Prediction methods, devices, and computer storage media for emergency location message delivery
By constructing an undirected node topology graph and a deep Q-network model, model distillation technology is used to predict the transmission path and time of emergency location messages in low-Earth orbit satellites. This solves the problem that existing technologies cannot predict the delivery time of location messages, enabling detailed and timely feedback in emergency scenarios and enhancing the will to survive of rescued personnel.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- IPLOOK NETWORKS CO LTD
- Filing Date
- 2023-12-15
- Publication Date
- 2026-07-31
AI Technical Summary
Existing emergency IoT positioning devices cannot provide predictions of the time required for location messages to be uploaded to the emergency response center, resulting in people waiting for rescue not receiving timely feedback on their distress calls, which affects their will to survive.
By constructing an undirected node topology graph and a deep Q-network model, and using model distillation technology to compress the large model into a small model, which is then deployed in low-Earth orbit satellites, the transmission path and time of emergency positioning messages are predicted, thus achieving accurate prediction of the delivery time of emergency positioning messages.
It provides detailed and timely feedback in emergency scenarios, helping users understand the rescue time and enhancing the will to survive of those being rescued.
Smart Images
Figure CN117668197B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of communications, and in particular to the transmission of emergency location messages for Internet of Things (IoT) devices based on low-Earth orbit (LEO) satellites. Background Technology
[0002] In this era of "Internet of Everything", IoT devices will be integrated into various intelligent scenarios, including transportation, aquaculture, and real-time monitoring. For example, in shipping, various goods are stored in containers equipped with IoT positioning devices. In the event of a maritime disaster, the goods can be retrieved through satellite positioning.
[0003] Satellite emergency communication is characterized by its wide coverage, lack of time and distance limitations, strong communication capabilities, and excellent flexibility. Since disasters and major safety incidents often occur in areas with complex terrain and harsh natural conditions (such as at sea, in mountains, and deserts), it poses significant challenges to emergency services. Satellite communication, however, is not constrained by terrain or weather conditions, thus improving the efficiency of rescue efforts. For example, if a ship sinks unexpectedly at sea, and the crew and some containers are washed ashore on a remote island, although they are temporarily safe, they need rescue within days due to scarcity of food and water. An IoT device equipped with positioning capabilities, subscribed to the IoT SAT service, connects via 5G satellite IoT connectivity. IoT SAT uses a low-Earth orbit (LEO) satellite constellation supporting store-and-forward operation. This positioning IoT device provides emergency linkage and positioning capabilities. When the emergency button on the IoT device is pressed, an indicator light illuminates, and survivors can wait for rescue.
[0004] However, in existing technologies, emergency IoT positioning devices only provide status indication functions for the sending of positioning messages, and fail to provide the estimated time required for the positioning messages to be uploaded to the emergency response center. For people waiting for rescue, providing the estimated time for message upload can provide timely feedback on their distress calls and enhance their will to survive.
[0005] Therefore, there is an urgent need for a method to solve the above problems. Summary of the Invention
[0006] The purpose of this invention is to provide a method, apparatus, and readable storage medium for predicting the delivery of emergency location messages for IoT devices based on low-Earth orbit satellites. This method can distill a large overall deep Q-network model into multiple smaller deep Q-network models through model distillation and deploy them in each group of low-Earth orbit satellites to predict the time required for emergency location messages of IoT devices to be sent to the emergency response center, thus providing users with more detailed and timely feedback in emergency scenarios.
[0007] To achieve the above objectives, this invention discloses a prediction method for emergency location message delivery to IoT devices based on low-Earth orbit (LEO) satellites, comprising: constructing an undirected node topology graph based on nodes of IoT devices, emergency response centers, gateway stations, satellite monitoring stations, and satellites, wherein the satellites include LEO satellites; obtaining link information corresponding to the undirected node topology graph; training a deep Q-network model by taking the undirected node topology graph and the link information as input states and the message transmission path from the IoT device to the emergency response center as output actions; and distilling the deep Q-network model as a teacher model to obtain a one-to-one correspondence with each group of LEO satellites. Furthermore, a student policy network is established, using the low-Earth orbit (LEO) satellite as the starting satellite. This student policy network is stored in the corresponding LEO satellite, where the starting satellite is the first LEO satellite to which the message sent by the IoT device is delivered. When the IoT device sends an emergency location message to an overpass LEO satellite, the overpass LEO satellite predicts the message delivery path of the emergency location message using the stored student policy network. Based on the message delivery path, the time required for the emergency location message to reach the emergency response center is calculated to obtain the predicted delivery time. The predicted delivery time is then returned to the IoT device, which receives and displays the predicted delivery time.
[0008] Preferably, calculating the time required for the emergency location message to reach the emergency response center based on the message transmission path to obtain the predicted delivery time specifically involves summing the propagation delays between all nodes in the message transmission path to obtain the predicted delivery time.
[0009] Preferably, distilling the large deep Q-network model as a teacher model to obtain a student policy network corresponding one-to-one with each group of low-Earth orbit (LEO) satellites and using the LEO satellites as starting satellites specifically includes: storing several task sources in the large deep Q-network model, each task source having a starting satellite identifier, state, and action corresponding to that starting satellite; setting an experience replay pool corresponding to each LEO satellite and placing the task source into the experience replay pool corresponding to its starting satellite; establishing a small deep Q-network model corresponding one-to-one with each group of LEO satellites using the large deep Q-network model as a teacher model; inputting the state and starting satellite identifier in the task source as input to the corresponding small deep Q-network model to predict the output action; obtaining the actions stored in the experience replay pool from the experience replay pool through a distillation loss function model; calculating the minimum loss based on the predicted action and the stored action; optimizing the parameters in the small deep Q-network model based on the minimum loss; iteratively learning to complete the training of the small deep Q-network model, thereby obtaining the student policy network.
[0010] Preferably, an undirected node topology graph is constructed based on the nodes of IoT devices, emergency response centers, gateway stations, satellite monitoring stations, and satellites, including: acquiring ephemeris data from the integrated space-ground communication network accessed by the IoT devices; acquiring the location data of the IoT devices, emergency response centers, gateway stations, satellite monitoring stations, and satellites in the integrated space-ground communication network; establishing corresponding nodes for the IoT devices, emergency response centers, gateway stations, satellite monitoring stations, and satellites based on the location data and the ephemeris data; and generating the undirected node topology graph based on all the nodes. The undirected node topology graph includes ephemeris data of satellite nodes, location data of gateway station nodes, node data of emergency response centers, node data of satellite monitoring stations, and location data of IoT device nodes collected and statistically analyzed by a third party.
[0011] Specifically, generating the undirected node topology graph based on all the nodes includes: calling the Network X toolkit to create the undirected node topology graph based on all the nodes.
[0012] Specifically, obtaining the link information corresponding to the undirected node topology graph includes: traversing all edges in the undirected node topology graph; determining the network service parameters between the two nodes connected by each edge; and using all the network service parameters as the link information.
[0013] Preferably, both the large deep Q-network model and the small deep network model are DQN network models. The DQN network model includes four layers: the first layer is the input layer, which includes a fully connected layer and an activation function layer; the second and third layers both include fully connected layers and activation function layers; and the fourth layer is the output layer, which includes a fully connected layer. The fully connected layer uses the nn.Linear function in PyTorch, and the activation function layer uses the ReLU function.
[0014] The present invention also discloses a predictive device for the delivery of emergency location messages to Internet of Things (IoT) devices based on low-Earth orbit (LEO) satellites, comprising: a processor and a memory, wherein the memory is used to store at least one executable instruction, and the executable instruction causes the processor to perform the operation of the predictive method for the delivery of emergency location messages to IoT devices based on LEO satellites as described above.
[0015] The present invention also discloses a computer storage medium storing at least one executable instruction, which causes a processor to perform the operation corresponding to the prediction method for emergency location message delivery to Internet of Things devices based on low-Earth orbit satellites as described above.
[0016] Compared with existing technologies, this paper discloses a method for predicting the delivery time of emergency location messages of IoT devices by compressing the large model of the deep Q network through model distillation to obtain a small model for easier deployment. The deep Q network of the small model is used to analyze the orbital information of satellite motion and the latency of the link. This method can predict the time required for the emergency location message of the IoT device to be sent to the emergency response center, and provide users with more detailed and timely feedback in emergency scenarios. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of an IoT emergency scenario.
[0018] Figure 2 It is an IoT emergency undirected node topology diagram.
[0019] Figure 3 is a diagram of the DQN prediction model network architecture.
[0020] Figure 4 is a flowchart of the model distillation process.
[0021] Figure 5 This is a flowchart of the prediction method for emergency location message delivery to IoT devices based on low-orbit satellites, as per the present invention. Detailed Implementation
[0022] To illustrate the technical content, structural features, objectives, and effects of the present invention in detail, the following description is provided in conjunction with the embodiments and accompanying drawings.
[0023] refer to Figure 5 The present invention discloses a prediction method for the delivery of emergency location messages to Internet of Things devices based on low-orbit satellites, including steps S11 to S18.
[0024] refer to Figure 1 Using shipping containers as IoT devices, this invention illustrates the integrated space-ground communication network into which the IoT devices are integrated. Two types of links can be observed: one where there are inter-satellite links between low-Earth orbit (LEO) satellites, and one where there are no inter-satellite links. The difference lies in the fact that satellites with inter-satellite links can communicate directly with the ground via transparent forwarding technology, while those without inter-satellite links communicate with the ground via satellite store-and-forward technology.
[0025] S11, construct an undirected node topology graph based on the nodes of IoT devices, emergency response center, gateway station, satellite monitoring station and satellite, and obtain the link information corresponding to the undirected node topology graph, wherein the satellite includes low-orbit satellites; refer to Figure 2This is an undirected node topology graph of the components of the present invention. Step S11 specifically includes: acquiring ephemeris data in the integrated space-ground communication network accessed by the IoT device; acquiring the location data of the IoT device, emergency response center, gateway station, satellite monitoring station, and satellite in the integrated space-ground communication network; establishing corresponding nodes for the IoT device, emergency response center, gateway station, satellite monitoring station, and satellite based on the location data and the ephemeris data; generating the undirected node topology graph based on all the nodes; the undirected node topology graph includes ephemeris data of satellite nodes, location data of gateway station nodes, node data of emergency response center, node data of satellite monitoring station, and location data of IoT device nodes collected and statistically analyzed by a third party.
[0026] This can be achieved by calling the Network X toolkit to create an undirected node topology graph based on all nodes. This graph can then be visualized using the Python language.
[0027] The step of obtaining the link information corresponding to the undirected node topology graph includes: traversing all edges in the undirected node topology graph; determining the network service parameters between the two nodes connected by each edge; and using all the network service parameters as the link information.
[0028] S12, taking the undirected node topology graph and the link information as the state, taking the message transmission path from the IoT device to the emergency response center as the action, obtaining the state and action data from the database, and using the state as input and the action as output to train the parameters of the DQN network model to obtain a trained deep Q network large model.
[0029] Both the large deep Q-network model and the small deep network model are DQN network models, and the structure of the DQN network model is as follows: Figure 3 As shown. (Refer to...) Figure 3The DQN network model includes a prediction network module, a task queue module, a loss function module, a target network module, an evaluation network module, an experience replay pool module, a reward module, an action module, a state module, and an environment module. The prediction network module obtains the state S from observations of the environment and iteratively trains it using feedback from the evaluation network. The input is state S, which contains the network model to be trained. The relevant parameters of the network model are shown in Table 1. The data flow queue module provides the function of processing computational tasks (here, the queue can be modeled as M / M / 1). To further reduce the correlation of input data, the target network module is used to reduce TD error, and the evaluation network module provides the calculation of the difference between the prediction network and the target network. The experience replay pool provides storage for experience tuples used for training. DQN is based on MDP, which includes state, action, transition function, and reward. The environment is constructed based on dual-connection link information (latency, packet loss rate, bandwidth, and service priority) and the previously built undirected node topology environment (using ephemeris data collected and statistically analyzed by third parties, gateway station node location data, 5G base station node location data along high-speed rail lines, link data between the core network and other nodes, mobile edge computing device node location data, and train UE node location data). Figure 2 The diagram shown is an undirected node topology.
[0030] The prediction network module provides the functionality for policy computation and selection. The DQN network model consists of an input layer, a fully connected layer, an activation function layer, and an output layer. The entire DQN network model is divided into four layers. The first layer includes a fully connected layer and an activation function layer. The fully connected layer uses the `nn.Linear` function from PyTorch to perform the following transformation on the input data: y = xAT + b. The activation function uses the ReLU function. The input size of this layer is the size of `inputs`, and the output size is `val`. The second layer includes a fully connected layer and an activation function layer, with both input and output sizes of `val`. The third layer is the same as the second layer. The fourth layer is the output layer, containing a fully connected layer, with an input size of `val` and an output size of `output`. The working principle of the prediction network module is shown in the following pseudocode: Input:State st Output: π(s) 1. Initialize Q-function Q, target Q-function Q^=Q 2 for each time step t 3 states (st), take action based on Q (based on epsilon greedy). 4. Receive reward rt, and proceed to the next state st+1. 5. Store (st, at, rt, st+1) into the buffer. 6. Randomly sample batch data (si, ai, ri, si+1) from the buffer. 7
[0031] 8. Update the parameters of Q so that Q(si,ai) approximates y(regression). 9. Reset Q^=Q every C steps.
[0032] S13, the deep Q network large model is distilled as a teacher model to obtain a deep Q network small model that corresponds one-to-one with each group of low-Earth orbit satellites and uses the low-Earth orbit satellite as the starting satellite as a student policy network, and the student policy network is stored in the corresponding low-Earth orbit satellite, where the starting satellite is the first low-Earth orbit satellite to which the message sent by the Internet of Things device is transmitted. refer to Figure 4 Suppose there is a set of m task sources: S1 ... Sm, where each task source comes from a teacher DQN network model. Based on different starting satellite nodes, the state, starting satellite identifier, and output (target output) are placed into the corresponding experience replay pool, denoted by QTi, where i = 1 ... m. The goal is to train the student policy network QS. Each student policy network corresponds to a unique starting satellite node and is trained independently. Data is stored in its own separate experience replay pool. The loss function distillation module learns sequentially from these N data storage units, calculates the loss function, and updates the corresponding student policy network. Each task has a separate output layer. During training and evaluation, the task ID (starting satellite identifier) needs to be constantly marked to switch to the correct output.
[0033] Step S13 specifically includes: the large deep Q-network model stores several task sources, each task source having a starting satellite identifier, state, and action; setting an experience replay pool corresponding to each low-Earth orbit satellite; placing the task source into the experience replay pool corresponding to its starting satellite; using the large deep Q-network model as the teacher model to establish a small deep Q-network model corresponding to each group of low-Earth orbit satellites; inputting the state and starting satellite identifier of the task source to the corresponding small deep Q-network model to predict the output action; obtaining the actions stored in the experience replay pool from the experience replay pool through a distillation loss function model; calculating the minimum loss based on the predicted action and the stored action; optimizing the parameters in the small deep Q-network model based on the minimum loss; iteratively learning to complete the training of the small deep Q-network model, thereby obtaining the student policy network.
[0034] The distillation loss function model is based on the formula
[0035] To optimize the parameters in the k-th deep Q-network small model The empirical replay pool corresponding to the k-th deep Q-network small model is the k-th empirical replay pool, f() is the softmax function, and τ represents the temperature applied to the softmax function for smoothing. For the action of the i-th policy stored in the k-th experience replay pool, The action predicted by the i-th policy of the k-th deep Q-network mini-model. = ( ), This is a dataset generated by a teacher model, where si represents the observation sequence. The values of actions are composed of vectors of unnormalized Q-values in the teacher model. For the parameters of the k-th deep Q-network mini-model, This is the KL divergence loss. The Q-value refers to the Q-value of the teacher model.
[0036] Model distillation is a method for transferring knowledge from a teacher model T to a student model S. The output distribution generated by the teacher network is used to form a more informative target for the student to learn. Specifically, the output of the teacher model Q network is converted into a probability distribution of actions using softmax, and then passed to the untrained student. Supervised regression is then used to train the target network (student), thereby producing the same output distribution as the original network (teacher).
[0037] S14, the IoT device node sends an emergency positioning message containing its own location information to an overhead low-Earth orbit satellite.
[0038] S15, after receiving the emergency positioning message, the overhead low-orbit satellite predicts the message transmission path of the emergency positioning message through the stored student policy network to obtain the optimal transmission path.
[0039] The actual transmission path of the emergency location message is the optimal transmission path predicted by the overpassing low-Earth orbit satellite through its own student policy network. The emergency location message is generated and sent by the IoT device. After receiving the emergency location message, the overpassing low-Earth orbit satellite operates its internally stored student policy network according to its space-ground integrated communication network to obtain the optimal transmission path. Following this optimal transmission path, it sends the emergency location message outward until it reaches the emergency response center. The emergency location message travels from the IoT device to the overpassing low-Earth orbit satellite along the optimal transmission path to the gateway node or satellite monitoring station node, and then from the gateway node or satellite monitoring station node to the emergency response center, thus completing the communication process.
[0040] S16, the overhead low-orbit satellite calculates the time required to deliver the emergency positioning message based on the optimal transmission path to obtain the predicted delivery time.
[0041] In step S16, the propagation delay between nodes in the optimal delivery path is calculated, and the propagation delay between all nodes in the optimal delivery path is summed to obtain the predicted delivery time.
[0042] All delays in the optimal transmission path include sending delay, propagation delay, processing delay, and queuing delay. Since messages have the highest priority in emergency scenarios, queuing delay can be ignored. Similarly, emergency messages are short messages, and sending and processing delays can be ignored. Therefore, the delays in this invention mainly refer to the propagation delay between nodes.
[0043] S17, the overhead low-orbit satellite returns the predicted delivery time to the IoT device node.
[0044] S18, the IoT device node receives the predicted delivery time and displays it.
[0045] The present invention also discloses a predictive device for the delivery of emergency location messages to Internet of Things (IoT) devices based on low-Earth orbit (LEO) satellites, comprising: a processor and a memory, wherein the memory is used to store at least one executable instruction, and the executable instruction causes the processor to perform the operation of the predictive method for the delivery of emergency location messages to IoT devices based on LEO satellites as described above.
[0046] The present invention also discloses a computer storage medium storing at least one executable instruction, which causes a processor to perform the operation corresponding to the prediction method for emergency location message delivery to Internet of Things devices based on low-Earth orbit satellites as described above.
[0047] AMF: Access and Mobility Management Function 3GPP: 3rd Generation Partnership Project UE: User Equipment RAN: Radio Access Network AF: Application Function UDM: Unified Data Management NEF: Network Exposure Function NWDAF: Network Data Analytics Function DQN: Deep Q-network It should be recognized that embodiments of the present invention can be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium. The method can be implemented using standard programming techniques—including implementation in a computer program on a non-transitory computer-readable storage medium configured to allow the computer to operate in a specific and predefined manner—according to the methods and drawings described in the specific embodiments. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit (ASIC).
[0048] Furthermore, the procedures described in this embodiment can be performed in any suitable order unless otherwise indicated by this embodiment or clearly contradicted by the context. The procedures (or variations and / or combinations thereof) described in this embodiment can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program includes a plurality of instructions executable by one or more processors.
[0049] Furthermore, the method can be implemented in any suitable type of computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices, etc. Aspects of the invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it is readable by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein. Furthermore, the machine-readable code, or portions thereof, can be transmitted via wired or wireless networks. The invention described in this embodiment includes these and other different types of non-transitory computer-readable storage media when such media comprises instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. When programmed according to the methods and techniques described in the invention, the invention also includes the computer itself.
[0050] A computer program can be applied to input data to perform the functions described in this embodiment, thereby transforming the input data to generate output data stored in non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including specific visual depictions of physical and tangible objects generated on the display.
[0051] The above-disclosed embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.
Claims
1. A predictive method for emergency location message delivery to IoT devices based on low-Earth orbit satellites, characterized in that: include: An undirected node topology graph is constructed based on the nodes of IoT devices, emergency response centers, gateway stations, satellite monitoring stations, and satellites, including low-orbit satellites, and the link information corresponding to the undirected node topology graph is obtained. The state is taken as input from the undirected node topology graph and the link information, and the action is taken as output from the message passing path between the IoT device and the emergency response center. A large deep Q network model is trained. Distill the deep Q-network large model as a teacher model to obtain a student policy network that corresponds one-to-one with each group of low-Earth orbit satellites and uses the low-Earth orbit satellite as the starting satellite. Store the student policy network in the corresponding low-Earth orbit satellite. The starting satellite is the first low-Earth orbit satellite to which the message sent by the Internet of Things device is transmitted. When the IoT device sends an emergency location message to an overhead low-Earth orbit satellite, the overhead low-Earth orbit satellite predicts the message delivery path of the emergency location message through a stored student policy network. Based on the aforementioned message transmission path, the time required for the emergency location message to reach the emergency response center is calculated to obtain the predicted delivery time. The predicted delivery time is returned to the IoT device; The IoT device receives and displays the predicted delivery time.
2. The method for predicting the delivery of emergency location messages as described in claim 1, characterized in that: The predicted delivery time is obtained by calculating the time required for the emergency location message to reach the emergency response center based on the message transmission path. Specifically, the propagation delay between all nodes in the message transmission path is summed to obtain the predicted delivery time.
3. The method for predicting the delivery of emergency location messages as described in claim 1, characterized in that: Distilling the deep Q-network large model as the teacher model to obtain the student policy network that corresponds one-to-one with each group of low-Earth orbit satellites and uses the low-Earth orbit satellites as the starting satellites specifically includes: The deep Q-network large model stores several task sources. Each task source has a starting satellite identifier, state, and corresponding action of the starting satellite. An experience replay pool is set for each low-orbit satellite, and the task source is placed into the experience replay pool corresponding to its starting satellite. Using the aforementioned large-scale deep Q-network model as the teacher model, establish small-scale deep Q-network models corresponding one-to-one with each group of low-Earth orbit satellites; The state and starting satellite identifier in the task source are fed as input to the corresponding deep Q-network mini-model to predict the output action. The actions stored in the experience replay pool are obtained from the experience replay pool through the distillation loss function model. The loss is minimized based on the predicted action and the stored action. The parameters in the deep Q-network mini-model are optimized based on the minimized loss. The deep Q-network mini-model is trained iteratively to obtain the student policy network.
4. The method for predicting the delivery of emergency location messages as described in claim 1, characterized in that: An undirected node topology graph is constructed based on the nodes of IoT devices, emergency response centers, gateway stations, satellite monitoring stations, and satellites, including: Obtain ephemeris data from the integrated space-ground communication network accessed by the IoT device; Acquire the location data of the IoT devices, emergency response center, gateway station, satellite monitoring station, and satellite in the integrated space-ground communication network; Based on the location data and ephemeris data, establish corresponding nodes for the IoT devices, emergency response center, gateway station, satellite monitoring station, and satellite; Generate the undirected node topology graph based on all the nodes; The undirected node topology graph includes satellite node ephemeris data, gateway station node location data, emergency response center node data, satellite monitoring station node data, and IoT device node location data collected and statistically analyzed by a third party.
5. The method for predicting the delivery of emergency location messages as described in claim 4, characterized in that: The step of generating the undirected node topology graph based on all the nodes includes: calling the Network X toolkit to create the undirected node topology graph based on all the nodes.
6. The method for predicting the delivery of emergency location messages as described in claim 4, characterized in that: The step of obtaining the link information corresponding to the undirected node topology graph includes: traversing all edges in the undirected node topology graph; determining the network service parameters between the two nodes connected by each edge; and using all the network service parameters as the link information.
7. The method for predicting the delivery of emergency location messages as described in claim 1, characterized in that: Both the large and small deep Q-network models are DQN network models. The DQN network model consists of four layers: the first layer is the input layer, which includes a fully connected layer and an activation function layer; the second and third layers each include a fully connected layer and an activation function layer; and the fourth layer is the output layer, which includes a fully connected layer. The fully connected layer uses the nn.Linear function in PyTorch, and the activation function layer uses the ReLU function.
8. A predictive device for delivering emergency location messages to Internet of Things (IoT) devices based on low-Earth orbit satellites, characterized in that: include: The processor and the memory, wherein the memory is used to store at least one executable instruction, the executable instruction causing the processor to perform the operation of the prediction method for emergency location message delivery to Internet of Things devices based on low-orbit satellites as described in any one of claims 1-7.
9. A computer storage medium, characterized in that: The storage medium stores at least one executable instruction that causes the processor to perform the operation corresponding to the prediction method for emergency location message delivery to IoT devices based on low-orbit satellites as described in any one of claims 1-7.