a one-to-one correspondence rule, specifically that each satellite node uniquely corresponds to one three-dimensional spatial feature point, and one three-dimensional spatial feature point corresponds to only one satellite node. There is no situation where one satellite node corresponds to multiple feature points, or one feature point corresponds to multiple satellite nodes. During mapping, the instantaneous three-dimensional position coordinates reported by the satellite node are directly used as the spatial coordinates of the feature point in three-dimensional space, ensuring that the spatial position of the feature point is completely consistent with the actual orbital position of the satellite node. At the same time, the three key state parameters of the satellite node, namely the link signal-to-noise ratio, queue occupancy rate, and total energy consumption, are used as the attribute values ​​of the feature point, realizing the precise association between the satellite node state and the feature point attribute. After the mapping of a single satellite node is completed, the feature points corresponding to all satellite nodes are precisely arranged according to their respective spatial coordinates, and all feature points form a feature point set with a clear distribution pattern in three-dimensional space.

[0016] Step 103: For any location in three-dimensional space, calculate the spatial distance between any location and each feature point in the feature point set. Using the reciprocal of the square of the spatial distance as the weight, sum the attribute values ​​of all feature points in the feature point set to obtain the state feature value at any location. After traversing all locations in three-dimensional space, obtain the state feature field. Specifically, this embodiment collects four core types of information: the three-dimensional location of satellite nodes, link signal-to-noise ratio, queue occupancy rate, and energy consumption. It also clarifies the specific parameters and value ranges of each type of information and maps them to a set of spatial feature points, following a one-to-one correspondence rule. This effectively solves the problem of dispersed satellite nodes and independent states in low-Earth orbit satellite constellations.

[0017] In a preferred embodiment of the present invention, step 2, determining a reference region in the state feature field and dividing the reference region according to the density of feature points and the gradient of link state changes to obtain multiple feature subdomains, mapping the feature point set to the feature subdomains to generate aggregated information blocks, may include: step 201, in the state feature field, determining a reference region with the distribution range of the spatial coordinates of all feature points as the boundary; dividing the reference region into multiple equally spaced grid cells, counting the number of feature points contained in each grid cell, and calculating the feature point density value at each location in the reference region based on the number of feature points, specifically including: on the basis of the already constructed state feature field, in order to realize the partitioned management of satellite network state, first determining the reference region for subsequent region division. The reference region is determined by using the maximum distribution range covered by the spatial coordinates of all feature points as its boundary. This ensures that the reference region completely includes all feature points corresponding to satellite nodes, avoiding any omissions. Specifically, the boundary is determined by taking the maximum and minimum longitude, maximum and minimum latitude, and maximum and minimum orbital altitude values ​​from the three-dimensional coordinates of all feature points. These six values ​​are used as the boundary parameters of the reference region. After determining the reference region, it is uniformly divided into multiple grid cells of the same size at equal intervals. The spacing between the grid cells is reasonably set according to the distribution density of the satellite nodes, with a spacing range of 10. The grid spans from km to 50km to ensure accurate reflection of the distribution differences of feature points without excessive computation due to overly dense grids. After segmentation, the number of feature points contained within each grid cell is counted. Based on the number of feature points in each grid cell, the feature point density value corresponding to each location within the reference area is calculated. Specifically, the number of feature points within the grid cell where the current location is located is combined with the area of ​​the grid cell to calculate the number of feature points per unit area, which is used as the feature point density value for that location. The feature point density value ranges from 0.001 points / km² to 0.01 points / km².

[0018] Step 202: Based on the state characteristic value at each location in the state characteristic field, calculate the spatial gradient value of the state characteristic value at each location within the reference area. Specifically, this includes: to divide the region in conjunction with the changing trend of the satellite network state, calculate the amplitude of the change of the state characteristic value at each location within the reference area in different spatial directions based on the state characteristic value corresponding to each location in the state characteristic field. The specific calculation process is as follows: select 8 adjacent locations around the current location, with the distance between adjacent locations being consistent with the grid cell distance in step 201; calculate the difference between the state characteristic values ​​of these adjacent locations and the current location respectively; and then determine the amplitude of the change of the state characteristic value at the current location in each direction based on the magnitude and direction of the difference. This amplitude of change is determined as the gradient value of the state characteristic value corresponding to the current location, with the gradient value ranging from 0 to 10.

[0019] Step 203 involves weighted fusion of feature point density values ​​and gradient change values ​​to obtain a comprehensive division index value for each location. Local maxima of the comprehensive division index value are used as boundary points. Connecting all boundary points yields multiple feature subdomains. Specifically, to achieve a reasonable division of the baseline region, the feature point density values ​​and gradient change values ​​for each location are weighted and fused according to a preset ratio. The preset ratio is 60% for feature point density values ​​and 40% for gradient change values. During the fusion process, based on the operational requirements of the satellite network in the marine environmental monitoring scenario, weight coefficients are set for each feature point density value and gradient change value. The weight coefficient for feature point density values ​​ranges from 0.5 to 0.7, and the weight coefficient for gradient change values ​​ranges from 0.3 to 0.5. In this study, a weight coefficient of 0.6 for feature point density values ​​and a weight coefficient of 0.4 for gradient change values ​​are preferred. The weight coefficients are determined by considering the satellite node distribution characteristics and network stability requirements to ensure a reasonable impact of both on the region division. The specific fusion method is as follows: the feature point density value is multiplied by its corresponding weight coefficient, the change gradient value is multiplied by its corresponding weight coefficient, and the two products are added together to obtain the comprehensive partition index value corresponding to the current position. The comprehensive partition index value ranges from 0 to 5. The comprehensive partition index value can comprehensively reflect the satellite distribution density and the degree of network state change at this position. Then, local maxima are selected from the comprehensive partition index values ​​of all positions. The selection criterion is that the comprehensive partition index value of the current position is greater than the comprehensive partition index values ​​of its 8 neighboring positions. The positions corresponding to these local maxima are the key dividing points of satellite distribution density and network state change. These local maxima are used as the boundary points of the region partitioning. Then, all the partitioning boundary points are connected in sequence according to the spatial distribution order to form multiple continuous partitioning boundaries. These boundaries divide the benchmark region into multiple independent and non-overlapping feature subdomains.

[0020] Step 204: Based on the spatial coordinates of each feature point, divide each feature point into its corresponding feature subdomain. This allows each feature subdomain to aggregate a group of feature points with similar spatial locations. Each feature subdomain and its aggregated feature points together constitute an aggregated information block. Specifically, to achieve centralized management of satellite nodes, based on the spatial coordinates of each feature point, determine the feature subdomain in which the feature point belongs. The determination process involves comparing the spatial coordinates of the feature point with the boundary ranges of each feature subdomain to determine which feature subdomain the feature point belongs to. Specifically, the comparison method involves determining whether the three-dimensional coordinates of the feature point are all within the longitude and latitude of a certain feature subdomain. Within the orbital height boundary, if all features are within the range, then the feature point belongs to that feature subdomain. Each feature point is then assigned to a feature subdomain that matches its spatial location. This division method aggregates a group of spatially similar and network-state-similar feature points within each feature subdomain, avoiding the clustering of spatially dispersed feature points with significant state differences in the same subdomain. The criteria for spatial proximity is a straight-line spatial distance of less than 50km between feature points, and the criteria for network state similarity is a difference in state feature values ​​of less than 10 between feature points. Each feature subdomain and all feature points contained within it together constitute an aggregated information block.

[0021] This embodiment adaptively divides spatial regions by combining feature point density and state change gradient, clarifying specific parameters and value ranges such as spacing and weight coefficients in the division process. It can automatically aggregate satellite nodes with similar spatial locations and network states into information blocks based on the actual spatial distribution of satellite nodes and changes in network state, clarifying specific judgment criteria for similarity, effectively improving the identification accuracy and management efficiency of dynamically changing satellite topology, and reducing the complexity of managing dispersed satellite nodes individually.

[0022] In a preferred embodiment of the present invention, step 3, statistically analyzing the feature points within the aggregated information block, extracting feature parameters, and generating dynamic adjustment coefficients based on the feature parameters, may include: Step 301, for each aggregated information block, obtaining all feature points contained within the aggregated information block; based on the spatial coordinates of all feature points, sequentially calculating the geometric moments of the feature point distribution within the aggregated information block, calculating the central moments of each order based on the geometric moments of each order, normalizing the central moments of each order to obtain the normalized central moments, and calculating seven invariant moment values ​​based on the fixed combination relationship of the normalized central moments. Specifically, this includes: for each aggregated information block, first obtaining all feature points contained within the block to ensure that no feature points are omitted or redundant, providing a complete data foundation for subsequent feature parameter extraction. Based on the spatial coordinates of all feature points, the geometric moments of each order of feature points in their spatial distribution are calculated sequentially. The calculation of geometric moments starts with the first-order geometric moment and proceeds step by step to the third-order geometric moment. The calculation of each order of geometric moment is based on the spatial coordinates of all feature points, reflecting the distribution position and morphological characteristics of the feature points in space. Among them, the first-order geometric moment reflects the centroid position of the feature point distribution, the second-order geometric moment reflects the dispersion of the feature point distribution, and the third-order geometric moment reflects the symmetry of the feature point distribution. After calculating each order of geometric moments, the central moments of each order are further calculated based on each order of geometric moments. The calculation of the central moments is based on the centroid of the feature point distribution. Specifically, the calculation method is to subtract the centroid coordinates from the coordinates of each feature point before calculating the corresponding order of geometric moments, eliminating the influence of spatial position translation on feature description and making the calculation results more stable.

[0023] The central moments of each order are normalized using a min-max normalization method. Specifically, the minimum value of all central moments is subtracted from each central moment, and then the result is divided by the difference between the maximum and minimum values ​​to eliminate the influence of scale variations in feature point distribution, resulting in normalized central moments. The normalized central moments range from 0 to 1. The normalized central moments are then combined according to a pre-defined fixed combination relationship: the first invariant moment is the square root of the second-order normalized central moment; the second invariant moment is the ratio of the absolute value of the third-order normalized central moment to the 1.5th power of the second-order normalized central moment; the third invariant moment is the ratio of the fourth-order normalized central moment to the square of the second-order normalized central moment; and the fourth invariant moment is the ratio of the fifth-order normalized central moment to the square of the second-order normalized central moment. The ratio of the normalized central moment to the 2.5th power of the second normalized central moment; the fifth invariant moment is the square of the product of the second and third normalized central moments; the sixth invariant moment is the ratio of the cube of the second normalized central moment to the fourth normalized central moment; and the seventh invariant moment is the ratio of the square of the third normalized central moment to the cube of the second normalized central moment. The combination relationship is set based on the description requirements of the feature point distribution shape, ensuring that the calculation results can stably describe the distribution shape of the feature points and do not change with spatial location or scale. Through this combination operation, seven invariant moment values ​​are finally obtained, each ranging from 0 to 1. These seven invariant moment values ​​together constitute the core descriptive parameters of the feature point distribution shape.

[0024] Step 302: Based on the attribute values ​​of all feature points, calculate the average signal-to-noise ratio (SNR), queue occupancy rate, and total energy consumption of all feature points within the aggregated information block. This involves: to comprehensively reflect the network operation status of the aggregated information block, statistically analyzing the three types of attribute values ​​based on the attribute values ​​of all feature points within the aggregated information block, and calculating the corresponding average values. Specifically, the process involves first collecting the SNR values ​​of all feature points within the aggregated information block, summing all SNR values, and then dividing the sum by the total number of feature points within the block to obtain the average SNR corresponding to the aggregated information block. This average value reflects the inter-satellite communication between satellite nodes within the block. The overall transmission quality of the link is assessed, with the average signal-to-noise ratio ranging from 10dB to 40dB. Then, the queue occupancy rate of all feature points within the block is collected, summed, and divided by the total number of feature points to obtain the average queue occupancy rate for the aggregated information block. This average reflects the overall data processing load of the satellite nodes within the block, and the average queue occupancy rate ranges from 0% to 100%. Finally, the total energy consumption of all feature points within the block is collected, summed, and divided by the total number of feature points to obtain the average energy consumption for the aggregated information block. This average reflects the overall energy consumption level of the satellite nodes within the block, and the average energy consumption ranges from 50W to 200W.

[0025] Step 303 involves combining the seven invariant moment values, the average signal-to-noise ratio, the average queue occupancy rate, and the average energy consumption into a feature parameter vector, which serves as the feature parameters of the aggregated information block. Specifically, to form feature parameters that can comprehensively describe the aggregated information block, the seven invariant moment values ​​obtained in step 301, the average signal-to-noise ratio, the average queue occupancy rate, and the average energy consumption obtained in step 302 are combined in a fixed order. The fixed combination order is as follows: first invariant moment value, second invariant moment value, third invariant moment value, fourth invariant moment value, fifth invariant moment value, sixth invariant moment value, seventh invariant moment value, average signal-to-noise ratio, average queue occupancy rate, and average energy consumption. The combination order is preset and kept consistent to ensure the comparability of feature parameters of different aggregated information blocks. Through this combination process, a feature parameter vector containing 10 dimensions is formed, and this multi-dimensional feature parameter vector is used as the feature parameters corresponding to the current aggregated information block.

[0026] Step 304: Normalize each component in the feature parameter vector to obtain normalized shape feature index, signal-to-noise ratio index, queue occupancy rate index, and energy consumption index. Specifically, since the components in the feature parameter vector are of different types and have significantly different numerical ranges, directly using them for subsequent calculations would lead to inconsistent influence scales of different parameter types, thus affecting the accuracy of the calculation results. Therefore, it is necessary to normalize each component in the feature parameter vector separately. The specific normalization method is min-max normalization, which involves subtracting the value of each component from the value of the corresponding component. The minimum value of the quantity in all aggregated information blocks is divided by the difference between the maximum and minimum values ​​of that component. The value of each component is then uniformly adjusted to the range of 0 to 1 to ensure that the numerical scale of all components is consistent. After normalization, the normalized shape feature index, signal-to-noise ratio index, queue occupancy rate index, and energy consumption index are obtained. The shape feature index corresponds to the normalized result of seven invariant moment values, the signal-to-noise ratio index corresponds to the normalized result of the average signal-to-noise ratio, the queue occupancy rate index corresponds to the normalized result of the average queue occupancy rate, and the energy consumption index corresponds to the normalized result of the average energy consumption.

[0027] Step 305 involves weighted summation of the normalized shape feature index, signal-to-noise ratio index, queue occupancy rate index, and energy consumption index to obtain the dynamic adjustment coefficient corresponding to the aggregated information block. Specifically, to generate dynamic adjustment coefficients usable for subsequent network regulation, the normalized shape feature index, signal-to-noise ratio index, queue occupancy rate index, and energy consumption index are multiplied by their respective preset weighting coefficients. The preset weighting coefficients are set based on the regulation requirements of the satellite network in the marine environmental monitoring scenario, determined according to the degree of influence of different indicators on network operation, ensuring that the influence weight of each indicator is reasonable. The weighting coefficient for the shape feature index ranges from 0.2 to 0.3, and the weighting coefficient for the signal-to-noise ratio index ranges from [missing value]. The weight coefficients for the queue occupancy rate index range from 0.3 to 0.4, the weight coefficients for the energy consumption index range from 0.1 to 0.2, and the sum of the four weight coefficients is 1. The preferred weight coefficients for the shape feature index are 0.25, the signal-to-noise ratio index is 0.35, the queue occupancy rate index is 0.25, and the energy consumption index is 0.15. The specific calculation process involves multiplying each normalized index by its corresponding weight coefficient to obtain the weighted value for each index. Then, the weighted values ​​of all indices are summed sequentially to complete the weighted summation operation, ultimately obtaining the dynamic adjustment coefficient corresponding to the current aggregated information block. The dynamic adjustment coefficient ranges from 0 to 1.

[0028] This embodiment obtains dynamic adjustment coefficients by normalizing and weighting different types of parameters, clarifying the normalization method, the weight of each indicator, and the range of values. It does not rely on centralized control of ground stations and can realize distributed autonomous control of satellite networks. It effectively reduces the amount of signaling interaction between satellites and between satellites and ground, reduces signaling interaction overhead, and can also achieve adaptive adjustment of satellite connectivity, group division, and routing strategies.

[0029] In a preferred embodiment of the present invention, step 4, correcting the average connectivity of satellite nodes according to the dynamic adjustment coefficient and dynamically grouping the low-Earth orbit satellite constellation to generate multiple satellite groups, may include: step 401, according to the dynamic adjustment coefficient, calculating the average number of communication links established by a satellite node with other surrounding satellite nodes within a preset historical period, and using the average number as the initial average connectivity of the satellite node, specifically including: according to the dynamic adjustment coefficient obtained in step 305, calculating the average number of communication links established by each satellite node with other surrounding satellite nodes within a preset historical period, and using the average number as the initial average connectivity of the satellite node. The preset historical time period is a pre-defined time range for accurately counting the establishment of communication links between satellite nodes, used to avoid statistical deviations caused by short-term link fluctuations. Specifically, it is set to 10 to 30 minutes, with 20 minutes being preferred in this case. The statistical process involves recording the duration of each communication link established between each satellite node and its neighboring satellite nodes within this historical time period, filtering out valid communication links. The criteria for a valid communication link are a link signal-to-noise ratio of not less than 10 dB and a link duration of not less than 10 seconds. The total number of valid communication links within this historical time period is counted, and the total number of times is divided by the total number of minutes in the historical time period to obtain the average number of times a satellite node establishes a valid communication link per minute. This average number is then multiplied by the total number of minutes in the historical time period to obtain the average number of valid communication links established by the satellite node within this historical time period. This average number is the initial average connectivity of the satellite node.

[0030] Step 402: Multiply the dynamic adjustment coefficient by the initial average connectivity to obtain the corrected average connectivity of the satellite nodes; based on the corrected average connectivity of each satellite node, sort all satellite nodes in descending order to obtain the sorted node sequence. Specifically, this includes: multiplying the dynamic adjustment coefficient obtained in step 305 by the initial average connectivity obtained in step 401 to obtain the corrected average connectivity of the satellite nodes. The specific calculation method is to multiply the dynamic adjustment coefficient corresponding to each satellite node by the initial average connectivity of that satellite node, and the product result is the corrected average connectivity. The corrected average connectivity can optimize the initial average connectivity by combining the spatial shape and operating status of the aggregated information block where the satellite node is located, so that the connectivity is more in line with the actual network control requirements; after calculating the corrected average connectivity of each satellite node, sort all satellite nodes in descending order according to the corrected average connectivity. The sorting process is to arrange all satellite nodes in descending order of their corrected average connectivity to form the sorted node sequence.

[0031] Step 403: Select nodes sequentially from the sorted node sequence. If the currently selected node has not yet been assigned to any satellite group, set the node as the new group center node. Starting from the new group center node, search in three-dimensional space for surrounding satellite nodes whose spatial distance from the new group center node is less than a preset distance threshold and whose link signal-to-noise ratio (SNR) with the new group center node is higher than a preset SNR threshold. Aggregate the searched surrounding satellite nodes and the new group center node into a satellite group. For the remaining satellite nodes that have not yet been assigned to any satellite group, repeat the node selection and group aggregation process. Until all satellite nodes are assigned to their corresponding satellite groups, generating multiple satellite groups, the process involves: selecting nodes sequentially from the sorted node sequence in chronological order; during the selection process, determining whether each selected node has already been assigned to any satellite group by querying its group identifier; if the group identifier is empty, the node is determined not to have been assigned and is then set as the new group center node; starting from the new group center node, searching in three-dimensional space for links whose spatial distance to the center node is less than a preset distance threshold and whose link signal-to-noise ratio (SNR) is higher than a preset SNR threshold. The search considers the surrounding satellite nodes. A preset distance threshold is a pre-set distance standard to ensure that satellite nodes within the group are spatially close, ranging from 30km to 80km, with 50km being preferred in this study. A preset signal-to-noise ratio (SNR) threshold is a pre-set SNR standard to ensure good inter-satellite link quality within the group, ranging from 15dB to 25dB, with 20dB being preferred in this study. The search process involves calculating the straight-line spatial distance between each surrounding satellite node and the central node, while simultaneously detecting the SNR value of the inter-satellite communication link between them. Satellite nodes with a distance less than the preset distance threshold and an SNR value higher than the preset SNR threshold are selected. The search process involves aggregating the surrounding satellite nodes and the new group center node into a satellite group, assigning a unique group identifier to the group, and marking all satellite nodes within the group with this group identifier. For the remaining satellite nodes in the sorted node sequence that have not yet been assigned to any satellite group, the above node selection and group aggregation process is repeated. That is, the unassigned nodes are selected sequentially from the remaining nodes as new group center nodes, and the search and aggregation operations are repeated until all satellite nodes are marked with group identifiers and assigned to the corresponding satellite groups, ultimately generating multiple independent and non-overlapping satellite groups.

[0032] This embodiment corrects the average connectivity of satellite nodes by dynamically adjusting the coefficient and performs dynamic group division. It can optimize the group structure according to the actual operating status of the satellite network, making the spatial distance and link quality of satellite nodes in the group more compatible, and improving the stability and collaborative efficiency of the satellite group.

[0033] In a preferred embodiment of the present invention, step 5 involves selecting one satellite as the group management node and the remaining satellites as member satellites within each satellite group, and constructing a collaborative deep reinforcement learning framework. Member satellites train their local models based on network state information and dynamic adjustment coefficients, obtaining local network weights after reaching a preset number of training rounds. A weighted average is then applied to the local network weights, incorporating a regularization constraint on the difference between the local network weights and the global network weights to be updated during the weighted averaging process, to obtain the updated global network weights. This step may include: step 501, comparing all corrected average connectivity values ​​within each satellite group, and setting the corrected average connectivity value... The largest satellite node is selected as the group management node, and the remaining satellite nodes in the satellite group other than the group management node are selected as member satellites. Specifically, in each satellite group generated in step 403, the corrected average connectivity of all satellite nodes in the group is compared one by one. The comparison process is to compare the corrected average connectivity of each satellite node in the group pairwise, and select the satellite node with the largest corrected average connectivity value. This satellite node is selected as the group management node of the satellite group. The group management node is responsible for subsequent global model initialization, weight aggregation and update operations. All other satellite nodes in the satellite group other than the group management node are selected as member satellites.

[0034] Step 502: The group management node initializes the global deep neural network model and distributes the global network weights of the global deep neural network model to each member satellite in the satellite group. After receiving the global network weights, each member satellite trains its local model based on its own network state information and dynamic adjustment coefficients. After reaching a preset number of training rounds, it obtains its local network weights. Specifically, the group management node of each satellite group initializes the global deep neural network model. The model construction process is as follows: a deep neural network containing an input layer, a hidden layer, and an output layer is built. The number of nodes in the input layer is set to 10, corresponding to each component of the 10-dimensional feature parameter vector obtained in step 303. The hidden layer has 3 layers, with 64, 32, and 16 nodes in each layer, respectively. The number of nodes in the output layer is 1, corresponding to the action value of inter-satellite routing decision. The ReLU function is used as the activation function to enhance the nonlinear fitting ability of the model. The mean squared error loss function is used as the loss function to measure the deviation between the model's predicted value and the actual value. After initialization, the group management node distributes the global network weights of the global deep neural network model to each member satellite in the satellite group. The distribution process is completed through inter-satellite communication links to ensure that the global network weights received by each member satellite are completely consistent. The initial weight values ​​between the input layer and the first hidden layer range from -0.5 to 0.5, between the first hidden layer and the second hidden layer range from -0.4 to 0.4, between the second hidden layer and the third hidden layer range from -0.3 to 0.3, between the third hidden layer and the output layer range from -0.2 to 0.2, and the initial value of the bias term for each layer ranges from -0.1 to 0.1.

[0035] After receiving the global network weights, each member satellite uses them as the initial weights for its local model. Based on the member satellite's own network state information and the dynamic adjustment coefficient corresponding to the aggregated information block it belongs to, the local model is trained. The local model training process involves inputting the member satellite's 3D position coordinates, link signal-to-noise ratio, queue occupancy rate, total energy consumption, and dynamic adjustment coefficients into the local deep neural network model. Forward propagation is used to calculate the model output value. Then, a loss function is used to calculate the deviation between the output value and the actual routing decision effect. The local network weights are adjusted using a backpropagation algorithm. The backpropagation algorithm specifically calculates the partial derivative of the loss function with respect to the output layer weights. The result of the partial derivative calculation is multiplied by the learning rate to obtain the output layer weight adjustment. The value range is from 0.001 to 0.01, with 0.005 being the preferred value in this study. Then, the partial derivatives of the loss function with respect to the weights of the third, second, first, and input layers are calculated sequentially. These partial derivatives are multiplied by the learning rate to obtain the weight adjustment for each layer. The weight adjustment for each layer must be controlled within the range of -0.01 to 0.01 to avoid excessive weight adjustments that could cause model oscillations. Finally, the current weights of each layer are subtracted from the corresponding adjustment amount to complete the local network weight update. Each forward and backward propagation round constitutes one training cycle. The preset number of training cycles is a pre-defined range to ensure sufficient local model training and avoid overtraining, ranging from 50 to 200 cycles, with 100 cycles being the preferred value in this study. Training stops after the preset number of training cycles is reached, yielding the trained local network weights.

[0036] Step 503: Send the local network weights to the group management node. After receiving the local network weights, the group management node calculates the aggregate weights corresponding to each member satellite based on the dynamic adjustment coefficient. It then multiplies the local network weights by the corresponding aggregate weights and sums them to obtain the initial weighted average aggregate result. Specifically, after each member satellite completes local model training and obtains its local network weights, it sends the local network weights to the group management node of its satellite group via the inter-satellite communication link. During transmission, the local network weights are encrypted to ensure data transmission security and prevent interference or tampering with the weight data. The sent local network weights must be consistent with the weight range of each layer after training: the weights of the input layer and the first hidden layer are -0.5 to 0.5; the weights of the first and second hidden layers are -0.4 to 0.4; the weights of the second and third hidden layers are -0.3 to 0.3; and the weights of the third hidden layer and the output layer are -0. The bias term ranges from -0.1 to 0.1. After receiving the local network weight sent by each member satellite, the group management node calculates the aggregate weight of each member satellite based on the dynamic adjustment coefficient corresponding to the aggregate information block where the member satellite is located. Specifically, the dynamic adjustment coefficient corresponding to the member satellite is divided by the sum of the dynamic adjustment coefficients of all member satellites in the satellite group to obtain the aggregate weight of the member satellite. The sum of the aggregate weights of all member satellites is 1, and the aggregate weight ranges from 0 to 1 to ensure that the aggregate weight can reflect the reliability of the local model of the member satellite. After calculating the aggregate weight, the local network weight of each member satellite is multiplied by the corresponding aggregate weight to obtain the weighted value of the local weight of the member satellite. The range of the weighted value is consistent with the range of the local network weight of the corresponding layer. Then, the weighted values ​​of all member satellites are accumulated sequentially to complete the weighted summation operation and obtain the initial aggregate result of the weighted average.

[0037] Step 504: Based on the initial aggregation result, calculate the difference between the local network weight and the global network weight. Multiply the difference by a preset regularization coefficient to obtain the regularization adjustment term for each member satellite. Add the regularization adjustment terms of all member satellites to the initial aggregation result to obtain the updated global network weight. Specifically, based on the initial aggregation result obtained in step 503, calculate the difference between the local network weight of each member satellite and the global network weight initialized in step 502. The calculation method is to subtract the global network weight from the local network weight of each member satellite to obtain the difference. The difference ranges from -1.0 to 1.0, where the weight difference between the input layer and the first hidden layer is -1.0 to 1.0, the weight difference between the first and second hidden layers is -0.8 to 0.8, the weight difference between the second and third hidden layers is -0.6 to 0.6, and the weight difference between the third hidden layer and the output layer is -0.4 to 0.4. Multiply this difference by a preset regularization coefficient. The regularization coefficient is used to obtain the regularization adjustment term for each member satellite. The preset regularization coefficient is a pre-set coefficient to constrain the difference between the local network weights and the global network weights, avoiding excessive local model training bias that could lead to global model instability. The value range is 0.001 to 0.01, and 0.005 is preferred in this case. This coefficient setting combines the stability requirements of low-Earth orbit satellite group collaborative training, effectively constraining bias without excessively limiting the adaptability of the local model. The regularization adjustment term ranges from -0.01 to 0.01. The regularization adjustment term is used to constrain the difference between the local network weights and the global network weights, avoiding excessive local model training bias that could lead to global model instability. After calculating the regularization adjustment term for each member satellite, the regularization adjustment terms of all member satellites are summed sequentially to obtain the regularization adjustment sum. The value range of the regularization adjustment sum is -0.1 to 0.1. Then, the regularization adjustment sum is superimposed on the initial aggregation result to obtain the updated global network weights.

[0038] This embodiment, by constructing a collaborative deep reinforcement learning framework within a satellite group, clarifies the division of labor between the group management node and member satellites, constructs a detailed global deep neural network model, and completes local training and weight aggregation, can integrate the operational experience of all satellite nodes within the group and improve the model's adaptability to the satellite network state.

[0039] In a preferred embodiment of the present invention, in step 6, based on the updated global network weights and in combination with the dynamic adjustment coefficient, perform the routing and forwarding decision of the inter-satellite data packets to complete the interaction of the Internet of Things data in the low-Earth orbit satellite communication network, which may include: Step 601, each satellite node loads the updated global network weights received into the local deep neural network to obtain a local routing decision network; for data forwarding, after the satellite node receives the Internet of Things data packet to be forwarded, according to the destination satellite node identifier of the data packet and the link status information of all neighbor satellite nodes, a state vector is combined and generated, specifically including: each satellite node receives the updated global network weights sent by the satellite group management node of its own satellite group through the inter-satellite communication link, loads the global network weights into its own local deep neural network, replaces the original weight parameters of the local model, and obtains a local routing decision network, ensuring that the routing decision network of each satellite node is consistent with the global model while taking into account the local operation characteristics; for accurately screening and adapting neighbor satellite nodes and improving the spatial adaptability of routing decisions, before forwarding the Internet of Things data for ocean environment monitoring, first execute the plane and sphere intersection area calculation algorithm to determine the effective spatial area within the communication range of the satellite node. The specific calculation process is as follows: taking the current satellite node as the center of the sphere and constructing a sphere with the maximum effective distance R of the inter-satellite communication as the radius, the value range of R is 30 km to 80 km, and 50 km is preferably selected this time. The sphere equation is , where ([[]] , , ) is the three-dimensional position coordinate of the current satellite node; construct a plane perpendicular to the line connecting the current satellite node and the center of the ocean monitoring target area, and the plane equation is , where A, B, and C are the three components of the plane normal vector, corresponding to the direction vector of the line connecting the current satellite node and the center of the monitoring target area, and D is the distance from the plane to the center of the sphere, with a value range of 0 km to R, and 25 km is preferably selected this time; calculate the distance from the center of the sphere to the plane ; judge the intersection situation. If d > R, there is no intersection area, and at this time, only all neighbor satellite nodes within the sphere are selected; if d = R, the intersection area is a point, and the neighbor satellite node corresponding to this point is selected; if d < R, the intersection area is a circular cross-section, calculate the cross-section radius , and the cross-section center coordinates are Only neighboring satellite nodes located within the circular cross-sectional area are selected as candidate nodes. After calculating the intersection area, the forwarding operation of the marine environmental monitoring IoT data is performed. After receiving the IoT data packet to be forwarded, each satellite node first extracts the destination satellite node identifier of the data packet. This identifier is a unique identifier for each satellite node and is used to determine the final receiving node of the data packet. At the same time, the link state information of the above-selected candidate neighboring satellite nodes is collected. The link state information specifically includes the link signal-to-noise ratio, link transmission delay, and link bandwidth between the neighboring satellite node and the selected satellite node. The destination satellite node identifier of the data packet is converted into a numerical form and combined with the numerical values ​​corresponding to the link state information of all candidate neighboring satellite nodes in a fixed order to generate a state vector for routing decision.

[0040] Step 602: Input the state vector into the routing decision network to calculate the action value of each neighboring satellite node; based on the action value and the dynamic adjustment coefficient corresponding to the aggregated information block to which the satellite node belongs as the exploration probability, determine one neighboring satellite node as the next-hop satellite node from all neighboring satellite nodes. Specifically, this includes: inputting the state vector generated in step 601 into the local routing decision network, performing forward propagation calculations through the network's input layer, hidden layer, and output layer in sequence to obtain the action value corresponding to each candidate neighboring satellite node. The action value is used to measure the rationality and efficiency of forwarding data packets to that neighboring satellite node. The higher the action value, the higher the transmission efficiency and the more stable the link when forwarding to that neighboring satellite node; calculate the action value of all candidate neighboring satellite nodes. After selecting the action value of a neighboring satellite node, the exploration probability is determined based on the action value and the dynamic adjustment coefficient corresponding to the aggregated information block to which the satellite node belongs. Specifically, the dynamic adjustment coefficient is used as the base value of the exploration probability. The higher the action value of a neighboring satellite node, the higher its exploration probability. The exploration probability is used to balance the stability and exploratory nature of routing decisions and avoid getting trapped in local optima. Based on the determined exploration probability, a neighboring satellite node is selected as the next-hop satellite node from all candidate neighboring satellite nodes. The selection method is to select the neighboring satellite node with the highest exploration probability as the next hop. If multiple neighboring satellite nodes have the same exploration probability, the neighboring satellite node with the highest link signal-to-noise ratio is selected to ensure that the selection of the next-hop node matches the actual link status.

[0041] Step 603: The IoT data packet is forwarded to the next-hop satellite node. Upon receiving the packet, the next-hop satellite node compares the destination satellite node identifier with its own identifier. If they match, data reception is complete. If they do not match, the next-hop satellite node is used as the new forwarding satellite node, and the packet forwarding and judgment process is repeated until the packet reaches the destination satellite node, completing the interaction of IoT data in the low-Earth orbit satellite communication network. Specifically, this includes: after determining the next-hop satellite node, the satellite node forwards the IoT data packet to be forwarded to the next-hop satellite node via the inter-satellite communication link. During the forwarding process, the link status is monitored in real time. Simultaneously, based on the calculation results of the intersection area between the plane and the sphere in step 601, it is verified whether the next-hop satellite node is still within the valid intersection area. If it exceeds the area, candidate nodes are immediately re-selected to ensure that the data packet can be transmitted completely and efficiently. After receiving the data packet, the next-hop satellite node first extracts the destination satellite node identifier from the data packet. The identifier is compared with its own unique identifier. The comparison process involves checking the identifier's value bit by bit to determine if the two are completely consistent. If the destination satellite node identifier matches its own identifier, it means that the satellite node is the final receiving node for the data packet, and the data reception is completed. The reception result is then fed back to the initial sending node of the data packet. If the destination satellite node identifier does not match its own identifier, the next-hop satellite node is used as the new forwarding satellite node, and the operations of steps 601 to 602 are repeated. That is, the new forwarding satellite node first performs the plane-sphere intersection region calculation algorithm to filter candidate neighbor satellite nodes, then extracts the destination identifier of the data packet, collects the link status of the candidate neighbor nodes, generates a state vector, and inputs it into the routing decision network to determine the new next-hop node. The data packet forwarding, region verification, and identifier judgment process is repeated until the data packet reaches the destination satellite node, and finally the interaction of IoT data in the low-Earth orbit satellite communication network is completed.

[0042] This embodiment performs routing and forwarding decisions based on the updated global network weights and dynamic adjustment coefficients. It combines the destination identifier of the data packet with the link status of neighboring nodes to generate decision criteria. By exploring the stability and exploratory nature of probability balance decision-making, it ensures that the routing decisions can fit the real-time network status, improve the efficiency and reliability of IoT data forwarding, reduce data transmission latency and packet loss rate, and ensure that marine environmental monitoring data can be exchanged in a timely and complete manner.

[0043] As shown in Figure 2, an embodiment of the present invention also provides an IoT low-Earth orbit satellite communication data interaction generation system, comprising: a data acquisition module, used to acquire network status information of each satellite node in a low-Earth orbit satellite constellation and convert it into a spatially distributed feature point set, and construct a state feature field based on the feature point set; a partitioning module, used to determine a reference region in the state feature field, and partition the reference region according to the density of feature points and the gradient of link state changes to obtain multiple feature subdomains, map the feature point set to the feature subdomains, and generate aggregated information blocks; an analysis module, used to perform statistical analysis on the feature points within the aggregated information blocks, extract feature parameters, and generate dynamic adjustment coefficients based on the feature parameters; and a processing module, used to correct the average connectivity of satellite nodes according to the dynamic adjustment coefficients, and The system dynamically divides the low-Earth orbit (LEO) satellite constellation into multiple satellite groups. A training module selects one satellite as the group management node and the remaining satellites as member satellites within each group, constructing a collaborative deep reinforcement learning framework. Member satellites train their local models based on network state information and dynamic adjustment coefficients, obtaining local network weights after a preset number of training rounds. A weighted average of these local network weights is then applied, incorporating a regularization constraint on the difference between the local network weights and the global network weights to be updated, resulting in updated global network weights. A routing module, based on the updated global network weights and the dynamic adjustment coefficients, executes routing and forwarding decisions for inter-satellite data packets, facilitating the interaction of IoT data within the LEO satellite communication network.

[0044] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0045] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0046] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0047] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for generating low-Earth orbit satellite communication data interaction for the Internet of Things, characterized in that, The method includes: Step 1, collecting network state information of each satellite node in the low-Earth orbit satellite constellation and converting it into a spatially distributed feature point set, and constructing a state feature field based on the feature point set; Step 2, determining a reference region in the state feature field, and dividing the reference region according to the density of feature points and the gradient of link state changes to obtain multiple feature subdomains, mapping the feature point set to the feature subdomains to generate aggregated information blocks; Step 3, performing statistical analysis on the feature points within the aggregated information blocks, extracting feature parameters, and generating dynamic adjustment coefficients based on the feature parameters; Step 4, correcting the average connectivity of satellite nodes according to the dynamic adjustment coefficients, and dynamically grouping the low-Earth orbit satellite constellation to generate... Multiple satellite groups; Step 5: Within each satellite group, select one satellite as the group management node and the remaining satellites as member satellites, and construct a collaborative deep reinforcement learning framework; Member satellites train local models based on network state information and dynamic adjustment coefficients, and obtain local network weights after reaching a preset number of training rounds; Weighted average the local network weights, and introduce regularization constraints on the difference between the local network weights and the global network weights to be updated during the weighted averaging process to obtain the updated global network weights; Step 6: Based on the updated global network weights and combined with the dynamic adjustment coefficients, execute the routing and forwarding decision of inter-satellite data packets to complete the interaction of IoT data in the low-Earth orbit satellite communication network.

2. The IoT low-orbit satellite communication data interaction generation method according to claim 1, characterized in that, The network state information of each satellite node in the low-Earth orbit (LEO) satellite constellation is collected and transformed into a spatially distributed feature point set. A state feature field is constructed based on the feature point set. This process includes: the ground station acquiring the network state information of each satellite node in the LEO satellite constellation, which includes the instantaneous three-dimensional position coordinates of the satellite node, the signal-to-noise ratio (SNR) of the link between the satellite node and its neighboring satellite nodes, the queue occupancy rate of the satellite node, and the total energy consumption of the satellite node per unit time; mapping each satellite node to a feature point in three-dimensional space, using the instantaneous three-dimensional position coordinates of the satellite node as the spatial coordinates of the feature point, and the SNR, queue occupancy rate, and total energy consumption as the attribute values ​​of the feature point; the feature points corresponding to all satellite nodes together constitute a spatially distributed feature point set; for any position in three-dimensional space, the spatial distance between any position and each feature point in the feature point set is calculated, and the reciprocal of the square of the spatial distance is used as the weight to perform a weighted summation of the attribute values ​​of all feature points in the feature point set to obtain the state feature value at any position; after traversing all positions in three-dimensional space, the state feature field is obtained.

3. The IoT low-orbit satellite communication data interaction generation method according to claim 2, characterized in that, A baseline region is determined in the state feature field, and the baseline region is divided into multiple feature subdomains based on the density of feature points and the gradient of link state changes. The feature point set is mapped to the feature subdomains to generate an aggregated information block. This includes: determining a baseline region in the state feature field with the distribution range of the spatial coordinates of all feature points as the boundary; dividing the baseline region into multiple equally spaced grid cells, counting the number of feature points contained in each grid cell, and calculating the feature point density value at each position in the baseline region based on the number of feature points; calculating the spatial gradient value of the state feature value at each position in the baseline region based on the state feature value at each position in the state feature field; weighting and fusing the feature point density value and the gradient value to obtain a comprehensive partitioning index value at each position, using the local maximum point of the comprehensive partitioning index value as the partitioning boundary point, and connecting all partitioning boundary points to obtain multiple feature subdomains; and assigning each feature point to the corresponding feature subdomain based on the spatial coordinates of each feature point, so that a group of feature points with similar spatial positions are aggregated in each feature subdomain. Each feature subdomain and the feature points aggregated within it together constitute an aggregated information block.

4. The IoT low-orbit satellite communication data interaction generation method according to claim 3, characterized in that, Statistical analysis is performed on feature points within the aggregated information block to extract feature parameters. Dynamic adjustment coefficients are then generated based on these parameters. This process includes: for each aggregated information block, obtaining all feature points contained within it; calculating the geometric moments of each order of feature point distribution within the aggregated information block based on the spatial coordinates of all feature points; calculating the central moments of each order based on these geometric moments; normalizing the central moments to obtain normalized central moments; and calculating seven invariant moment values ​​based on the fixed combination relationship of the normalized central moments; and calculating the average signal-to-noise ratio and queue occupancy of all feature points within the aggregated information block based on their attribute values. The average value of the rate and the average value of the total energy consumption are used to obtain the average signal-to-noise ratio, average queue occupancy rate, and average energy consumption. The seven invariant moment values, the average signal-to-noise ratio, the average queue occupancy rate, and the average energy consumption are combined into a feature parameter vector, which serves as the feature parameters of the aggregated information block. Each component in the feature parameter vector is normalized to obtain the normalized shape feature index, signal-to-noise ratio index, queue occupancy rate index, and energy consumption index. The normalized shape feature index, signal-to-noise ratio index, queue occupancy rate index, and energy consumption index are weighted and summed to obtain the dynamic adjustment coefficient corresponding to the aggregated information block.

5. The IoT low-orbit satellite communication data interaction generation method according to claim 4, characterized in that, The average connectivity of satellite nodes is corrected based on a dynamic adjustment coefficient, and the low-Earth orbit satellite constellation is dynamically grouped to generate multiple satellite groups. This includes: calculating the average number of communication links established between a satellite node and other surrounding satellite nodes within a preset historical period based on the dynamic adjustment coefficient, using this average number as the initial average connectivity of the satellite node; multiplying the dynamic adjustment coefficient by the initial average connectivity to obtain the corrected average connectivity of the satellite node; sorting all satellite nodes in descending order based on the corrected average connectivity of each satellite node to obtain a sorted node sequence; and sequentially selecting nodes from the sorted node sequence. If the previously selected node has not yet been assigned to any satellite group, then the node is set as the new group center node. Starting from the new group center node, a search is conducted in three-dimensional space for surrounding satellite nodes whose spatial distance from the new group center node is less than a preset distance threshold and whose link signal-to-noise ratio with the new group center node is higher than a preset signal-to-noise ratio threshold. The searched surrounding satellite nodes and the new group center node are aggregated into a satellite group. For the remaining satellite nodes that have not yet been assigned to any satellite group, the node selection and group aggregation process is repeated until all satellite nodes are assigned to the corresponding satellite groups, generating multiple satellite groups.

6. The IoT low-orbit satellite communication data interaction generation method according to claim 5, characterized in that, Step 5 includes: within each satellite group, comparing all corrected average connectivity values ​​in the satellite group, selecting the satellite node with the highest corrected average connectivity value as the group management node, and designating the remaining satellite nodes in the satellite group other than the group management node as member satellites; the group management node initializes a global deep neural network model and distributes the global network weights of the global deep neural network model to each member satellite in the satellite group; after receiving the global network weights, each member satellite trains its local model based on its own network state information and dynamic adjustment coefficients, and obtains local network weights after reaching a preset number of training rounds; the local network weights are sent to the group management node; after receiving the local network weights, the group management node calculates the aggregate weights corresponding to each member satellite according to the dynamic adjustment coefficients, multiplies the local network weights by the corresponding aggregate weights, and sums them to obtain an initial aggregate result with a weighted average; based on the initial aggregate result, the difference between the local network weights and the global network weights is calculated, the difference is multiplied by a preset regularization coefficient to obtain the regularization adjustment term for each member satellite, and the regularization adjustment terms of all member satellites are summed and superimposed on the initial aggregate result to obtain the updated global network weights.

7. The IoT low-orbit satellite communication data interaction generation method according to claim 6, characterized in that, Based on the updated global network weights and combined with dynamic adjustment coefficients, the routing and forwarding decisions for inter-satellite data packets are executed to complete the interaction of IoT data in the low-Earth orbit satellite communication network. This includes: each satellite node loading the received updated global network weights into its local deep neural network to obtain a local routing decision network; for data forwarding, after receiving an IoT data packet to be forwarded, the satellite node generates a state vector based on the destination satellite node identifier and the link state information of all neighboring satellite nodes; the state vector is input into the routing decision network to calculate the action value of each neighboring satellite node; based on the action value and the dynamic adjustment coefficient corresponding to the aggregation information block to which the satellite node belongs, a neighboring satellite node is determined as the next-hop satellite node from all neighboring satellite nodes as the exploration probability; the IoT data packet is forwarded to the next-hop satellite node, and upon receiving it, the next-hop satellite node compares the destination satellite node identifier with its own identifier; if they match, data reception is completed; if they do not match, the next-hop satellite node is used as the new forwarding satellite node, and the data packet forwarding and judgment process is repeated until the data packet reaches the destination satellite node, completing the interaction of IoT data in the low-Earth orbit satellite communication network.

8. An Internet of Things (IoT) low-Earth orbit satellite communication data interaction generation system, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to collect network status information of each satellite node in the low-Earth orbit satellite constellation, and convert it into a spatially distributed feature point set, and construct a state feature field based on the feature point set; The partitioning module is used to determine a baseline region in the state feature field and divide the baseline region into multiple feature subdomains based on the density of feature points and the gradient of link state changes. The feature point set is mapped to the feature subdomains to generate an aggregated information block. The analysis module is used to perform statistical analysis on the feature points in the aggregated information block, extract feature parameters, and generate dynamic adjustment coefficients based on the feature parameters. The processing module is used to correct the average connectivity of satellite nodes according to the dynamic adjustment coefficient and to dynamically divide the low-Earth orbit satellite constellation into multiple satellite groups. The training module is used to select one satellite as the group management node and the remaining satellites as member satellites in each satellite group, and to build a collaborative deep reinforcement learning framework. The member satellites train the local model based on network state information and dynamic adjustment coefficients, and obtain the local network weights after reaching a preset number of training rounds. The local network weights are weighted and averaged. During the weighting and averaging process, a regularization constraint is introduced on the difference between the local network weights and the global network weights to be updated, so as to obtain the updated global network weights. The routing module is used to make routing and forwarding decisions for inter-satellite data packets based on the updated global network weights and combined with dynamic adjustment coefficients, thereby enabling the interaction of IoT data in the low-Earth orbit satellite communication network.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images