High-low orbit satellite fusion communication scheduling method for emergency scenarios

Abstract

The application relates to the technical field of satellite communication, in particular to a high-low orbit satellite fusion communication scheduling method for an emergency scene, which comprises the following steps: acquiring terminal pose, masking profile, meteorological and ephemeris data; respectively calculating a first timestamp generated by a terrain entity block and a second timestamp caused by link margin exhaustion due to dynamic plume aerosol attenuation, taking the minimum value of the two as a predicted switching timestamp and delimiting a rolling time domain window; subsequently, taking physical masking as a topological constraint, weighting aerosol attenuation cost and transmission delay cost to construct a high-low orbit joint graph model; finally, outputting target satellite identifiers and antenna pointing data through algorithm optimization, and driving the antenna to be redirected in advance for access. The application realizes satellite network switching in an emergency environment through link breakage early warning and cost optimization.

Description

Technical Field

[0001] This invention relates to the field of satellite communication technology, specifically to a high- and low-orbit satellite fusion communication scheduling method for emergency scenarios. Background Technology

[0002] With the rapid development of satellite communication technology, heterogeneous communication networks integrating low-Earth orbit (LEO) and high-Earth orbit (HEO) satellites have become an indispensable means of communication support in emergency rescue scenarios for natural disasters such as forest fires and earthquakes. Among them, LEO satellites are the preferred access layer due to their low orbital altitude and low transmission latency, while HEO satellites are the backbone backup layer due to their wide coverage and strong resilience.

[0003] Existing satellite network scheduling and handover technologies typically rely on satellite ephemeris data to predict satellite transit trajectories and visibility windows. Some scheduling systems also incorporate basic terrain elevation data, using geometric line-of-sight calculations to determine whether macroscopic terrain features such as mountains will obstruct communication terminals. When the currently serving satellite moves to the edge of the line-of-sight range, or when the terminal detects that the received signal strength indicator is below a set threshold, the system triggers a network handover process, migrating services to the next visible satellite.

[0004] However, in extreme emergency rescue scenarios such as forest fires, the communication environment exhibits extremely high dual dynamism: on the one hand, there are high-frequency changes in the topology caused by the high-speed motion of low-orbit satellites; on the other hand, there is the rapidly evolving physical environment at the fire site. Objectively, emergency sites not only have static physical obstructions such as mountain terrain, but also a large amount of dynamically drifting dense smoke aerosols. High concentrations of aerosol particles and complex wind convection will cause strong scattering and absorption of microwave radio frequency signals, leading to a nonlinear and rapid attenuation of available link margin.

[0005] Existing scheduling mechanisms primarily rely on predictions based on ideal spatial geometric line-of-sight or passive link reconnection after signal quality deterioration. When faced with complex scenarios involving the coupling of high-speed satellite movement and dynamic evolution of environmental media such as aerosols, relying solely on geometrical visual judgments or single signal threshold triggers fails to accurately reflect the dynamic availability window of the underlying radio frequency link under multidimensional environmental interference. This often leads to discrepancies between predicted link availability and actual availability, resulting in communication interruptions during signaling handshakes and handovers, impacting the continuity of disaster relief and rescue operations.

[0006] In summary, how to achieve seamless switching and scheduling of communication networks in extreme emergency environments with rapidly changing topologies and dynamic multidimensional physical media interference (such as the combination of terrain obstruction and aerosol attenuation) is a technical problem that urgently needs to be solved in this field.

[0007] To address this, a high- and low-orbit satellite fusion communication scheduling method for emergency scenarios is proposed. Summary of the Invention

[0008] The purpose of this invention is to provide a high-low orbit satellite fusion communication scheduling method for emergency scenarios, enabling satellite network switching in emergency environments through link failure early warning and cost optimization. This includes acquiring terminal pose, masking contour, meteorological, and ephemeris data; calculating a first timestamp caused by terrain obstruction and a second timestamp caused by dynamic plume aerosol attenuation leading to link margin depletion, taking the minimum of the two as the predicted switching timestamp and defining a rolling time-domain window; subsequently, using physical masking as a topological constraint, constructing a high-low orbit joint graph model by weighting aerosol attenuation cost and transmission delay cost; and finally, outputting target satellite identifiers and antenna pointing data through algorithmic optimization to drive early antenna redirection for access.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A high- and low-Earth orbit satellite integrated communication scheduling method for emergency scenarios includes: Acquire terminal pose data, masking contour data, and meteorological feature data; acquire real-time ephemeris data of serving low-Earth orbit satellites, candidate low-Earth orbit satellites, and high-Earth orbit backbone satellites. Line-of-sight occlusion calculations are performed based on terminal pose data, masking contour data, and real-time ephemeris data of the serving LEO satellites to generate a first timestamp; meteorological feature data are input into the plume model to generate plume concentration field data, and the predicted link attenuation is calculated by combining the real-time ephemeris data of the serving LEO satellites. The time node corresponding to when the predicted link attenuation reaches the threshold is extracted to generate a second timestamp. Extract the minimum value between the first and second timestamps as the predicted switching timestamp; set a rolling time-domain window that includes the predicted switching timestamp, and use the masking contour data and plume concentration field data as topology and link cost constraints to fuse real-time ephemeris data of candidate low-Earth orbit satellites and high-Earth orbit backbone satellites to construct a high-low orbit map model. In the high and low orbit map model, an optimization algorithm is used to find the target satellite identifier and antenna pointing data with the minimum link cost. Based on the target satellite identifier and antenna pointing data, the terminal communication link is redirected to access the corresponding target candidate low orbit satellite or high orbit backbone satellite.

[0010] Preferably, the process of acquiring terminal pose data, masking contour data, meteorological feature data, and real-time ephemeris data includes: acquiring the three-dimensional position coordinates and velocity vector calculated by the emergency response terminal based on the satellite positioning module, and combining them with the attitude pitch angle data collected by the inertial measurement unit to generate the terminal pose data; receiving the three-dimensional entity boundary matrix scanned and sent by the edge computing node through a penetrating detection device as the masking contour data, wherein the penetrating detection device adopts a synthetic aperture radar mounted on an unmanned aerial vehicle platform; extracting real-time wind field velocity data and aerosol concentration data collected by environmental monitoring equipment deployed in the emergency area as the meteorological feature data; parsing the satellite system broadcast ephemeris message received by the emergency response terminal, and extracting the real-time ephemeris data of the serving low-Earth orbit satellites, candidate low-Earth orbit satellites, and high-Earth orbit backbone satellites.

[0011] Preferably, the process of generating the first timestamp includes: constructing a time-varying line-of-sight spatial vector from the emergency terminal to the serving low-Earth orbit satellite based on the three-dimensional position coordinates in the terminal pose data and the real-time ephemeris data of the serving low-Earth orbit satellite; performing a terminal antenna coordinate system transformation on the time-varying line-of-sight spatial vector by combining the attitude pitch angle data in the terminal pose data to generate a radio frequency link pointing baseline; traversing and comparing the radio frequency link pointing baseline with the three-dimensional entity boundary matrix in the masking contour data within a set time step to calculate the spatial geometric intersection state of the radio frequency link pointing baseline being blocked by the entity terrain; extracting the time node parameter of the first spatial intersection between the radio frequency link pointing baseline and the three-dimensional entity boundary matrix, and generating the time node parameter as the first timestamp.

[0012] Preferably, the plume model includes: an input feature layer that receives the wind field velocity data and the aerosol concentration data, and combines the masking contour data to construct an initial spatiotemporal feature map containing the topological relationships of three-dimensional spatial nodes and edges; The spatial graph convolutional layer receives the initial spatiotemporal feature map, uses the physical masking contour as the blocking boundary, and performs spatial convolution aggregation on the aerosol concentration features between adjacent nodes according to the wind field velocity direction to generate a spatial diffusion feature tensor. The time series extraction layer receives the spatial diffusion feature tensor, performs time series memory calculations along a set time window step, and extracts the dynamic evolution features of aerosol concentration over time. The output mapping layer receives dynamic evolution features and outputs a three-dimensional aerosol concentration distribution matrix containing multiple consecutive time nodes within the future prediction period. The three-dimensional aerosol concentration distribution matrix is ​​used as the plume concentration field data.

[0013] Preferably, the process of generating the second timestamp includes: constructing a dynamic line-of-sight link trajectory that changes over time based on the terminal pose data and the real-time ephemeris data of the serving low-Earth orbit satellite; spatially matching the dynamic line-of-sight link trajectory with plume concentration field data to extract aerosol concentration parameters distributed along the dynamic line-of-sight link trajectory; using the aerosol concentration parameters as a search index to search in a preset environment and radio frequency attenuation mapping table to obtain the predicted link attenuation for each time series; comparing the predicted link attenuation with a preset link margin depletion threshold to extract the time node when the predicted link attenuation first reaches the link margin depletion threshold, and generating the second timestamp.

[0014] Preferably, the construction process of the high-low orbit graph model includes: comparing a first timestamp and a second timestamp, and determining the minimum of the two as the predicted handover timestamp; dividing and generating a rolling time domain window starting from the current system time and ending at the predicted handover timestamp plus a preset handover signaling interaction margin duration; within the rolling time domain window, mapping the terminal, each candidate low-orbit satellite, and the high-orbit backbone satellite as communication nodes in the graph model, and constructing initial dynamic directed edges based on the radio frequency transceiver payload matching relationship between nodes; and mapping the masking contour data and each of the dynamic directed edges... Spatial collision detection is performed on edges, and the topological connectivity of dynamic directed edges that physically block the masking contour data is marked as disconnected, forming topological constraints. The aerosol concentration integral value of the path traversed by the undisconnected dynamic directed edges in the plume concentration field data is extracted and converted into radio frequency attenuation penalty cost. The spatial transmission delay between each communication node is extracted as the baseline transmission cost. The baseline transmission cost and the radio frequency attenuation penalty cost are weighted and summed to calculate the comprehensive link cost weight of each dynamic directed edge, forming link cost constraints and generating a high- and low-orbit joint graph model.

[0015] Preferably, the process of obtaining the target satellite identifier and antenna pointing data includes: using the node mapped by the terminal in the high and low orbit map model as the source point, and using the available satellite communication nodes within the rolling time domain window as the traversal space, running a path optimization algorithm to calculate the cumulative link cost from the source point to each available satellite communication node; comparing and obtaining the optimal communication path where the cumulative link cost reaches a minimum value, and extracting the satellite node pointed to by the optimal communication path as the target satellite identifier; retrieving the corresponding real-time satellite ephemeris data according to the target satellite identifier, performing a three-dimensional spatial coordinate system transformation and calculation on the real-time satellite ephemeris data and the terminal pose data, and calculating the azimuth and elevation angles of the phased array antenna driving the terminal as antenna pointing data.

[0016] Preferably, the process of redirecting access to the corresponding target candidate low-Earth orbit (LEO) satellite or high-Earth orbit (HEO) backbone satellite includes: parsing the network hierarchy attribute of the target satellite identifier to determine its orbital type; when the target satellite identifier belongs to the candidate LEO satellite, driving the phased array antenna of the terminal to activate beam pre-synchronization and establish a LEO relay link before the predicted switching timestamp arrives, based on the antenna pointing data, and switching the terminal's data stream to be transmitted to the target candidate LEO satellite; when the target satellite identifier belongs to the HEO backbone satellite, driving the phased array antenna of the terminal to switch to a preset anti-attenuation communication frequency band and establish a cross-orbit direct link, based on the antenna pointing data, and redirecting the terminal's data stream to be transmitted to the HEO backbone satellite.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This application establishes a link failure early warning mechanism under multi-dimensional physical environment coupling by extracting the minimum value of the first hard obstruction timestamp caused by line-of-sight blocking and the second soft attenuation timestamp caused by aerosol accumulation depleting link margin as the predictive handover benchmark. Compared with the traditional passive scheduling strategy that relies solely on the satellite's geometric line-of-sight angle, this mechanism can accurately capture the earliest communication failure node caused by the superposition of mountain obstruction and dynamic plume evolution at the emergency site. This dual timestamp competition calculation method effectively avoids unpredictable disconnections caused by sudden environmental interference, providing a high-confidence time-domain decision reference for subsequent signaling interaction and network forward-looking handover.

[0018] 2. By taking the topological disconnect caused by the masking contour as a hard constraint and by weighting and summing the attenuation cost of plume concentration mapping with the spatial transmission delay to obtain the comprehensive link weight, this application constructs a joint time-varying graph model for heterogeneous satellite homogenization. In the path optimization process, the system can adaptively seek the optimal balance between "low latency requirement for low orbit" and "high anti-interference requirement for high orbit" based on the comprehensive cost, realizing the global planning and efficient allocation of cross-orbit space network resources under dynamic constraints.

[0019] 3. By outputting target satellite identifiers and antenna pointing data within a time-domain window, and driving phased array antennas to perform differentiated pre-synchronization or cross-orbit direct connection operations based on the target's orbital type, this application achieves a complete control closed loop from network layer algorithm optimization to physical layer RF hardware execution. Before the predicted switching node arrives, the system can pre-calculate spatial coordinates and schedule antennas to complete beam pointing, realizing a relay communication logic of establishing communication before disconnection. Adaptive frequency band adjustments and data stream redirection strategies are implemented for satellite targets at different levels, ensuring the smooth continuity of emergency rescue operations. Attached Figure Description

[0020] Figure 1A schematic diagram of the high and low orbit satellite fusion communication scheduling method for emergency scenarios provided by the present invention; Figure 2 This is a schematic diagram of the plume model process provided by the present invention; Figure 3 This is a schematic diagram illustrating the construction process of the high-low orbit map model provided by the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figures 1 to 3 This invention provides a high- and low-orbit satellite fusion communication scheduling method for emergency scenarios, the technical solution of which is as follows: To achieve the above objectives, the present invention provides the following technical solution: A high- and low-Earth orbit satellite integrated communication scheduling method for emergency scenarios, the specific process of which is as follows: Figure 1 As shown, it includes: Acquire terminal pose data, masking contour data, and meteorological feature data; acquire real-time ephemeris data of serving low-Earth orbit satellites, candidate low-Earth orbit satellites, and high-Earth orbit backbone satellites. Line-of-sight occlusion calculations are performed based on terminal pose data, masking contour data, and real-time ephemeris data of the serving LEO satellites to generate a first timestamp; meteorological feature data are input into the plume model to generate plume concentration field data, and the predicted link attenuation is calculated by combining the real-time ephemeris data of the serving LEO satellites. The time node corresponding to when the predicted link attenuation reaches the threshold is extracted to generate a second timestamp. Extract the minimum value between the first and second timestamps as the predicted switching timestamp; set a rolling time-domain window that includes the predicted switching timestamp, and use the masking contour data and plume concentration field data as topology and link cost constraints to fuse real-time ephemeris data of candidate low-Earth orbit satellites and high-Earth orbit backbone satellites to construct a high-low orbit map model. In the high and low orbit map model, an optimization algorithm is used to find the target satellite identifier and antenna pointing data with the minimum link cost. Based on the target satellite identifier and antenna pointing data, the terminal communication link is redirected to access the corresponding target candidate low orbit satellite or high orbit backbone satellite.

[0023] Example 1: Furthermore, the process of acquiring terminal pose data, masking contour data, meteorological feature data, and real-time ephemeris data includes: acquiring the three-dimensional position coordinates and velocity vectors calculated by the emergency response terminal based on the satellite positioning module, and combining them with the attitude pitch angle data collected by the inertial measurement unit to generate the terminal pose data; receiving the three-dimensional entity boundary matrix scanned and sent by the edge computing node through a penetrating detection device as the masking contour data, wherein the penetrating detection device adopts a synthetic aperture radar mounted on an unmanned aerial vehicle platform; extracting real-time wind field velocity data and aerosol concentration data collected by environmental monitoring equipment deployed in the emergency area as the meteorological feature data; parsing the satellite system broadcast ephemeris messages received by the emergency response terminal, and extracting the real-time ephemeris data of the serving low-Earth orbit satellites, candidate low-Earth orbit satellites, and high-Earth orbit backbone satellites.

[0024] First, the emergency response terminal receives spatial positioning signals through its built-in Global Navigation Satellite System (GNSS) positioning module, calculates and obtains its current three-dimensional position coordinates (longitude, latitude, and elevation), as well as the velocity vector composed of its direction of motion and speed. Simultaneously, the terminal collects real-time spatial attitude information of the device body through its built-in inertial measurement unit (IMU), including the current pitch, roll, and yaw angles. Based on a unified system clock, the system timestamps and calibrates the calculated three-dimensional position coordinates, velocity vector, and attitude angle data using a unified system clock. The aligned parameters are then encapsulated into a unified data structure to generate real-time terminal attitude data. Specifically, the spatial reference system calibration involves: constructing a local rectangular coordinate system (northeast-sky) based on the local ground plane of the terminal's location; projecting and transforming the longitude, latitude, and elevation coordinates to the origin of this local rectangular coordinate system; and using the three orthogonal axes of this coordinate system as references, performing a rotation matrix mapping on the attitude angles output by the IMU, thus unifying the terminal's absolute position coordinates and the relative attitude of the antenna pointing to the same three-dimensional spatial reference system.

[0025] Secondly, above the emergency site, a drone platform equipped with a synthetic aperture radar (SAR) performs flight scanning detection along a preset route. The SAR emits penetrating microwave pulses towards the ground and receives echo signals from the ground and vegetation after penetrating the smoke. Edge computing nodes deployed on-site receive these echo signals and use radar imaging algorithms to resolve them into a three-dimensional spatial point cloud containing terrain undulations and the physical height of obstacles. Specifically, the edge computing nodes receive the echo signals and use range-Doppler imaging algorithms to resolve them into a three-dimensional spatial point cloud containing terrain undulations and the physical height of obstacles. Specifically, the edge computing nodes first perform range pulse compression processing on the received raw radar echo data, using matched filtering technology to compress the linear frequency modulated pulse signal into a narrow pulse to improve range resolution. Then, they perform azimuth Fourier transform on the range-compressed data, completing range migration correction in the range-Doppler domain to eliminate the range-azimuth coupling effect caused by the drone platform's movement. Finally, azimuth compression is completed through azimuth matched filtering to generate a two-dimensional SAR complex image. The elevation information of the three-dimensional spatial point cloud is obtained by interferometry of two complex SAR images acquired by repeated scanning of adjacent flight paths by UAVs. After registration of the two complex images, the interferometric phase difference is calculated. Then, the interferometric phase is converted into terrain elevation values ​​through phase unwrapping and elevation inversion formulas. Finally, the two-dimensional geographic location coordinates and elevation values ​​are fused to generate the three-dimensional spatial point cloud. The aforementioned SAR imaging and interferometric processing flow employs mature, publicly available algorithms. Edge computing nodes can be deployed based on open-source SAR processing libraries (such as ISCE or SNAP). The synthetic aperture radar operates in the P-band (center frequency approximately 430MHz) or L-band (center frequency approximately 1.25GHz). The wavelength range of this band is 0.24m to 0.7m, significantly larger than the particle size of fire smoke aerosols (typically 0.1μm to 10μm), differing by approximately 4 to 5 orders of magnitude. This satisfies the Rayleigh scattering condition, making the scattering attenuation of the radar signal in this band negligible, thus enabling effective penetration detection of smoke at the scene. This frequency band selection is a prerequisite for the synthetic aperture radar in this scheme to effectively acquire terrain masking data in dense fire smoke environments. Subsequently, the edge computing nodes perform meshing on the 3D spatial point cloud, constructing a 3D entity boundary matrix representing the spatial connectivity state and entity blocking boundaries. The emergency response terminal receives this matrix through a local ad hoc network link and uses it as masking contour data. During the meshing process, the edge computing nodes divide the detection airspace into multiple uniform 3D voxel grids according to a preset spatial resolution. Then, the number of radar echo point clouds contained in each 3D voxel grid is counted. If the number of point clouds in a certain grid exceeds a preset entity density threshold, the attribute of the corresponding element in the 3D entity boundary matrix of that grid is assigned the value of blocking state; otherwise, it is assigned the value of connected state. This transforms the disordered discrete point cloud into a 3D binary spatial matrix that can be used by the computer for link geometric tracing.

[0026] Subsequently, various IoT environmental monitoring devices deployed throughout the emergency area continuously collected physical parameters of the on-site environment using their onboard anemometers and particulate sensors. The wireless receiving port of the emergency response terminal continuously listened to the environmental telemetry data frames transmitted by these monitoring devices. The system unpacked the intercepted telemetry data frames and separated the payload, directly extracting the real-time wind velocity data (current wind speed and direction) and aerosol concentration data (density of suspended particulate matter per unit volume) encapsulated within the payload. These two types of physical quantities were combined as meteorological characteristic data. During the combination process, the system extracted the deployment location identifiers of each IoT environmental monitoring device and bound the intercepted wind velocity data and aerosol concentration data to their corresponding geographical coordinates. Then, using the terminal's location as the center, a spatial interpolation algorithm was used to combine the aforementioned data with geographical coordinates. Discrete meteorological monitoring point data are mapped and expanded into a gridded meteorological feature distribution map covering the current emergency area. Specifically, for each grid point of the target grid, all meteorological monitoring points within its surrounding radius R (500m in this embodiment) are searched. The wind velocity data and aerosol concentration data of each monitoring point are weighted by the inverse square of the Euclidean distance from each monitoring point to the grid point. The inverse power of the distance is set to 2 to balance the smoothness of interpolation and the preservation of local features. For grid points without any monitoring points within the radius R, their meteorological feature values ​​are marked as missing and inherit the values ​​of the nearest available grid point.

[0027] Finally, the radio frequency receiving channel of the emergency terminal continuously receives downlink broadcast signals from the satellite communication system. The baseband processing procedure demodulates and decodes the downlink signal to restore the complete system broadcast ephemeris message data stream. According to the frame structure definition of the satellite communication protocol, the system removes the message header and checksum, locates and reads the ephemeris data block, and then determines the orbital attributes of each satellite based on the extracted six root parameters: First, it reads the semi-major axis parameter to distinguish between high and low Earth orbit types. High-altitude satellites determined to be in geostationary orbit are archived as real-time ephemeris data of high-orbit backbone satellites covering the region. For satellites determined to be in low Earth orbit, the system calculates the spatial elevation angle based on the currently generated terminal pose data. Targets with elevation angles within the current service connection range are archived as real-time ephemeris data of serving low Earth orbit satellites, while targets with elevation angles below the current connection threshold but within the preliminary visibility range are archived as real-time ephemeris data of candidate low Earth orbit satellites. From these, the six root parameters and time perturbation parameters characterizing the satellite's spatial orbit are extracted one by one and archived as real-time ephemeris data of serving low Earth orbit satellites currently providing connections, candidate low Earth orbit satellites within the preliminary visibility range, and high-orbit backbone satellites covering the region.

[0028] By employing a multi-sensor fusion positioning, a multi-source physical perception mechanism that combines UAV-mounted synthetic aperture radar and IoT monitoring devices, this application tightly anchors the macroscopic scheduling algorithm to the precise underlying hardware entity, ensuring the real-time acquisition of terminal pose, physical terrain contours, and dynamic meteorological parameters in extreme environments.

[0029] Further, the process of generating the first timestamp includes: constructing a time-varying line-of-sight spatial vector from the emergency terminal to the serving low-Earth orbit satellite based on the three-dimensional position coordinates in the terminal pose data and the real-time ephemeris data of the serving low-Earth orbit satellite; performing a terminal antenna coordinate system transformation on the time-varying line-of-sight spatial vector by combining the attitude pitch angle data in the terminal pose data to generate a radio frequency link pointing baseline; traversing and comparing the radio frequency link pointing baseline with the three-dimensional entity boundary matrix in the masking contour data within a set time step, calculating the spatial geometric intersection state of the radio frequency link pointing baseline being blocked by the entity terrain; extracting the time node parameter of the first spatial intersection between the radio frequency link pointing baseline and the three-dimensional entity boundary matrix, and generating the time node parameter as the first timestamp.

[0030] First, the three-dimensional position coordinates of the current moment are extracted from the terminal pose data as a fixed ray starting point. At the same time, the real-time ephemeris data of the serving low-Earth orbit satellite is analyzed, and the orbit extrapolation algorithm is used to calculate the set of coordinate points of the satellite's trajectory over a continuous period of time in the future. Simultaneously, the system extracts the velocity vector from the terminal pose data. Using the initial ray starting point as a reference, kinematic extrapolation is performed on the future three-dimensional position of the terminal according to the set time step, generating a set of predicted coordinate points of the terminal that correspond one-to-one with the satellite's trajectory in the time series. Subsequently, in a unified Earth spatial rectangular coordinate system, the terminal position coordinates as the starting point are connected one by one with the coordinate points of each satellite on the predicted trajectory, and the direction vector is calculated. This generates a time-varying line-of-sight spatial vector that changes spatial direction over time, pointing directly from the emergency terminal to the serving low-Earth orbit satellite.

[0031] Secondly, since the emergency response terminal inevitably tilts during actual use or movement, the system extracts the pitch angle and related yaw and roll data from the terminal's pose data, constructs a three-dimensional spatial rotation mapping matrix, and projects the time-varying line-of-sight spatial vector generated in the absolute geographic coordinate system onto the relative coordinate system with the terminal's physical antenna panel as the origin through this rotation mapping matrix. After attitude rotation correction, this vector accurately corresponds to the beam center normal direction of the electromagnetic wave actually emitted by the antenna array, thereby generating a true radio frequency link pointing reference line.

[0032] Next, a tiny time step is set as the progressive unit of the loop traversal. At each time step, the system projects the radio frequency link pointing baseline corresponding to the current moment onto the three-dimensional space where the masking contour data is located. Specifically, since the three-dimensional entity boundary matrix is ​​constructed under an absolute geospatial reference system, the system calls the inverse transformation matrix of the three-dimensional space rotation mapping matrix constructed in the previous steps, and combines it with the real-time three-dimensional position translation coordinates of the terminal to inversely transform and map the radio frequency link pointing baseline in the relative coordinate system to the absolute geospatial rectangular coordinate system where the three-dimensional entity boundary matrix is ​​located, so as to ensure that the ray pointing is strictly aligned with the coordinate scale of the physical terrain. Employing 3D spatial ray tracing logic, the system sequentially retrieves voxel grids within the 3D entity boundary matrix penetrated by the ray along the extension direction of the RF link pointing to the baseline. A 3D digital differential analysis algorithm is introduced, and based on the spatial grid resolution of the 3D entity boundary matrix, the continuous RF link pointing to the baseline is discretized into a series of spatially adjacent 3D grid coordinate index sequences. The system then retrieves the corresponding 3D voxel grids according to the order of these coordinate index sequences. By checking the attribute labels of each grid along the ray's path, it determines whether the RF link pointing to the baseline passes through entity terrain grids marked as obstructed, thereby determining the link's spatial geometric intersection state at the current time point.

[0033] Finally, following the forward sequence of the timeline, the above-mentioned ray tracing and mesh comparison process, with the set time step as the unit, is continuously executed. When, at a certain future time node, it is detected that the attribute of a certain voxel mesh along the path of the radio frequency link pointing to the baseline changes abruptly from a connected state to an obstructed state for the first time during the extension process, it is determined that the range link is physically cut off by the boundary of a physical entity such as a mountain. The system then terminates the comparison loop, extracts and records the exact time node parameter at the time of this first spatial intersection, directly assigns the time node parameter and outputs it as the first timestamp representing the end of the satellite's physical visibility window.

[0034] By incorporating the three-dimensional coordinates and pitch angle from the terminal pose into time-varying line-of-sight vector calculations, and establishing a ray-tracing comparison mechanism between the RF link pointing baseline and the three-dimensional entity boundary matrix, the transformation of communication line-of-sight obstruction prediction from ideal geometry to actual physical dynamics is realized. It can correct beam pointing deviations caused by terminal tilt in real time, and by performing high-frequency traversal intersection determination in the voxelized masking contour, it captures the precise instantaneous nature of terrain obstruction, thus providing a quantitative and reliable time-based basis for determining the physical visibility window of low-Earth orbit satellites constrained by terrain in emergency scenarios.

[0035] Furthermore, the plume model includes: an input feature layer that receives the wind field velocity data and the aerosol concentration data, and combines the masking contour data to construct an initial spatiotemporal feature map containing the topological relationships of three-dimensional spatial nodes and edges; The spatial graph convolutional layer receives the initial spatiotemporal feature map, uses the physical masking contour as the blocking boundary, and performs spatial convolution aggregation on the aerosol concentration features between adjacent nodes according to the wind field velocity direction to generate a spatial diffusion feature tensor. The time series extraction layer receives the spatial diffusion feature tensor, performs time series memory calculations along a set time window step, and extracts the dynamic evolution features of aerosol concentration over time. The output mapping layer receives dynamic evolution features and outputs a three-dimensional aerosol concentration distribution matrix containing multiple consecutive time nodes within the future prediction period. This three-dimensional aerosol concentration distribution matrix is ​​then used as the plume concentration field data. The specific process is as follows: Figure 2 As shown.

[0036] In the operation process of the input feature layer, the physical space of the emergency site is first divided into multiple uniformly arranged three-dimensional voxel nodes according to the set three-dimensional spatial resolution to construct a basic spatial grid coordinate system. Then, the three-dimensional entity boundary attributes in the masking contour data are extracted, and voxel nodes inside the entity terrain blockage are removed or marked as unreachable. Spatial connectivity edges are established for the remaining adjacent voxel nodes without physical entity barriers, thereby constructing an initial graph topology structure with physical terrain constraints. Next, the timestamps are aligned, and the three-dimensional wind speed vector in the wind field velocity data and the concentration scalar value in the aerosol concentration data are assigned to the corresponding voxel nodes one by one according to the spatial coordinate mapping relationship, as the initial state attributes of each spatial node in the graph network. This generates an initial spatiotemporal feature map containing three-dimensional spatial nodes, edge topology relationships, and dynamic meteorological attributes.

[0037] During the processing of the spatial graph convolutional layer, the initial spatiotemporal feature map is received. The topological disconnect boundary constructed by the physical masking contour strictly restricts the information interaction between nodes. For any central node in the graph network, the system calculates the upwind and downwind relationships in fluid dynamics based on the wind velocity direction of that node and its connected adjacent nodes. According to the calculated wind velocity projection component onto the central node, asymmetric convolution weights are dynamically assigned to adjacent edges in different directions, i.e., giving upwind adjacent nodes higher information transmission weights. The dynamic assignment of asymmetric convolution weights to adjacent edges in different directions... The specific process of scaling is as follows: Extract the spatial direction vector formed by the central node and adjacent nodes; calculate the cosine of the spatial angle between this spatial direction vector and the current wind velocity vector of the central node; input this cosine of the spatial angle as an independent variable into a preset exponential scaling function, so that the output value when the angle represents the upwind direction is exponentially amplified, while the output value when the angle represents the downwind direction is attenuated; subsequently, input the scaled weight values ​​of adjacent nodes into a normalization function for processing, ensuring that the sum of the weights of all connected edges of the central node is 1. The specific mathematical expression of the exponential scaling function is: Where θ is the angle between the spatial direction vector between the central node and the adjacent nodes and the current wind velocity vector of the central node, and k is the temperature coefficient that controls the sensitivity of wind direction. In this embodiment, k is set to 2.0. The weight is maximized when θ = 0° (completely upwind) and minimized when θ = 180° (completely downwind). The normalized values ​​are used as the final asymmetric convolution weights. Based on the assigned weight parameters, aerosol concentration features from adjacent nodes are extracted and weighted, then subjected to a nonlinear mapping transformation using a modified linear unit activation function. Due to the inherent non-negativity of aerosol concentration in the physical environment, this modified linear unit activation function filters out negative noise features that may arise during the convolution weighted summation, retaining only positive concentration diffusion characteristics. The output is a spatial diffusion feature tensor that incorporates the directional driving effect of the wind field. The physical applicability of using graph convolutional networks to process aerosol diffusion in this embodiment is based on the following engineering premises: the spatial resolution of plume diffusion prediction must match the resolution of the constructed three-dimensional voxel mesh, and the maximum diffusion distance of aerosols within a single system scheduling cycle does not exceed the physical side length of a voxel. Under these premises, node-level graph convolutional aggregation can reasonably approximate the diffusion behavior of continuous fluids. For extreme weather scenarios where wind speeds are extremely high at emergency sites, causing aerosol diffusion distances to exceed multiple voxels in a single cycle, the system mainly uses the aforementioned multi-layer spatial graph convolutional layer stacking (such as the 3-layer cascaded structure in this embodiment) to achieve long-distance information transmission and approximate compensation. If the local wind speed is detected to exceed the extreme value threshold for stable convergence of the graph model, the system will automatically downgrade and load and use a Gaussian plume model based on pure physics as a backup prediction method.

[0038] In the computational steps of the time series extraction layer, spatial diffusion feature tensors generated under multiple consecutive time sections in the past are extracted according to the set time window step size. These tensors are then arranged in chronological order to construct a time series input sequence. Subsequently, a recurrent network structure with a memory mechanism (such as a long short-term memory network or a gated recurrent unit framework) is called to process the sequence frame by frame. In the computation of each time step, the system uses the spatial diffusion feature tensor of the current moment and the hidden state data transmitted in the previous moment as input. The forgetting logic inside the network is used to truncate historical feature signals that are no longer relevant, such as aerosol dissipation or sedimentation. At the same time, the newly added diffusion trend features of the current moment are absorbed into the backbone state sequence through the update logic. The hidden layer state is continuously updated iteratively, and finally the dynamic evolution features of aerosol concentration over time are extracted from the time series sequence. In this embodiment, the time series extraction layer adopts a two-layer bidirectional long short-term memory network with 128 hidden units per layer and a time window step size of 6 time sections, that is, the spatial diffusion feature tensors of the past 6 consecutive moments are used as the input sequence. The spatial graph convolutional layer adopts a 3-layer stacked structure, with each layer having a kernel size of 1×1 (node-level aggregation). The number of output channels is 32, 64, and 64 respectively. The number of nodes in the fully connected layer in the multi-layer regression decoding channel is set to the total number of voxel grids N. Then, there are 3 layers of 3D transposed convolutional layers, each with a kernel size of 3×3×3. The number of output channels is 64, 32, and 1 respectively, with a stride of 1 and the same padding method.

[0039] In the generation stage of the output mapping layer, the dynamic evolution feature matrix output from the time series extraction layer is received and imported into the multi-layer regression decoding channel. According to the set future prediction period length and the step size of consecutive time nodes, the dynamic evolution features are dimensionally expanded and reshaped into tensors. The multi-layer regression decoding channel includes cascaded fully connected mapping layers and 3D transposed convolutional layers. Specifically, firstly, the time-series extracted dynamic evolution features are mapped into high-dimensional feature vectors matching the initial voxel grid number through the fully connected mapping layer. Then, according to the set 3D spatial resolution rules, the high-dimensional feature vectors are reshaped into 3D feature blocks. Using continuously stacked 3D... The transposed convolutional layer performs progressive spatial upsampling and feature decoding on the 3D feature block, restoring it to the real 3D grid scale under the physical constraints of the physical terrain. Through multi-layer mapping calculation, the high-dimensional abstract temporal features are progressively restored and projected back into the 3D grid space with the same dimension as the initial input layer. On the output tensor slices corresponding to each prediction time node, the system calculates and restores the predicted aerosol concentration value of each voxel node at that future time, and reconstructs a 3D aerosol concentration distribution matrix containing multiple consecutive prediction time sections in the future. Finally, this continuous multi-dimensional matrix is ​​output as plume concentration field data characterizing the distribution of the decay medium in the future space environment.

[0040] Before the plume model is applied for real-time prediction, a model training process is performed, specifically including: The training sample set is constructed as follows: Based on the HYSPLIT or WRF-Chem numerical meteorological simulation platform, meteorological boundary conditions from historical fire or industrial accident scenarios are imported. Simulation data of a continuous 72-hour aerosol concentration field is generated with a time step of 5 minutes, resulting in at least 500 time-series samples. Each sample contains input features from 12 consecutive time sections and the corresponding real concentration distribution as labels. Historical meteorological and topographic data are input into the plume model as input features, and the predicted 3D aerosol concentration distribution matrix is ​​calculated. The joint loss function is defined as: ,in To predict the mean square error between the concentration matrix and the true matrix, The physical penalty term is calculated by extracting the sum of the absolute values ​​of the predicted concentrations on the voxel nodes marked as entity blocking attributes by the masked contours in the predicted concentration matrix, which forces the model to learn the physical boundary condition of 'zero concentration inside the entity'. The value is set to 0.1. Using the gradient descent optimization algorithm, backpropagation is performed based on the joint loss function to iteratively update the network node weight parameters in the spatial graph convolutional layer, the time series extraction layer, and the multi-layer regression decoding channel. After 5 consecutive iterations... The change is less than 10 -4 When convergence is reached, the training of the plume model is completed.

[0041] By constructing a plume model that combines spatial graph convolution and time series extraction, physical terrain masking is used as the blocking boundary, and spatiotemporal joint feature calculation of aerosol diffusion is realized based on wind field velocity. This enables the output of three-dimensional aerosol concentration distribution prediction results for future periods, thus providing a reliable data benchmark for the system to quantify dynamic radio frequency link attenuation, plan ahead, and trigger satellite network switching scheduling.

[0042] Further, the process of generating the second timestamp includes: constructing a dynamic line-of-sight link trajectory that changes over time based on the terminal pose data and the real-time ephemeris data of the serving low-Earth orbit satellite; spatially matching the dynamic line-of-sight link trajectory with plume concentration field data to extract aerosol concentration parameters distributed along the dynamic line-of-sight link trajectory; using the aerosol concentration parameters as a search index to search in a preset environment and radio frequency attenuation mapping table to obtain the predicted link attenuation for each time series; comparing the predicted link attenuation with a preset link margin depletion threshold to extract the time node when the predicted link attenuation first reaches the link margin depletion threshold, and generating the second timestamp.

[0043] In the process of constructing a dynamic line-of-sight link trajectory, the system takes the current moment as the starting point and discretizes the future time period into a series of continuous time nodes according to the set prediction time step. At each time node, the predicted three-dimensional spatial position coordinates of the emergency terminal are calculated by extrapolation using the terminal pose data, and the real-time ephemeris data of the serving low-orbit satellite is analyzed simultaneously. The three-dimensional spatial orbit coordinates of the low-orbit satellite at the corresponding moment are calculated using the space orbit dynamics model. Then, in a unified geocentric rectangular coordinate system, the terminal spatial position coordinates corresponding to each time node are connected to the satellite orbit coordinates point-to-point with a straight line to generate a spatial line segment representing the geometric path of electromagnetic wave transmission. Then, according to the set spatial resolution, the spatial line segment is discretized into a series of continuous three-dimensional coordinate point sets, thereby forming a dynamic line-of-sight link trajectory that moves with the time sequence.

[0044] During the extraction of aerosol concentration parameters, the system retrieves a three-dimensional aerosol concentration distribution matrix (i.e., plume concentration field data) containing multiple future time sections. For each time node, it extracts the set of three-dimensional coordinate points corresponding to the dynamic line-of-sight link trajectory at that node. Using a spatial coordinate mapping algorithm, the above three-dimensional coordinate point set is mapped and positioned one by one into the plume concentration field data grid of the corresponding time section. The specific mapping process is as follows: extract the physical coordinates of the origin of the plume concentration field data grid and the set three-dimensional grid spatial resolution; calculate the three-dimensional physical offset distance of each discrete point in the three-dimensional coordinate point set relative to the origin; then, divide the three-dimensional physical offset distance by the spatial resolution in the corresponding direction, and perform rounding down or rounding operation on the division result to obtain the corresponding three-dimensional integer index value. This three-dimensional integer index value is then used in the plume concentration field data matrix. The system precisely locates the corresponding voxel grid. Then, it reads the aerosol concentration value corresponding to each three-dimensional voxel grid along the trajectory and, combined with the actual physical distance step between adjacent coordinate points, performs path integration calculation on these concentration values ​​along the entire dynamic line-of-sight link trajectory. The specific numerical implementation of the path integration calculation is as follows: extract the Euclidean distance between two adjacent three-dimensional coordinate points as the micro-element step size, multiply the read voxel grid aerosol concentration value with the micro-element step size to obtain the local equivalent concentration on the micro-segment; then, it traverses all micro-segments on the dynamic line-of-sight link trajectory, sums up the calculated local equivalent concentrations, and obtains the total equivalent concentration value encountered by the microwave signal when penetrating the smoke area through this integration calculation. This accumulated equivalent value is used as the aerosol concentration parameter in the corresponding time series.

[0045] In the process of obtaining the predicted link attenuation, the system pre-configures an environment-RF attenuation mapping table. This table records the empirical mapping relationship between different orders of magnitude of aerosol cumulative concentration and additional microwave signal path attenuation (in decibels) under the terminal's current operating communication frequency band. The mapping table is constructed as follows: based on Mie scattering theory, the average particle size distribution parameters of typical aerosol particles in an emergency scenario and the electromagnetic wave wavelength of the terminal's current operating communication frequency band are input to calculate the absorption cross section and scattering cross section coefficient of aerosol at a unit concentration for that frequency band's RF signal; based on these coefficients, under a preset concentration... Within a given range, calculations are performed at predetermined concentration steps to determine the theoretical attenuation decibel values ​​for different cumulative aerosol concentrations. The concentration baseline and theoretical attenuation decibel values ​​are then stored to form a discrete mapping table. Following the chronological order of the time series, the cumulative aerosol concentration parameters calculated in the previous step are sequentially used as search indices in this table. Through data lookup and linear interpolation calculations between adjacent intervals, the attenuation decibel values ​​of the radio frequency signal at the corresponding time node when crossing the plume region due to scattering and absorption are output one by one. These values ​​are then used as the predicted link attenuation for each time series. In the typical forest fire scenario addressed in this application, the average particle size distribution of aerosol particles ranges from 0.3 μm to 3 μm, the real part of the refractive index ranges from 1.45 to 1.55 (corresponding to typical values ​​for carbonaceous particles), and the imaginary part ranges from 0.01 to 0.05. The concentration reference in the reference table uses the particle mass concentration (mg / m³) per unit volume (1 cubic meter), with a step size of 0.5 mg / m³ and a coverage range of 0 to 200 mg / m³. Based on the above parameters, the additional path attenuation in the Ku band (12 GHz) is calculated according to Mie scattering theory. The typical result is that when the cumulative concentration integral value (concentration × path length, in mg / m²) is 100 mg / m², it corresponds to an additional attenuation of approximately 2.5 dB. The above parameters can be calibrated and replaced according to on-site measurement data in different actual scenarios.

[0046] In the traversal determination step of generating the second timestamp, the system pre-calculates the maximum attenuation margin that the current communication link can tolerate without interruption based on the current terminal radio frequency front-end transmission parameters, spatial propagation model and receiver performance, and sets it as the link margin exhaustion threshold.

[0047] It should be noted that the calculation formula is based on the logarithmic domain link budget theory of communication engineering standards. In the physical layer of microwave radio frequency propagation, the final received power of the terminal is obtained by multiplying and dividing the transmit power with various spatial gains and losses in the absolute linear domain (such as watts). To avoid high-complexity floating-point multiplication and division calculations in the underlying hardware, the system uniformly converts and unifies all physical parameters to the logarithmic domain (dBm based on 1 milliwatt, and dB / dBi representing the ratio). According to the mathematical properties of logarithmic operations, multiplication and division in the linear domain are equivalently converted to addition and subtraction of the corresponding physical quantities in the logarithmic domain. Therefore, the system can obtain the accurate attenuation margin directly through a single linear addition and subtraction operation while maintaining the uniformity of logarithmic dimensions, thereby effectively reducing the real-time computing power overhead of the emergency response terminal. The specific calculation formula is as follows: in, This represents the maximum attenuation margin that the current communication link can tolerate (i.e., the threshold for exhausting the link margin). Indicates the transmit power of the radio frequency front end. This represents the theoretical gain of the transmitting antenna. This represents the theoretical gain of the receiving antenna. This represents the baseline free space loss between transmitting and receiving nodes. This indicates the inherent feeder and equipment insertion loss of the system. This represents the receiver sensitivity threshold required to maintain an extremely low bit error rate. To simplify real-time calculations, the link margin depletion threshold M is calculated using a static benchmark: the measured distance between the terminal and the currently serving low-Earth orbit satellite at the system initialization moment is used as a benchmark, and the value is calculated by substituting it into the free-space path loss formula. A baseline value is used, and the M value at this baseline distance is taken as the fixed link margin depletion threshold within the entire rolling time-domain window. Within the window, only the dynamic attenuation of aerosols is compared with this threshold, while the amount caused by satellite orbital motion is ignored. Minor changes are negligible (within a typical 5-minute window of low-Earth orbit satellite transit, the change in free space loss caused by elevation angle changes usually does not exceed 0.5 dB, which can be considered as system margin).

[0048] Subsequently, the system moves forward along the time axis, comparing the predicted link attenuation (also in dB) for each time node with the link margin depletion threshold one by one. When the predicted link attenuation for a certain time node is detected to be greater than or equal to the link margin depletion threshold for the first time, it is determined that the attenuation effect of aerosol has caused the communication link to be unable to maintain a basic connection. The comparison loop is then terminated, and the exact time node parameters that triggered the over-limit state are extracted, converted into a standard time format, and a second timestamp representing the soft attenuation link breakage warning is generated.

[0049] Further, the construction process of the high-Earth orbit and low-Earth orbit graph model includes: comparing a first timestamp and a second timestamp, and determining the minimum of the two as the predicted handover timestamp; dividing and generating a rolling time domain window starting from the current system time and ending at the predicted handover timestamp plus a preset handover signaling interaction margin duration; within the rolling time domain window, mapping the terminal, each candidate low-Earth orbit satellite, and the high-Earth orbit backbone satellite as communication nodes in the graph model, and constructing initial dynamic directed edges based on the radio frequency transceiver payload matching relationship between nodes; and performing masking contour data and each of the dynamic directed edges... Linear collision detection marks the topological connectivity of dynamic directed edges that physically block the masking contour data as disconnected, forming topological constraints. The aerosol concentration integral value along the path of the undisconnected dynamic directed edges in the plume concentration field data is extracted and converted into radio frequency attenuation penalty cost. The spatial transmission delay between each communication node is extracted as the baseline transmission cost. The baseline transmission cost and the radio frequency attenuation penalty cost are weighted and summed to calculate the comprehensive link cost weight of each dynamic directed edge, forming link cost constraints. This generates a high-orbit / low-orbit joint graph model. The specific process is as follows: Figure 3 As shown.

[0050] In the process of determining the predicted handover timestamp and defining the rolling time-domain window, the absolute time values ​​corresponding to the first and second timestamps are first extracted and compared. The time node with the smaller time span, i.e., closer to the current time, is taken as the predicted handover timestamp, thus establishing the earliest critical moment when the current service link will inevitably be interrupted. Then, the current system clock time is obtained as the starting point of the time interval. Based on the predicted handover timestamp value, a margin duration pre-set according to the network layer protocol for signaling handshakes and route reconfiguration is added to obtain the ending point of the time interval. Between this starting point and the ending point, a continuous time axis is divided according to a set discrete sampling step size, generating the rolling time-domain window.

[0051] During the node mapping and initial connectivity construction phase of the graph model, within the aforementioned rolling time-domain window, the system parses the identifiers and orbital parameters of the terminal, each candidate low-Earth orbit satellite, and high-Earth orbit backbone satellite, assigning an independent virtual node index to each communication entity in the graph structure. Next, the system traverses all node combinations, reading the operating frequency band range, polarization mode, and transmit / receive duplex mode supported by the physical radio frequency payload of each node. Only when the transmit frequency bandwidth and polarization mode of the source node are completely overlapped and compatible with the receive channel of the target node, does the system establish an initial dynamic directed edge for these two nodes in the adjacency matrix. This preemptively eliminates invalid links caused by physical layer hardware mismatches, completing the initial topology construction. The specific compatibility determination logic is as follows: extract the highest and lowest operating frequency parameters of the source node's transmit channel and the target node's receive channel, respectively, and calculate the overlap and intersection of their frequency band intervals; when the bandwidth value of the overlap and intersection is greater than or equal to the minimum modulation bandwidth threshold required by the terminal's current pending service data stream, and the radio frequency polarization identifiers of the two are consistent, it is determined that the physical channels are completely overlapped and compatible. In the heterogeneous network scenario of this application, low-Earth orbit (LEO) candidate satellites communicate using the Ku band (11.7 GHz to 12.7 GHz) or the Ka band (26.5 GHz to 40 GHz), while high-Earth orbit (HEO) backbone satellites use the L band (1.5 GHz to 1.6 GHz) or the S band (2.0 GHz to 2.4 GHz). Since the RF front-end of the terminal phased array antenna can only operate in one band at a time, direct directed edges between LEO candidate satellite nodes and HEO backbone satellite nodes are not established during the initial graph construction phase; that is, no direct link is formed between the two types of nodes by default. All directed edges extend from the terminal node to each LEO candidate satellite node (Ku / Ka matching) or each HEO backbone satellite node (L / S matching), forming a star topology with the terminal as a single relay point. The path optimization algorithm traverses these two types of candidate terminal nodes.

[0052] In the spatial collision detection stage that forms topological constraints, for each established initial dynamic directed edge, the system extracts the 3D spatial straight line trajectory connecting the source node and the target node at a specific time step. Using a spatial ray penetration algorithm, this trajectory is projected onto a 3D entity boundary matrix composed of masking contour data for mesh intersection testing. The specific mesh intersection testing process is as follows: using a set 3D mesh edge length as the discrete resolution, a 3D digital differential analysis algorithm is used to resolve the continuous spatial straight line trajectory from the source node to the target node into a series of closely adjacent 3D mesh coordinate index sequences. The system reads the voxel mesh attribute status at the corresponding index position in the 3D entity boundary matrix one by one according to the spatial extension order of this sequence. By determining whether the spatial straight line penetrates the voxel mesh marked as having entity blocking attributes, the system accurately identifies physical links blocked by mountain terrain. For dynamic directed edges that have penetrated and collided, the system directly sets the corresponding connectivity status identifier to empty or assigns an infinite value in the adjacency matrix of the graph structure, logically forcibly severing the connectivity of the edge and forming a hard topological constraint.

[0053] In the comprehensive link cost calculation stage of generating the high-low orbit joint graph model, the system first filters out links marked as disconnected, and only extracts features from effective dynamic directed edges that maintain connectivity. The system locates the spatial trajectory of the effective directed edges in the plume concentration field data, and discretizes and accumulates the aerosol concentration values ​​of the passed grid along this trajectory segment to obtain the total aerosol concentration integral value on the path. This integral value is then multiplied by a set frequency band attenuation conversion coefficient to convert it into a radio frequency attenuation penalty cost representing signal quality degradation. Simultaneously, the system calculates the three-dimensional Euclidean distance between connected nodes, divides it by the propagation speed of electromagnetic waves in a vacuum to obtain the absolute physical delay, and converts it into a baseline transmission cost. Finally, a pre-configured weight allocation ratio factor is extracted. This pre-configured weight allocation ratio factor is looked up through a service type-weight mapping table, which is pre-built offline during the system initialization phase. The specific construction process includes: First, the system builds an offline simulation network environment for this emergency scenario and imports historical real masking contour data, typical aerosol evolution data and satellite ephemeris data as simulation background. Secondly, a weight search space is established in the simulation environment, and multiple sets of candidate weight ratio combinations that satisfy the linear constraint equation W1+W2=1 are generated in the interval from 0 to 1 with a set small step size (such as 0.05). Next, for each of the preset typical service types (such as real-time voice, video stream, and telemetry data), the routing optimization algorithm is run in the simulation environment by applying the above-mentioned candidate weight ratio combinations, and network connectivity and data stream transmission simulation tests are continuously performed. During the testing process, the system collected end-to-end network performance metrics for various service types. Specifically, for voice or video streaming services with extremely high real-time requirements, average end-to-end latency was extracted as the core fitness evaluation metric; for telemetry data services with extremely high anti-interference requirements, the number of reconnections due to attenuation and the overall packet loss rate were extracted as core fitness evaluation metrics. Finally, by comparing various indicators, candidate weight ratio combinations that minimize the real-time service core adaptability evaluation indicator (average latency) and candidate weight ratio combinations that minimize the telemetry service core adaptability evaluation indicator (link breakage and packet loss rate) are selected. The system binds and stores each service type identifier with its corresponding optimal candidate weight ratio combination obtained from testing. This generates and distributes the service type and weight mapping table for emergency response terminals to look up and call during real-time scheduling. The cost is multiplied by the baseline transmission cost and the radio frequency attenuation penalty cost, and then summed to obtain a single comprehensive link cost value. Since the time dimension represented by the baseline transmission cost and the decibel dimension represented by the radio frequency attenuation penalty cost belong to different physical dimensions... The system pre-calculates the delay and attenuation extreme values ​​of all available connected links within the current time window, performs range normalization on these two types of cost data, and linearly maps them to a dimensionless standard interval of zero to one. Then, it extracts the pre-configured weight allocation ratio factor (including delay weight and attenuation weight, with the sum of the two configurations being 1), multiplies the delay weight by the baseline transmission cost after range normalization, and multiplies the attenuation weight by the radio frequency attenuation penalty cost after range normalization. Finally, it sums the products of the two to obtain a single dimensionless comprehensive link cost value, and assigns this value as the final weight to the corresponding dynamic directed edge, thereby generating a high-low orbit joint graph model that includes dynamic network topology and environmental attenuation characteristics.

[0054] Furthermore, the process of acquiring the target satellite identifier and antenna pointing data includes: using the node mapped by the terminal in the high and low orbit map model as the source point, and using the available satellite communication nodes within the rolling time domain window as the traversal space, running a path optimization algorithm to calculate the cumulative link cost from the source point to each available satellite communication node; comparing and obtaining the optimal communication path where the cumulative link cost reaches a minimum value, and extracting the satellite node pointed to by the optimal communication path as the target satellite identifier; retrieving the corresponding real-time satellite ephemeris data based on the target satellite identifier, performing a three-dimensional spatial coordinate system transformation and calculation on the real-time satellite ephemeris data and the terminal pose data, and calculating the azimuth and elevation angles of the phased array antenna driving the terminal as antenna pointing data.

[0055] In the process of calculating the cumulative link cost using the path optimization algorithm, the system first initializes a link cost record table, sets the cost of the source node mapped to the terminal in the high-Earth orbit joint graph model to zero, and sets the initial cumulative cost of all other available satellite communication nodes within the rolling time window to infinity. Simultaneously, a predecessor node tracing table is established. Subsequently, the system uses a shortest path search logic based on a priority queue (such as Dijkstra's algorithm logic) to traverse the graph model. In each iteration, the untraversed node with the smallest cumulative cost in the current record table is selected as the active node. All valid dynamic directed edges originating from this active node are retrieved, and the comprehensive link cost weight corresponding to each directed edge is extracted and added to the current cumulative cost of the active node. If the new cost value is strictly less than the original cumulative cost of the target neighbor node in the record table, the new cost value overwrites the cumulative cost record of the target neighbor node, and the predecessor pointer of the target neighbor node in the predecessor node tracing table is updated to the current active node. This node expansion and cost relaxation update operation is continuously performed until all connected nodes within the rolling time window have been traversed.

[0056] In the process of extracting the target satellite identifier, the updated link cost record table is queried, and the final cumulative link cost values ​​of all available satellite communication nodes (including each candidate low-Earth orbit satellite and high-Earth orbit backbone satellite) that serve as endpoint nodes are extracted. A linear comparison is performed on these final cost values, and endpoint satellite nodes whose values ​​have reached the global minimum are selected. After determining the optimal endpoint, the predecessor node tracing table is called, and the topology is backtracked in reverse along the recorded predecessor pointers starting from the minimum endpoint satellite node until the terminal source point is reached. This reconstructs the complete optimal communication path sequence. From this sequence, the satellite node that directly establishes a first-hop radio frequency connection with the terminal source point is selected, and its corresponding network layer unique identification code is extracted as the final output target satellite identifier.

[0057] In the process of performing the three-dimensional spatial coordinate system transformation calculation, based on the extracted target satellite identifier, the real-time ephemeris data corresponding to the satellite at the predicted switching timestamp is retrieved. Using an orbital dynamics extrapolation model, the three-dimensional position coordinates of the satellite in the geocentric fixed coordinate system in Earth's absolute space are resolved. Specifically, using the SGP4 orbital dynamics extrapolation model, with the NORAD double-row orbital elements from the satellite's real-time ephemeris data as input, the three-dimensional position coordinates of the satellite in the geocentric fixed coordinate system corresponding to the predicted switching timestamp are resolved. The SGP4 model incorporates the assumptions for Earth's non-spherical gravitational fields J2 / J3 / J4. Analytical compensation for perturbation terms, atmospheric drag perturbations, and lunar gravitational perturbations, within a short extrapolation window of 5 to 30 minutes, results in a low-Earth orbit satellite position prediction accuracy better than 1 km, sufficient to meet the accuracy requirements for initial beam alignment of phased array antennas. After establishing physical layer synchronization, the terminal performs single-pulse angle measurement closed-loop tracking by receiving satellite downlink beacon signals. Simultaneously, it extracts geodetic longitude, latitude, and elevation information from the terminal pose data, converting this geospatial information into three-dimensional coordinate points in a geocentric fixed coordinate system. The specific formula for converting these three-dimensional coordinate points into a geocentric fixed coordinate system is as follows: in, , , Represents a three-dimensional coordinate point; This indicates the extracted geodetic longitude; Indicates latitude; This represents the first eccentricity of the pre-defined Earth ellipsoid; express The calculated radius of curvature of the zonal loop, where This is the pre-defined semi-major axis of the Earth's ellipsoid.

[0058] Subsequently, the system subtracts the terminal's geocentric fixed coordinates from the satellite's geocentric fixed coordinates, generating a spatial line-of-sight vector pointing directly from the terminal to the target satellite through three-dimensional spatial vector subtraction. Next, a geographic rotation projection matrix is ​​constructed based on the latitude and longitude parameters of the terminal's location. The spatial line-of-sight vector in the aforementioned geocentric fixed coordinate system is multiplied by this projection matrix, rotating and mapping it to a local geographic rectangular coordinate system with the terminal's physical location as the origin (northeast celestial). This geographic rotation projection matrix is ​​based on the geodetic longitude of the terminal's location. with latitude The specific matrix structure and rotation mapping calculations are as follows: in, , , These are the east direction axis components, north direction axis components, and zenith direction axis components after being mapped to the local geographic rectangular coordinate system of the northeast sky; , , These are the three components of the spatial line-of-sight vector pointing directly from the terminal to the target satellite in the geocentric fixed coordinate system. Indicates the geodetic longitude of the terminal's location; This indicates the geodetic latitude of the terminal's location.

[0059] In the step of calculating the azimuth and elevation angles of the driving phased array antenna, under the local geographic rectangular coordinate system of the northeast sky, three spatial projection components of the aforementioned spatial line-of-sight vector on the east, north, and zenith axes are extracted. These three projection components are then processed using inverse trigonometric function logic to calculate the absolute geographic azimuth and absolute geographic elevation angles of the target satellite relative to the geographic horizontal plane where the terminal is located. The specific calculation formula using inverse trigonometric function logic is as follows: in, Indicates absolute geographical azimuth; Indicates absolute geographical pitch angle; , , These are the components of the east, north, and zenith directions, respectively, mapped to the local geographic rectangular coordinate system of the northeast celestial sphere; based on... and The quadrant of the positive and negative signs of the components will determine the calculated absolute geographic azimuth. The mapping is uniformly applied to a value range of 0° to 360°, and the absolute geographic pitch angle is constrained. The value range is from -90° to 90°.

[0060] Finally, to eliminate the tilting error of the rescue terminal caused by terrain undulations or movement, the real-time roll, pitch, and yaw angle data measured by the inertial measurement unit (IMU) are extracted from the terminal's pose data. This data is used to construct a three-dimensional attitude inverse rotation transformation matrix. The construction logic of the three-dimensional attitude inverse rotation transformation matrix is ​​as follows: Assuming the yaw, pitch, and roll angles measured by the IMU, and following a predetermined "yaw-pitch-roll" sequence, rotation matrices around the terminal's Z-axis, Y-axis, and X-axis are constructed respectively; multiplying these three basic rotation matrices yields... The comprehensive rotation matrix of the terminal body attitude is obtained. Then, using the property of orthogonal matrices, the comprehensive rotation matrix is ​​transposed to obtain its inverse matrix. The transposed matrix is ​​used as the final three-dimensional attitude inverse rotation transformation matrix for attitude stripping. The calculated absolute geographic azimuth and elevation angles are input into the inverse rotation transformation matrix for attitude stripping. The geographic reference angles are converted into relative angles with respect to the center normal of the phased array antenna physical panel. Finally, the calculated set of relative geometric angle data is directly output as the final antenna pointing data to drive the underlying phase shifter hardware.

[0061] Furthermore, the process of redirecting access to the corresponding target candidate low-Earth orbit (LEO) satellite or high-Earth orbit (HEO) backbone satellite includes: parsing the network hierarchy attribute of the target satellite identifier to determine its orbital type; when the target satellite identifier belongs to the candidate LEO satellite, driving the phased array antenna of the terminal to activate beam pre-synchronization and establish a LEO relay link before the predicted switching timestamp arrives, based on the antenna pointing data, and switching the terminal's data stream to be transmitted to the target candidate LEO satellite; when the target satellite identifier belongs to the HEO backbone satellite, driving the phased array antenna of the terminal to switch to a preset anti-attenuation communication frequency band and establish a cross-orbit direct connection link, based on the antenna pointing data, and redirecting the terminal's data stream to be transmitted to the HEO backbone satellite.

[0062] In the process of parsing the network-level attributes of the target satellite identifier, the system directly extracts the orbital feature code encapsulated inside the target satellite identifier. The specific extraction method is as follows: read the header field of the data packet of the target satellite identifier, perform a bitwise AND operation on the header field using a preset binary mask, and extract a specific byte segment used to characterize the satellite orbital attribute as the orbital feature code. The orbital feature code is compared with a local constellation hierarchy mapping table pre-constructed based on publicly available satellite orbital parameters. Based on the table lookup matching result, it is determined whether the target satellite belongs to a candidate low-Earth orbit satellite or a high-Earth orbit backbone satellite.

[0063] When the target satellite is determined to belong to a candidate LEO satellite, the system converts the antenna pointing data into a phase control signal. The system then sends the antenna pointing data to the underlying baseband beamforming controller through a standard peripheral interface. The controller directly outputs the corresponding digital phase control signal based on its built-in array arrangement algorithm, driving the phased array antenna to align with the target satellite. Before the predicted switching timestamp arrives, the system sends a physical layer synchronization pilot sequence to the target candidate LEO satellite through this beam to initiate a handshake. After synchronization is established, the system modifies its local medium access control layer forwarding table, directly switching the output port of the service data stream to be sent to the LEO relay link, thus achieving data stream migration that is established before being disconnected.

[0064] When the target satellite is determined to belong to a high-orbit backbone satellite, the system also drives the phased array antenna to align with the high-orbit target position based on antenna pointing data. Simultaneously, the system sends a local oscillator adjustment command to the radio frequency front-end, directly switching the carrier communication frequency band to a preset, attenuation-resistant microwave band with strong aerosol penetration capability. This is because the Ku / Ka band used in conventional low-orbit satellite communication has a short wavelength, making it easily absorbed and scattered by high-concentration smoke particles at fire scenes. Therefore, the system, based on the multi-band common-aperture phased array architecture used in the terminal radio frequency front-end (in which the Ku / Ka band and L / S band each have independent frequency synthesizers, power amplifiers, and low-noise amplifier links), triggers the corresponding frequency band's radio frequency switching channel, rather than relying on the large span of a single voltage-controlled oscillator. The system hard-switches the carrier to a pre-configured L-band (e.g., 1.5 GHz to 1.6 GHz) or S-band (e.g., 2.0 GHz to 2.4 GHz). Utilizing the longer electromagnetic wavelength of this low-frequency band, aerosol particles are placed in the deep Rayleigh scattering region relative to the wavelength. The scattering cross-section of the particles to electromagnetic waves is inversely proportional to the fourth power of the wavelength and decreases sharply. At the same time, by selecting circular polarization, the depolarization loss of linearly polarized waves on non-spherical particles is reduced, thereby achieving strong aerosol penetration and anti-attenuation at the physical level. A cross-orbit random access request is then sent. After successfully establishing a cross-orbit direct link, the system updates the routing and forwarding rules at the network layer, modifying the target address of the data stream to be sent to the high-orbit system gateway address, thereby redirecting the service data offloaded to the high-orbit backbone satellite.

[0065] Example 2: The rescue team carried a vehicle-mounted phased array satellite terminal deep into the canyon to conduct close-up reconnaissance of the fire. Due to the obstruction of steep mountains on both sides, the visible view from the sky was extremely limited. At the same time, the intense burning of vegetation inside the fire generated a large amount of high-concentration aerosols, which formed a dynamically drifting plume of dense smoke under the influence of complex valley wind fields.

[0066] In this scenario, a hovering drone scans the canyon terrain in real time using penetrating synthetic aperture radar, transmitting a 3D entity boundary matrix to the terminal. Simultaneously, environmental monitoring nodes deployed forward transmit real-time data on sudden wind direction changes and sharp increases in local aerosol levels. The terminal, combining its own position and the ephemeris of the low-Earth orbit satellites currently providing services, initiates a link failure prediction and reconstruction process.

[0067] Based on the terminal pose and the motion trajectory of the serving low-Earth orbit satellite, a two-dimensional link break prediction is performed in parallel: On the one hand, the system performs space ray tracing along the radio frequency link pointing to the baseline and predicts that at some point in the future, due to the high-speed passage of the low-orbit satellite, the link line of sight will be physically cut off by the mountain peaks on the east side of the canyon, thus generating the first timestamp representing the "hard blockage".

[0068] On the other hand, the system inputs storm-level flow velocity and high-concentration aerosol data into the plume model. The model predicts that, affected by the rapid change in the wind field, a high-density plume of smoke will laterally cut into the current RF link trajectory before the mountain peaks block it. The system performs concentration path integration along the trajectory and looks up the mapping table, discovering that the electromagnetic wave scattering and absorption attenuation caused by the smoke will increase rapidly and exhaust the maximum attenuation margin of the terminal link before the mountain peaks block it. This generates a second timestamp characterizing the soft attenuation.

[0069] After comparing the minimum values, the system determined that the second timestamp (aerosol soft attenuation) occurred earlier than the first timestamp (mountain hard occlusion), and thus established the second timestamp as the predicted switching timestamp to define the rolling time domain window.

[0070] In the graph model construction and cost optimization stage, physical collision detection is first performed using the entity boundary matrix issued by the radar. All candidate satellite links that are blocked by the mountains on both sides within the time window are decisively eliminated, forming hard topological constraints.

[0071] Subsequently, the system extracts the business data stream currently being sent by the assault team, namely, high-definition real-time video of the fire spread. Based on the pre-set business type and weight mapping table in Implementation Example 1, the system determines that the video stream belongs to a business with extremely high real-time requirements. Therefore, it adaptively allocates the first weight factor (latency weight) to a large proportion and the second weight factor (attenuation weight) to a small proportion. Based on this, it performs range normalization and weighted summation on the remaining effective links to generate a high-low orbit joint graph model.

[0072] Running the shortest path optimization algorithm in the above graph model, since the remaining visible low-orbit satellites above the canyon are all covered by a large area of ​​dense smoke, the calculated comprehensive link cost is at an extremely high level. After global comparison, the system selects a high-orbit backbone satellite in geostationary orbit as the optimal endpoint and outputs its corresponding target satellite identifier.

[0073] Since the target belongs to a high-orbit backbone satellite, before the predicted switching time stamp (i.e., when the dense smoke completely blocks the current low-orbit link) arrives, the system drives the vehicle-mounted phased array antenna to directly align with the high-orbit satellite based on the azimuth and elevation angles calculated by the three-dimensional coordinate system. At the same time, it sends a local oscillator adjustment command to the radio frequency front end to quickly switch the carrier communication frequency band from the high-frequency band to a preset anti-attenuation microwave frequency band (such as the L / S band) with strong aerosol penetration capability. Finally, it modifies the network layer routing to successfully redirect and offload the fire reconnaissance video stream to the high-orbit direct link before the current low-orbit service is completely paralyzed, ensuring the absolute unimpeded lifeline of disaster relief in the scenario of superimposed extreme geological and meteorological disasters.

[0074] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high- and low-orbit satellite fusion communication scheduling method for emergency scenarios, characterized in that, include: Acquire terminal pose data, masking contour data, and meteorological feature data; acquire real-time ephemeris data of serving low-Earth orbit satellites, candidate low-Earth orbit satellites, and high-Earth orbit backbone satellites. Line-of-sight occlusion calculations are performed based on terminal pose data, masking contour data, and real-time ephemeris data of the serving LEO satellite to generate a first timestamp. Specifically, a time-varying line-of-sight spatial vector from the emergency terminal to the serving LEO satellite is constructed based on the three-dimensional position coordinates in the terminal pose data and the real-time ephemeris data of the serving LEO satellite. Attitude pitch angles and related yaw and roll data are extracted from the terminal pose data to construct a three-dimensional spatial rotation mapping matrix. Based on the terminal's current spatial attitude represented by the pitch, yaw, and roll angles, the time-varying line-of-sight spatial vector is spatially attitude-mapped. The time-varying line-of-sight spatial vector is then projected onto a relative coordinate system with the terminal's physical antenna panel as the origin through the three-dimensional spatial rotation mapping matrix to generate a radio frequency link pointing baseline. Line-of-sight occlusion calculations are performed based on the radio frequency link pointing baseline and the masking contour data. Meteorological characteristic data is input into a plume model to generate plume concentration field data. Combined with the real-time ephemeris data of the serving LEO satellite, the predicted link attenuation is calculated. The time node corresponding to when the predicted link attenuation reaches a threshold is extracted to generate a second timestamp. Specifically, the meteorological characteristic data is input into the plume model to generate plume concentration field data. A dynamic line-of-sight (LAS) trajectory is constructed based on the terminal pose data and the real-time ephemeris data of the serving low-Earth orbit satellite, and the LAS trajectory is spatially matched with the plume concentration field data. The aerosol concentration values ​​corresponding to the three-dimensional voxel grid traversed by the LAS trajectory are read, and combined with the actual physical distance step between adjacent coordinate points, path integrals are calculated along the LAS trajectory for the aerosol concentration values. The product of the aerosol concentration value and the corresponding actual physical distance step is then calculated along the LAS trajectory. The total concentration equivalent value encountered by the microwave signal when penetrating the smoke area is obtained by accumulating the dynamic line-of-sight link trajectory. This total concentration equivalent value is used as the aerosol concentration parameter for the corresponding time series. The aerosol concentration parameter is the cumulative concentration integral value formed by integrating the aerosol volume mass concentration along the dynamic line-of-sight link trajectory. The aerosol concentration parameter is used as a search index to search in a preset environment and radio frequency attenuation mapping table. Based on the mapping relationship between the cumulative aerosol concentration and the additional attenuation of the microwave signal path in the environment and radio frequency attenuation mapping table, the predicted link attenuation amount for each time series is obtained. Based on the current terminal RF front-end's transmission parameters, space propagation model, and receiver performance, the maximum attenuation margin that the current communication link can tolerate without interruption is calculated in advance, and this maximum attenuation margin is set as the link margin depletion threshold. Specifically, the free space loss corresponding to the actual distance between the terminal and the currently serving low-Earth orbit satellite at the system initialization moment is used as the baseline free space loss in the link budget. The available attenuation margin of the current communication link is determined by combining the terminal's RF front-end's transmission power, transmit antenna gain, receive antenna gain, feeder and equipment insertion loss, and receiver sensitivity. The change in free space loss caused by the change in the relative position of the terminal and the serving low-Earth orbit satellite during the rolling prediction process based on the system initialization moment under the current service connection state is included as the propagation loss change relative to the baseline free space loss in the determination of the link margin. The predicted link attenuation is compared with the link margin depletion threshold, and the time node when the predicted link attenuation first reaches the link margin depletion threshold is extracted to generate the second timestamp. Extract the minimum value between the first and second timestamps as the predicted switching timestamp; set a rolling time-domain window that includes the predicted switching timestamp, and use the masking contour data and plume concentration field data as topology and link cost constraints to fuse real-time ephemeris data of candidate low-Earth orbit satellites and high-Earth orbit backbone satellites to construct a high-low orbit map model. In the high and low orbit map model, an optimization algorithm is used to find the target satellite identifier and antenna pointing data with the minimum link cost. Based on the target satellite identifier and antenna pointing data, the terminal communication link is redirected to access the corresponding target candidate low orbit satellite or high orbit backbone satellite.

2. The high- and low-orbit satellite fusion communication scheduling method for emergency scenarios according to claim 1, characterized in that, The process of acquiring terminal pose data, masking contour data, meteorological feature data, and real-time ephemeris data includes: acquiring the three-dimensional position coordinates and velocity vectors calculated by the emergency response terminal based on the satellite positioning module, and combining them with the attitude pitch angle data collected by the inertial measurement unit to generate the terminal pose data; receiving the three-dimensional entity boundary matrix scanned and sent by the edge computing node through a penetrating detection device as the masking contour data, wherein the penetrating detection device adopts a synthetic aperture radar mounted on an unmanned aerial vehicle platform; extracting real-time wind field velocity data and aerosol concentration data collected by environmental monitoring equipment deployed in the emergency area as the meteorological feature data; parsing the satellite system broadcast ephemeris messages received by the emergency response terminal, and extracting the real-time ephemeris data of the serving low-Earth orbit satellites, candidate low-Earth orbit satellites, and high-Earth orbit backbone satellites.

3. The high- and low-orbit satellite fusion communication scheduling method for emergency scenarios according to claim 1, characterized in that, The process of generating the first timestamp includes: constructing a time-varying line-of-sight spatial vector from the emergency terminal to the serving low-Earth orbit satellite based on the three-dimensional position coordinates in the terminal pose data and the real-time ephemeris data of the serving low-Earth orbit satellite; performing a terminal antenna coordinate system transformation on the time-varying line-of-sight spatial vector by combining the attitude pitch angle data in the terminal pose data to generate a radio frequency link pointing baseline; traversing and comparing the radio frequency link pointing baseline with the three-dimensional entity boundary matrix in the masking contour data within a set time step to calculate the spatial geometric intersection state of the radio frequency link pointing baseline being blocked by the entity terrain; extracting the time node parameter of the first spatial intersection between the radio frequency link pointing baseline and the three-dimensional entity boundary matrix, and generating the time node parameter as the first timestamp.

4. The high- and low-orbit satellite fusion communication scheduling method for emergency scenarios according to claim 2, characterized in that, The plume model includes: an input feature layer that receives the wind field velocity data and the aerosol concentration data, and combines the masking contour data to construct an initial spatiotemporal feature map containing the topological relationships of three-dimensional spatial nodes and edges; The spatial graph convolutional layer receives the initial spatiotemporal feature map, uses the physical masking contour as the blocking boundary, and performs spatial convolution aggregation on the aerosol concentration features between adjacent nodes according to the wind field velocity direction to generate a spatial diffusion feature tensor. The time series extraction layer receives the spatial diffusion feature tensor, performs time series memory calculations along a set time window step, and extracts the dynamic evolution features of aerosol concentration over time. The output mapping layer receives dynamic evolution features and outputs a three-dimensional aerosol concentration distribution matrix containing multiple consecutive time nodes within the future prediction period. The three-dimensional aerosol concentration distribution matrix is ​​used as the plume concentration field data.

5. The high- and low-orbit satellite fusion communication scheduling method for emergency scenarios according to claim 1, characterized in that, The process of generating the second timestamp includes: constructing a dynamic line-of-sight link trajectory that changes over time based on the terminal pose data and the real-time ephemeris data of the serving low-Earth orbit satellite; spatially matching the dynamic line-of-sight link trajectory with plume concentration field data to extract aerosol concentration parameters distributed along the dynamic line-of-sight link trajectory; using the aerosol concentration parameters as a search index to search in a preset environment and radio frequency attenuation mapping table to obtain the predicted link attenuation for each time series; comparing the predicted link attenuation with a preset link margin depletion threshold to extract the time node when the predicted link attenuation first reaches the link margin depletion threshold, and generating the second timestamp.

6. The high- and low-orbit satellite fusion communication scheduling method for emergency scenarios according to claim 1, characterized in that, The construction process of the high-Earth orbit and low-Earth orbit graph model includes: comparing a first timestamp and a second timestamp, and determining the minimum of the two as the predicted handover timestamp; dividing and generating a rolling time domain window starting from the current system time and ending at the predicted handover timestamp plus a preset handover signaling interaction margin duration; within the rolling time domain window, mapping the terminal, each candidate low-Earth orbit satellite, and the high-Earth orbit backbone satellite as communication nodes in the graph model, and constructing initial dynamic directed edges based on the radio frequency transceiver payload matching relationship between nodes; and performing masking contour data and each of the dynamic directed edges... The system performs line collision detection, marking the topological connectivity of dynamic directed edges that physically block the masking contour data as disconnected, thus forming topological constraints. It extracts the aerosol concentration integral value along the path of undisconnected dynamic directed edges in the plume concentration field data, converting it into radio frequency attenuation penalty cost. It extracts the spatial transmission delay between each communication node as the baseline transmission cost. Finally, it weights and sums the baseline transmission cost with the radio frequency attenuation penalty cost to calculate the comprehensive link cost weight of each dynamic directed edge, forming link cost constraints and generating a high-low orbit joint graph model.

7. The high- and low-orbit satellite fusion communication scheduling method for emergency scenarios according to claim 1, characterized in that, The process of acquiring the target satellite identifier and antenna pointing data includes: using the node mapped by the terminal in the high and low orbit map model as the source point, and using the available satellite communication nodes within the rolling time domain window as the traversal space, a path optimization algorithm is used to calculate the cumulative link cost from the source point to each available satellite communication node; the optimal communication path that reaches the minimum value of the cumulative link cost is compared and obtained, and the satellite node pointed to by the optimal communication path is extracted as the target satellite identifier; the corresponding real-time satellite ephemeris data is retrieved according to the target satellite identifier, and the real-time satellite ephemeris data and the terminal pose data are transformed and solved in a three-dimensional spatial coordinate system to calculate the azimuth and elevation angles of the phased array antenna driving the terminal, which are used as antenna pointing data.

8. The high- and low-orbit satellite fusion communication scheduling method for emergency scenarios according to claim 1, characterized in that, The process of redirecting access to the corresponding target candidate low-Earth orbit (LEO) satellite or high-Earth orbit (HEO) backbone satellite includes: parsing the network hierarchy attribute of the target satellite identifier to determine its orbital type; when the target satellite identifier belongs to the candidate LEO satellite, the phased array antenna of the terminal is driven by the antenna pointing to establish a low-Earth orbit relay link before the predicted switching timestamp arrives, and the terminal's data stream to be transmitted is switched to the target candidate LEO satellite; when the target satellite identifier belongs to the HEO backbone satellite, the phased array antenna of the terminal is driven by the antenna pointing to switch to a preset anti-attenuation communication frequency band and establish a cross-orbit direct link, and the terminal's data stream to be transmitted is redirected and offloaded to the HEO backbone satellite.

Images