Ad hoc network information processing method for massive Internet of Things devices

Abstract

The invention relates to the technical field of information processing, in particular to an ad hoc network information processing method for massive Internet of Things equipment, which comprises the following steps: acquiring a network topology structure of a target ad hoc network, and performing graph structure modeling on the network topology structure to obtain a network undirected graph corresponding to the network topology structure; and calculating the node weight of each node in the network undirected graph based on the node position of each node in the network undirected graph, and calculating the node label probability of each node in the network undirected graph based on the node weight. According to the method, the importance of the network nodes is evaluated by calculating the weights of the nodes and based on the node label probability, so that the whole network can be divided into a plurality of control domains according to the characteristics of different nodes, the network is divided into a plurality of sub-target ad hoc networks, each control domain is independently managed, and the reliability of the network is improved. The division improves the flexibility of network management, so that the management of different areas in the network can be finer.

Description

Technical Field

[0001] The present invention relates to the field of information processing technology, and in particular to a self-organizing network information processing method for massive Internet of Things devices. Background Art

[0002] Massive IoT devices refer to a huge number of physical devices connected through the Internet. These devices usually have low power consumption, low data transmission requirements, low bandwidth requirements, and have intelligent processing and sensing capabilities. With the development of IoT technology, especially in the fields of smart cities, industrial automation, smart homes, smart agriculture, etc., the scale of massive IoT devices is expanding.

[0003] Traditional methods usually adopt static network topology and cannot dynamically adapt to changes in the network environment. In actual Internet of Things (IoT) applications, the number of devices is huge and constantly changing. Traditional methods are often unable to effectively cope with such dynamic changes, resulting in inefficient and non-scalable network management. Traditional methods often rely on simple global optimization in network management, and cannot finely divide different areas for independent management. The role of each node and the importance of the network are often not fully evaluated, resulting in chaotic management levels and inefficiency. Traditional methods often rely on preset rules or fixed control strategies for network management and congestion control, lacking the ability of real-time prediction and dynamic adjustment. Once the network is congested, it is often difficult to identify in advance and take effective countermeasures, resulting in decreased network performance. Traditional methods are usually based on simple resource allocation rules, and fail to fully consider the complex relationship between nodes and the optimization of resource utilization. When facing a network with a large number of devices, traditional methods often lead to unbalanced, wasteful or insufficient resource scheduling, affecting network throughput and device energy efficiency. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a self-organizing network information processing method for massive Internet of Things devices.

[0005] The technical solution adopted to solve the above technical problems is: a self-organizing network information processing method for massive IoT devices, including:

[0006] Acquire the network topology structure of the target ad hoc network, and perform graph structure modeling on the network topology structure to obtain a network undirected graph corresponding to the network topology structure;

[0007] Calculating a node weight of each node in the network undirected graph based on the node position of each node in the network undirected graph, and calculating a node label probability of each node in the network undirected graph based on the node weight;

[0008] Calculating a penalty coefficient of each node in the network undirected graph based on the node label probability, and comparing the penalty coefficient of each node with a preset penalty threshold table to divide the target ad hoc network into a plurality of control domains, that is, dividing the target ad hoc network into a plurality of sub-target ad hoc networks, and the sub-target ad hoc networks correspond to the control domains;

[0009] Taking the control domain as a unit, monitoring the status of the sub-target self-organizing network in the control domain according to a preset monitoring period to obtain status data corresponding to the sub-target self-organizing network;

[0010] Performing congestion prediction on the state data based on a pre-trained congestion prediction model to obtain a congestion level label corresponding to the sub-target ad hoc network;

[0011] A congestion control model is constructed based on the state data and congestion level label corresponding to the sub-target self-organizing network, and the congestion control model is solved based on a reinforcement learning algorithm to obtain a congestion control solution corresponding to the sub-target self-organizing network.

[0012] Preferably, the network undirected graph includes a node set and an edge set, wherein the nodes in the node set correspond to Internet of Things devices, and the edges in the edge set correspond to communication links between two Internet of Things devices.

[0013] Preferably, the calculation formula of the node weight of each node in the network undirected graph is as follows:

[0014]

[0015] Among them, W i represents the node weight of the i-th node in the network undirected graph, TL i represents the maximum value of the delay between the ith node and all its neighboring nodes in the network undirected graph, and v j represents the jth node in the network undirected graph, N S Represents the set of neighbor nodes of the i-th node in the network undirected graph, d ij It represents the distance between the i-th node and the j-th node in the network undirected graph, and ρ represents the propagation speed of electromagnetic waves.

[0016] Preferably, the calculation formula for the node label probability of each node in the network undirected graph is as follows:

[0017]

[0018] Among them, l i represents the node label, and l i =0,1,P(l i =1) indicates node label li The node label probability is 1, and P(l i =0) indicates node label l i The node label probability is 0, and P(l i =0) = 1-P(l i =1), W min and W max Indicates the minimum and maximum values ​​of the node weight.

[0019] Preferably, calculating the penalty coefficient of each node in the network undirected graph based on the node label probability includes:

[0020] A node association function is constructed based on the node weights, wherein the node association function is as follows:

[0021]

[0022] Among them, AC(i,j) represents the node association between two nodes in the network undirected graph, η represents the distance coefficient, and Rt represents the communication range of the node, σ represents the standard deviation of the node weight;

[0023] A node association penalty function is constructed based on the node association function, wherein the association penalty function is as follows:

[0024]

[0025] Among them, ACP i represents the node association penalty coefficient of the i-th node in the network undirected graph, V represents the node set of the network undirected graph, δ(l i ,l j ) represents the association indicator function, and

[0026] A node penalty function is constructed based on the node label probability, wherein the node penalty function is as follows:

[0027] SCP i =-ln(P(l i = l j ));

[0028] Among them, SCP i represents the node penalty coefficient of the i-th node in the network undirected graph, P(l i = l j ) represents the probability that the node label of the i-th node in the network undirected graph belongs to the real node label;

[0029] The penalty function of the network undirected graph is constructed based on the node's own penalty function and the node-associated penalty function, wherein the penalty function of the network undirected graph is as follows:

[0030] LF i =β·SCP i +ACP i ;

[0031] Among them, LF i It represents the penalty coefficient of the ith node in the network undirected graph, and β represents the preset important factor.

[0032] Preferably, the expression of the state data corresponding to the sub-target ad hoc network is as follows:

[0033] SD = {RTT avg ,Ack num ,Thr avg ,cwnd cl ,v,RTT,MinRTT,Jitter e ,loss};

[0034] Among them, SD represents the state data corresponding to the sub-target self-organizing network, RTT avg Indicates the average packet delay during the monitoring period, Ack num Indicates the number of valid confirmation messages within the monitoring period, Thr avg Indicates the average throughput during the monitoring period, cend cl Indicates the current congestion window size during the monitoring period, v indicates the data transmission rate at the end of the monitoring period, RTT indicates the last RTT during the monitoring period, MinRTT indicates the minimum RTT within the recent preset time, and Jitter e It indicates the round-trip delay jitter value, and loss indicates the communication link packet loss rate.

[0035] Preferably, performing congestion prediction on the state data based on a pre-trained congestion prediction model to obtain a congestion level label corresponding to the sub-target ad hoc network includes:

[0036] Feature extraction is performed on the state data to obtain a congestion feature set, wherein the expression of the congestion feature set is as follows:

[0037]

[0038] Among them, FS represents the congestion feature set;

[0039] Congestion prediction is performed on the congestion feature set based on a pre-trained congestion prediction model to obtain a congestion level label corresponding to the sub-target ad hoc network, wherein the congestion prediction model adopts a support vector machine model.

[0040] Preferably, constructing a congestion control model based on the state data and congestion level label corresponding to the sub-target ad hoc network includes:

[0041] Constructing the state space of the congestion control model based on the state data corresponding to the sub-target ad hoc network;

[0042] Constructing a reward function of the congestion control model based on the state data and congestion level label corresponding to the sub-target ad hoc network;

[0043] The action space of the congestion control model is constructed based on all communication links of the sub-target ad hoc network. Preferably, the state space of the congestion control model is expressed as follows:

[0044] s t =(SD t ,SD t-1 ,...,SD t-k );

[0045] Among them, s t represents the state space of the congestion control model, SD t Represents the state data corresponding to the sub-target ad hoc network at time t;

[0046] The reward function of the congestion control model is as follows:

[0047]

[0048] Among them, reward represents the reward function of the congestion control model, p represents the congestion level corresponding to the sub-target ad hoc network, and p th Indicates the preset congestion level threshold.

[0049] Preferably, solving the congestion control model based on a reinforcement learning algorithm to obtain a congestion control solution corresponding to the sub-target ad hoc network includes:

[0050] Through the monitored state space s t , select actions from the action space according to the strategy π and execute them in the state space environment, thereby receiving a reward and entering the next state space s t+1 , the policy π is the state space s t Mapping to action space;

[0051] Based on the reinforcement learning algorithm, the accumulated reward is maximized to obtain the strategy π corresponding to the maximum reward, wherein the strategy π is the congestion control scheme corresponding to the sub-target self-organizing network, wherein the learning formula of the accumulated reward is as follows:

[0052]

[0053] Among them, E t represents the accumulated reward, and T represents the number of iterations of the reinforcement learning algorithm.

[0054] The beneficial effects of the present invention are as follows: (1) The present invention obtains an undirected graph representation by graph structure modeling of the network topology of the self-organizing network. This modeling method can clearly describe the network nodes and connection relationships, making subsequent management and optimization more controllable, and by calculating the weight of the node and evaluating the importance of the network node based on the node label probability, this can help divide the entire network into multiple control domains according to the characteristics of different nodes, and by dividing the network into multiple sub-target self-organizing networks, each control domain is managed independently and matched with its corresponding sub-target self-organizing network. This division improves the flexibility of network management and makes the management of different areas in the network more refined; (2) The present invention uses a pre-trained congestion prediction model to predict the congestion level of the network status data, so that The congestion situation of each sub-target self-organizing network is understood in real time, and measures are taken in advance to avoid the occurrence of congestion. Based on the congestion prediction, the congestion control model is solved by the reinforcement learning algorithm. This means that the congestion control strategy can be dynamically optimized according to the real-time network status during the continuous learning process, thereby reducing network congestion, increasing throughput, and improving the performance of the overall network; (3) The present invention can make adaptive adjustments according to the changing network status through the reinforcement learning algorithm. When the network conditions change, the model can automatically adjust the control scheme, avoiding the limitations of traditional static methods. In addition, since the characteristics of self-organizing networks are that the number of devices is large and widely distributed, the use of graph structure modeling and control domain division methods can effectively manage each sub-target self-organizing network independently, thereby ensuring stable operation in the case of massive devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic flow chart of the steps of an overall method in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] Embodiment 1, as Figure 1 As shown, the self-organizing network information processing method for massive IoT devices proposed by the present invention includes:

[0057] S1. Obtain the network topology of the target ad hoc network, and perform graph structure modeling on the network topology to obtain an undirected network graph corresponding to the network topology;

[0058] S2, calculating the node weight of each node in the network undirected graph based on the node position of each node in the network undirected graph, and calculating the node label probability of each node in the network undirected graph based on the node weight;

[0059] S3, calculating the penalty coefficient of each node in the network undirected graph based on the node label probability, and comparing the penalty coefficient of each node with the preset penalty threshold table to divide the target ad hoc network into multiple control domains, that is, dividing the target ad hoc network into multiple sub-target ad hoc networks, and the sub-target ad hoc networks correspond to the control domains;

[0060] S4, taking the control domain as a unit, monitoring the status of the sub-target self-organizing network in the control domain according to a preset monitoring period to obtain status data corresponding to the sub-target self-organizing network;

[0061] S5. Perform congestion prediction on the state data based on the pre-trained congestion prediction model to obtain a congestion level label corresponding to the sub-target ad hoc network;

[0062] S6. Construct a congestion control model based on the state data and congestion level label corresponding to the sub-target self-organizing network, and solve the congestion control model based on the reinforcement learning algorithm to obtain the congestion control solution corresponding to the sub-target self-organizing network.

[0063] In the present invention, an ad hoc network is a wireless network structure, in which devices (nodes) in the network can establish, maintain and manage connections by themselves without relying on fixed infrastructure or central control. Each node not only acts as a data source, but also forwards data from other nodes. Therefore, the state of the node and the network topology change dynamically. Network topology refers to the connection mode between each node in the network and between them. In the Internet of Things, the topology structure describes how devices (such as sensors, terminals, etc.) are connected and interact. The topology may be wired or wireless, and may be fixed or dynamically changing. The network topology can be represented by graph theory, in which nodes represent devices or sensors in the network, and edges represent communication connections between nodes. Graph structure modeling uses the concept of graph theory to represent the network topology. The network topology is converted into an undirected graph, that is, the connection between nodes has no directionality, indicating the relationship between devices communicating with each other. The node label is an identifier of the node, which may indicate the state or category of the node. The node label probability refers to the probability that each node belongs to a specific category (for example, an important node, a low-power node, etc.). These labels may affect the processing and communication strategies of the node. The penalty threshold table is a preset standard that defines thresholds corresponding to different penalty coefficients to help decide how to divide the network. If the penalty coefficient of a node exceeds the threshold, the node is marked as undirected. value, then it may be divided into different sub-networks (or control domains); a control domain refers to a certain area or sub-network in the network, usually composed of nodes with certain common characteristics (such as similar functions, tasks or distances). Each control domain has a certain degree of autonomy and can centrally process control tasks, monitoring data, etc.; a sub-target ad hoc network is a sub-network after the target ad hoc network is divided. Each sub-target ad hoc network represents a collection of devices in a specific area or sub-domain of the network. These sub-networks or sub-target ad hoc networks are usually divided according to the control domain. The nodes in each sub-target ad hoc network can independently transmit and manage data; the congestion prediction model is based on historical data and the prediction model of the current state of the network, which is used to predict the network congestion that may occur in a certain period of time in the future; the monitoring period refers to the time interval at which the system regularly monitors the status of a certain network domain (such as the sub-target self-organizing network in the control domain); the congestion control model is a control algorithm or strategy used to adjust the resource allocation, load balancing, routing strategy, etc. in the network to alleviate or solve the problem of network congestion. By making real-time adjustments to the network status (such as bandwidth, delay, load, etc.), the transmission efficiency of the network can be improved; reinforcement learning is a machine learning method in which the agent learns how to maximize the cumulative reward by selecting actions through interaction with the environment.

[0064] Embodiment 2, the self-organizing network information processing method for massive Internet of Things devices proposed in the present invention, compared with embodiment 1, this embodiment also includes: the network undirected graph includes a node set and an edge set, wherein the nodes in the node set correspond to the Internet of Things devices, and the edges in the edge set correspond to the communication links between two Internet of Things devices.

[0065] In this embodiment, an undirected graph is a graph structure in which each edge has no direction, which means that the two endpoints of the edge are equal. The nodes represent devices, and the edges represent communication links between devices. The characteristic of an undirected graph is that if node A has an edge with node B, then node B can also communicate with node A in the same way without directional restrictions.

[0066] In an optional embodiment, the calculation formula of the node weight of each node in the network undirected graph is as follows:

[0067]

[0068] Among them, W i represents the node weight of the i-th node in the network undirected graph, TL i represents the maximum value of the delay between the ith node and all its neighboring nodes in the network undirected graph, and v j represents the jth node in the network undirected graph, N S Represents the set of neighbor nodes of the i-th node in the network undirected graph, d ij It represents the distance between the i-th node and the j-th node in the network undirected graph, and ρ represents the propagation speed of electromagnetic waves.

[0069] In an optional embodiment, the calculation formula for the node label probability of each node in the network undirected graph is as follows:

[0070]

[0071] Among them, l i represents the node label, and l i =0,1,P(l i =1) indicates node label l i The node label probability is 1, and P(l i =0) indicates node label l i The probability of a node label being 0, and P(l i =0) = 1-P(l i =1), W min and W max Indicates the minimum and maximum values ​​of the node weight.

[0072] In an optional embodiment, the penalty coefficient of each node in the network undirected graph is calculated based on the node label probability, including:

[0073] A1. Construct a node association function based on node weights, where the node association function is as follows:

[0074]

[0075] Among them, AC(i,j) represents the node association between two nodes in the network undirected graph, η represents the distance coefficient, and Rt represents the communication range of the node, σ represents the standard deviation of the node weight;

[0076] A2. Construct a node association penalty function based on the node association function, where the association penalty function is as follows:

[0077]

[0078] Among them, ACP i represents the node association penalty coefficient of the i-th node in the network undirected graph, V represents the node set of the network undirected graph, δ(l i ,l j ) represents the association indicator function, and

[0079] A3. Construct the node's own penalty function based on the node label probability, where the node's own penalty function is as follows:

[0080] SCP i =-ln(P(l i = l j ));

[0081] Among them, SCP i represents the node penalty coefficient of the i-th node in the network undirected graph, P(l i = l j ) represents the probability that the node label of the i-th node in the network undirected graph belongs to the real node label;

[0082] A4. Construct a penalty function of a network undirected graph based on the node's own penalty function and the node-related penalty function. The penalty function of the network undirected graph is as follows:

[0083] LF i =β·SCP i +ACP i ;

[0084] Among them, LF i It represents the penalty coefficient of the ith node in the network undirected graph, and β represents the preset important factor.

[0085] In an optional embodiment, the expression of the state data corresponding to the sub-target ad hoc network is as follows:

[0086] SD = {RTT avg ,Ack num ,Thr avg ,cwnd cl ,v,RTT,MinRTT,Jittere ,loss};

[0087] Among them, SD represents the state data corresponding to the sub-target self-organizing network, RTT avg Indicates the average packet delay during the monitoring period, Ack num Indicates the number of valid confirmation messages within the monitoring period, Thr avg Indicates the average throughput during the monitoring period, cwnd cl Indicates the current congestion window size during the monitoring period, v indicates the data transmission rate at the end of the monitoring period, RTT indicates the last RTT during the monitoring period, MinRTT indicates the minimum RTT within the recent preset time, and Jitter e It indicates the round-trip delay jitter value, and loss indicates the communication link packet loss rate.

[0088] In an optional embodiment, congestion prediction is performed on the state data based on a pre-trained congestion prediction model to obtain a congestion level label corresponding to the sub-target ad hoc network, including:

[0089] B1. Extract features from the state data to obtain a congestion feature set, where the expression of the congestion feature set is as follows:

[0090]

[0091] Among them, FS represents the congestion feature set;

[0092] B2. Congestion prediction is performed on the congestion feature set based on a pre-trained congestion prediction model to obtain a congestion level label corresponding to the sub-target ad hoc network, wherein the congestion prediction model adopts a support vector machine model.

[0093] It should be noted that support vector machine is a common supervised learning algorithm, which is widely used in classification and regression problems. In the congestion prediction model, support vector machine is used to perform pattern recognition and classification based on training data; the basic idea of ​​support vector machine is to separate data points of different categories by constructing an optimal hyperplane. In congestion prediction, support vector machine can classify network status into different categories (such as normal, light congestion, heavy congestion, etc.) according to the extracted congestion features; congestion level label refers to the classification label of network congestion status, indicating which congestion status the network is currently in (such as no congestion, light congestion, moderate congestion, heavy congestion, etc.).

[0094] In an optional embodiment, a congestion control model is constructed based on the state data and congestion level label corresponding to the sub-target ad hoc network, including:

[0095] C1, construct the state space of the congestion control model based on the state data corresponding to the sub-target ad hoc network;

[0096] C2, constructing the reward function of the congestion control model based on the state data and congestion level labels corresponding to the sub-target ad hoc networks;

[0097] C3. Construct the action space of the congestion control model based on all communication links of the sub-target ad hoc network.

[0098] It should be noted that in reinforcement learning, the action space refers to the set of all possible actions that the agent can take. In the congestion control model, the action space represents all operations that the model can choose to control network congestion. These actions are usually defined by the control strategy in the network.

[0099] In an optional embodiment, the state space expression of the congestion control model is as follows:

[0100] s t =(SD t ,SD t-1 ,...,SD t-k );

[0101] Among them, s t represents the state space of the congestion control model, SD t Represents the state data corresponding to the sub-target ad hoc network at time t;

[0102] The reward function of the congestion control model is as follows:

[0103]

[0104] Among them, reward represents the reward function of the congestion control model, p represents the congestion level corresponding to the sub-target ad hoc network, and p th Indicates the preset congestion level threshold.

[0105] In an optional embodiment, the congestion control model is solved based on a reinforcement learning algorithm to obtain a congestion control solution corresponding to the sub-target ad hoc network, including:

[0106] D1, through the monitored state space s t , select actions from the action space according to the strategy π and execute them in the state space environment, thereby receiving a reward and entering the next state space s t+1 , the policy π is the state space s t Mapping to action space;

[0107] D2. Based on the reinforcement learning algorithm, learn to maximize the cumulative reward to obtain the strategy π corresponding to the maximum reward, where strategy π is the congestion control scheme corresponding to the sub-goal self-organizing network. The learning formula for the cumulative reward is as follows:

[0108]

[0109] Among them, R t represents the accumulated reward, and T represents the number of iterations of the reinforcement learning algorithm.

[0110] It should be noted that reinforcement learning algorithms are algorithms used to train intelligent agents to learn strategies in a specific environment. Common reinforcement learning algorithms include Q-learning, deep Q network (DQN), policy gradient method, etc. The core idea of ​​these algorithms is to learn through trial and error, and gradually adjust the strategy in the process of interaction with the environment to maximize the cumulative reward; strategy refers to a decision rule for network congestion control. In the sub-target self-organizing network, the strategy will determine how to optimize network performance by controlling network traffic, routing selection, bandwidth allocation, etc., so as to avoid or reduce network congestion. Reinforcement learning continuously optimizes these strategies to obtain the best congestion control effect.

[0111] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A method for processing information of a self-organizing network of massive IoT devices, characterized in that: include: Acquire the network topology of the target ad hoc network, and perform graph structure modeling on the network topology to obtain a network undirected graph corresponding to the network topology; Calculating a node weight of each node in the network undirected graph based on the node position of each node in the network undirected graph, and calculating a node label probability of each node in the network undirected graph based on the node weight; Calculating a penalty coefficient of each node in the network undirected graph based on the node label probability, and comparing the penalty coefficient of each node with a preset penalty threshold table to divide the target ad hoc network into a plurality of control domains, that is, dividing the target ad hoc network into a plurality of sub-target ad hoc networks, and the sub-target ad hoc networks correspond to the control domains; Taking the control domain as a unit, monitoring the status of the sub-target self-organizing network in the control domain according to a preset monitoring period to obtain status data corresponding to the sub-target self-organizing network; Performing congestion prediction on the state data based on a pre-trained congestion prediction model to obtain a congestion level label corresponding to the sub-target ad hoc network; A congestion control model is constructed based on the state data and congestion level label corresponding to the sub-target self-organizing network, and the congestion control model is solved based on a reinforcement learning algorithm to obtain a congestion control solution corresponding to the sub-target self-organizing network.

2. The method for processing information of a self-organizing network for massive IoT devices according to claim 1, characterized in that: The network undirected graph includes a node set and an edge set, wherein the nodes in the node set correspond to Internet of Things devices, and the edges in the edge set correspond to communication links between two Internet of Things devices.

3. The method for processing information of a self-organizing network for massive IoT devices according to claim 2, characterized in that: The calculation formula of the node weight of each node in the network undirected graph is as follows: Among them, W i represents the node weight of the i-th node in the network undirected graph, TL i represents the maximum value of the delay between the ith node and all its neighboring nodes in the network undirected graph, and v j represents the jth node in the network undirected graph, N S Represents the set of neighbor nodes of the i-th node in the network undirected graph, d ij It represents the distance between the i-th node and the j-th node in the network undirected graph, and ρ represents the propagation speed of electromagnetic waves.

4. The method for processing information of a self-organizing network for massive IoT devices according to claim 3, characterized in that: The calculation formula for the node label probability of each node in the network undirected graph is as follows: Among them, l i represents the node label, and l i =0,1,P(l i =1) indicates node label l i The node label probability is 1, and P(l i =0) indicates node label l i The probability of a node label being 0, and P(l i =0) = 1-P(l i =1), W min and W max Indicates the minimum and maximum values ​​of the node weight.

5. The method for processing information of a self-organizing network for massive IoT devices according to claim 4, characterized in that: Calculating the penalty coefficient of each node in the network undirected graph based on the node label probability includes: A node association function is constructed based on the node weights, wherein the node association function is as follows: Among them, AC(i,j) represents the node association between two nodes in the network undirected graph, η represents the distance coefficient, and Rt represents the communication range of the node, σ represents the standard deviation of the node weight; A node association penalty function is constructed based on the node association function, wherein the association penalty function is as follows: Among them, ACP i represents the node association penalty coefficient of the i-th node in the network undirected graph, V represents the node set of the network undirected graph, δ(l i ,l j ) represents the association indicator function, and A node penalty function is constructed based on the node label probability, wherein the node penalty function is as follows: SCP i =-ln(P(l i =l j )); Among them, SCP i represents the node penalty coefficient of the i-th node in the network undirected graph, P(l i = l j ) represents the probability that the node label of the i-th node in the network undirected graph belongs to the real node label; The penalty function of the network undirected graph is constructed based on the node's own penalty function and the node-associated penalty function, wherein the penalty function of the network undirected graph is as follows: LF i =β·SCP i +ACP i ; Among them, LF i It represents the penalty coefficient of the ith node in the network undirected graph, and β represents the preset important factor.

6. The method for processing information of a self-organizing network for massive IoT devices according to claim 1, characterized in that: The expression of the state data corresponding to the sub-target ad hoc network is as follows: SD={RTT avg ,Ack num ,Thr avg ,cwnd cl ,v,RTT,MinRTT,Jitter e ,loss}; Among them, SD represents the state data corresponding to the sub-target self-organizing network, RTT avg Indicates the average packet delay during the monitoring period, Ack num Indicates the number of valid confirmation messages within the monitoring period, Thr avg Indicates the average throughput during the monitoring period, cwnd cl Indicates the current congestion window size during the monitoring period, v indicates the data transmission rate at the end of the monitoring period, RTT indicates the last RTT during the monitoring period, MinRTT indicates the minimum RTT within the recent preset time, and Jitter e It indicates the round-trip delay jitter value, and loss indicates the communication link packet loss rate.

7. The method for processing information of a self-organizing network for massive IoT devices according to claim 6, characterized in that: Congestion prediction is performed on the state data based on a pre-trained congestion prediction model to obtain a congestion level label corresponding to the sub-target ad hoc network, including: Feature extraction is performed on the state data to obtain a congestion feature set, wherein the expression of the congestion feature set is as follows: Among them, FS represents the congestion feature set; Congestion prediction is performed on the congestion feature set based on a pre-trained congestion prediction model to obtain a congestion level label corresponding to the sub-target ad hoc network, wherein the congestion prediction model adopts a support vector machine model.

8. The method for processing information of a self-organizing network for massive IoT devices according to claim 7, characterized in that: A congestion control model is constructed based on the state data and congestion level label corresponding to the sub-target ad hoc network, including: Constructing the state space of the congestion control model based on the state data corresponding to the sub-target ad hoc network; Constructing a reward function of the congestion control model based on the state data and congestion level label corresponding to the sub-target ad hoc network; The action space of the congestion control model is constructed based on all communication links of the sub-target ad hoc network.

9. The method for processing information of a self-organizing network for massive IoT devices according to claim 8, characterized in that: The state space expression of the congestion control model is as follows: s t (SD) t ,SD t-1 ,...,SD t-k ); Among them, s t represents the state space of the congestion control model, SD t Represents the state data corresponding to the sub-target ad hoc network at time t; The reward function of the congestion control model is as follows: Among them, reward represents the reward function of the congestion control model, p represents the congestion level corresponding to the sub-target ad hoc network, and p th Indicates the preset congestion level threshold.

10. The method for processing information of a self-organizing network for massive IoT devices according to claim 9, characterized in that: The congestion control model is solved based on a reinforcement learning algorithm to obtain a congestion control solution corresponding to the sub-target ad hoc network, including: Through the monitored state space s t , select actions from the action space according to the strategy π and execute them in the state space environment, thereby receiving a reward and entering the next state space s t+1 , the policy π is the state space s t Mapping to action space; Based on the reinforcement learning algorithm, the accumulated reward is maximized to obtain the strategy π corresponding to the maximum reward, wherein the strategy π is the congestion control scheme corresponding to the sub-target self-organizing network, wherein the learning formula of the accumulated reward is as follows: Among them, E t represents the accumulated reward, and T represents the number of iterations of the reinforcement learning algorithm.

Images