Media access control (MAC) protocol reconstruction method and device based on task driving, equipment and medium

Abstract

The invention relates to an MAC protocol reconstruction method and device based on task driving, equipment and a medium. The method comprises the following steps: on one hand, considering the difference of service features brought by node task heterogeneity, finely splitting a network load into features of three dimensions, and fully reflecting the performance of an MAC protocol in different network environments by a constructed network load model; on the other hand, a closed-loop software-defined distributed subnet node system architecture is constructed, the architecture provides a scheme that distributed nodes cooperatively complete a reconstruction process, different task requirements are considered in the reconstruction process to solve the problem of consideration of protocol performance, a data set is constructed, and reconstruction decisions of different protocols are realized by utilizing a classification model; finally, the optimal MAC protocol based on task requirements and considering different performance dimensions can be obtained through reconstruction, and the adaptability and reliability of the MAC protocol to the dynamically changing network environment and task requirements are improved.

Description

Technical Field

[0001] The present application relates to the field of network communication technology, and in particular to a task-driven MAC protocol reconstruction method, device, equipment and medium. Background Art

[0002] Wireless Ad Hoc Network (WAN) is a distributed network that does not rely on fixed infrastructure. Its nodes can move freely and automatically configure connections. It has the characteristics of being decentralized, self-organizing, and having a dynamic topology. It is suitable for scenarios such as emergency response and disaster relief. Research on WAN has received widespread attention in recent years.

[0003] The multiple access control (MAC) protocol plays a vital role in the process of networking and information transmission, and directly affects the communication performance of the network. Generally speaking, MAC protocols suitable for wireless ad hoc networks can be divided into contention type and scheduling type. Contention type MAC protocols obtain transmission opportunities by preempting channels, and scheduling type protocols use channel division to orderly access channels. The essential difference between the two types of protocols lies in the different channel utilization rates under different network conditions. Poisson distribution is usually used to model the service arrival rate. It is believed that the use of contention type protocols in low-load environments and scheduling type protocols in high-load environments can give full play to the respective advantages of the two types of protocols.

[0004] However, due to the essential differences in the access mechanisms of MAC protocols, there is no protocol that meets all network environments and requirements. In practical applications, the dynamic changes in network environments and the time-varying nature of task requirements pose challenges to a single fixed MAC protocol. Therefore, an intelligent MAC protocol reconstruction algorithm is very necessary for dynamic network environments. Summary of the invention

[0005] Based on this, it is necessary to provide a task-driven MAC protocol reconstruction method, device, equipment and medium to address the above technical problems. By remodeling the network load, a network load model based on task characteristics is constructed, and on this basis, the MAC protocol is reconstructed, and a closed-loop distributed node system architecture is designed to reconstruct the optimal MAC protocol based on task requirements and taking into account different performance dimensions.

[0006] A task-driven MAC protocol reconstruction method, the method comprising: Considering the multi-domain distributed wireless self-organizing network scenario, a network load model is constructed through multi-dimensional feature extraction, Gaussian random process modeling and feature fusion of network load. Based on the network load model, a software-defined distributed subnet node system architecture is constructed, which consists of the task plane, data plane and joint control plane. Among them, the task plane is used to cache the communication services generated during the execution of node tasks and cluster tasks in the form of data frames to the data plane to form a queue. The data plane queue contains the service data frames generated by the node tasks, and also contains the control data frames generated by the feature interaction during the MAC protocol reconstruction process and the channel reservation and preemption behavior in the MAC protocol process; The joint control plane includes the observation plane, the knowledge plane and the control plane. The observation plane is used to observe and count the services of each node, obtain the service characteristics of each node in the current reconstruction cycle through maximum likelihood estimation, and obtain the service characteristics of each node in the next reconstruction cycle through Kalman filtering prediction. Then, the service characteristics of all single-hop reachable nodes obtained by broadcast interaction are integrated to obtain the network characteristics of the next reconstruction cycle and pass them to the knowledge plane. The knowledge plane is used to carry the pre-trained classification model and output the optimal MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane based on the network features input by the observation plane and the task requirements input by the task plane. The control plane is used to carry the MAC protocol component library and determine whether the reconstruction strategy output by the knowledge plane is the same as the MAC protocol of the current reconstruction cycle. If they are the same, the reconstruction process is not performed; if they are not the same, the components are searched and the protocol parameters are adjusted to complete the MAC protocol component-level reconstruction and enter the next reconstruction cycle.

[0007] In one of the embodiments, a multi-domain distributed wireless self-organizing network scenario is considered, and a network load model is constructed by extracting multi-dimensional features of the network load, Gaussian random process modeling, and feature fusion, including: Considering the multi-domain distributed wireless self-organizing network scenario, the multi-dimensional characteristics of the network load are divided and extracted, including the network node scale, node activity level and node business characteristics; The service characteristics of each node in the network are modeled using a multi-dimensional Gaussian random process with the same dimension, and the service characteristics of all nodes in the network are weighted and fused using a weighted fusion algorithm of multi-dimensional Gaussian components to obtain the global network service characteristics. The network load model is constructed based on the scale of network nodes, the global network activity obtained by averaging the activity levels of all nodes, and the global network business characteristics.

[0008] In one embodiment, a multi-dimensional Gaussian random process with the same dimension is used to model the service characteristics of each node in the network, including: Considering that each node in the network performs different tasks, we use The Gaussian random process models the business characteristics of each node in the network, which can be expressed as ,in, Reason tThe set of all sample functions of the business characteristics of a single node at a certain moment, , For the i A sample function, d is the number of sample functions, that is, the number of dimensions of the Gaussian random process is consistent with the number of sample functions, For the cycle; It is used to describe the random process in which the length of the business generated by the node changes over time. The normal probability density function is expressed as: ; in, for The expected value of Represents any sample function exist Random variables at time; for The expected value of Represents any sample function In another Random variables at time; is a random variable The variance of is a random variable The variance of represents the expected calculation; is the normalized covariance matrix The determinant of ; Indicates the node generated Business length, Indicates that the task is mapped to d Business characteristic random variables of dimensions; Indicates any d a moment; and Both represent dimensions; Further considering that the tasks of each node in the network are different and the business types are relatively fixed, the business characteristics of each node are set to be wide-sense stationary within a limited period of time, and the business characteristics of each node in the network are modeled as The business characteristics of a stationary Gaussian process satisfy the following constraints: ; ; in, for Expected value; is the expected value constant, which is used to indicate that the expected value of a stationary Gaussian process does not change with time; Represents the autocorrelation function at time and The value at for and The expected value of the product of is the time difference, that is ; is the autocorrelation function, which depends only on the time difference .

[0009] In one embodiment, the service characteristics of each node after modeling are based on the service length as a random vector at any time. Dimensional Gaussian distribution, nodes The business length of each dimension is described as dimensional random vector , the probability density function of its business characteristics is expressed as: ; in, for t Time Node The probability density function of the business characteristics; Representation Node The mean vector of different types of business lengths, Representation Node No. i The mean of the business length, , superscript T represents transpose; is the covariance matrix, which is used to represent the node The correlation between the mean vectors of different types of business lengths, Represents the covariance matrix The determinant of ; Through each node After using the Gaussian distribution to describe the service characteristics of different nodes, the probability density function set of the service characteristics of each node in the network is expressed as ;in, is the network node size, and .

[0010] In one of the embodiments, a weighted fusion algorithm of multi-dimensional Gaussian components is used to perform weighted fusion on the service characteristics of all nodes in the network to obtain global network service characteristics; use The weighted fusion algorithm of Gaussian components considers that the nodes in the network meet the independent and identically distributed conditions. The goal is to fuse the business characteristics of all nodes in the network to obtain a Global network service characteristics of Gaussian distribution; Assume Node The Gaussian component of the business characteristics is ,node The activity level is , the activity level indicates the node The probability of generating business in a unit time; Perform normalization to obtain the weight of each node’s business characteristics , expressed as: ; in, satisfy and ; Based on weight Perform weighted fusion on the service features of all nodes in the network to obtain the global network service features ,and conform to Gaussian distribution, that is ; Among them, the mean vector of global network service characteristics With the covariance matrix Respectively expressed as: ; ; in, ,in, express The i means, and ; Superscript T Indicates transpose.

[0011] In one embodiment, a network load model is constructed based on the scale of network nodes, the global network activity obtained by averaging the activity of all nodes, and the global network service characteristics, including: According to the scale of network nodes , the global network activity level is obtained by taking the average activity level of all nodes And the mean vector of global network service characteristics Modeling is performed to construct a network load model, where: t The expected network load at a given moment is expressed as: ; in, is the time-varying network load, for Expected value of global network activity , represents the average probability of all nodes in the network generating business per unit time; For Node activity level; For Node No. i The mean length of each business.

[0012] In one embodiment, the observation plane observes and counts the services of each node, and obtains the service characteristics of the current reconstruction period of each node through maximum likelihood estimation, including: The observation plane maps tasks and services with the task plane to observe and count the nodes in the network in the current observation period. Number of data generated within Observation dataset with data length ; Among them, the reconstruction cycle Include time slots, observation period Include time slots, and satisfy Right now ; Calculate the nodes based on the statistical results of the number of times data is generated n The activity level is , and the data length observation data set Use it as an observation sample to estimate and predict business characteristics; use dimensional stationary Gaussian random process for nodes n The business characteristics of the node are modeled and the n Business characteristics The dimensional Gaussian distribution is reduced to a one-dimensional Gaussian distribution description, and then the node is obtained by maximum likelihood estimation based on the prior information that the data frame length variable at the sampling time obeys the Gaussian distribution. n Mean of business characteristics and standard deviation Specifically, let , for the dataset , its likelihood function is the product of the probability density functions of all observations, expressed as: ; in, Represents the observation period Neidi i The length of the data generated, and ; Taking the logarithm of the likelihood function, we get the log-likelihood function , and About and Find the partial derivative and set it equal to zero, and the solution is and Respectively expressed as: ; ; Based on the above solution, we get and , determine the node n The business characteristics in the current reconstruction cycle are subject to ;in, , is the network node size.

[0013] In one embodiment, the observation plane obtains the service features of each node in the next reconstruction period through Kalman filtering prediction, and then performs feature fusion on the service features of all single-hop reachable nodes obtained through broadcast interaction to obtain the network features of the next reconstruction period and pass them to the knowledge plane, including: Set Nodes within the reconstruction cycle The real business characteristics of , through observation and maximum likelihood estimation, the business characteristics are obtained ;in, For Node In the The true mean of the business characteristics of the reconstruction period, and set The change of follows a random walk process; For Node In the The observed values ​​of the business characteristics of the reconstruction period represent the average data length and set The true mean Noisy measurement of For Node In the The variance of the service characteristics of the reconstruction period reflects the degree of discreteness of the data frame; The goal of Kalman filtering is to and and its corresponding variance and , prediction node In the The business characteristics of the reconstruction cycle are subject to , and then The prediction results of the reconstruction cycle are weightedly integrated with the service characteristics of all single-hop reachable nodes obtained by broadcast interaction to obtain the Finally, the network node scale, the global network activity obtained by averaging the activity of all nodes in the network, and the global network service characteristics are combined into the first The network features of the reconstruction cycle are transferred to the knowledge plane; Among them, the Kalman filter prediction process is not performed in the first two reconstruction cycles when the network is initially established, the new nodes added to the network at the end of the reconstruction cycle do not participate in the algorithm process of the observation surface, and the observation surface only performs the process of observation, maximum likelihood estimation and feature interactive fusion in the first two reconstruction cycles.

[0014] In one embodiment, the Kalman filter estimates the state through two steps of prediction and update. First, the prediction step obtains the The prior estimates of the mean and variance of the business characteristics of the reconstruction period are expressed as follows: ; ; in, For the A priori estimate of the mean of the business characteristics of the reconstruction period, For the The posterior estimate of the mean of the business characteristics of the reconstruction period; For the The variance of the business characteristics of the reconstruction cycle A priori estimate of For the A posterior estimate of the variance of the business characteristics of the reconstruction period; is the process noise variance; Then, the updated Kalman gain is obtained based on the update step of the posterior estimate , the expression is: ; According to the Kalman gain Respectively The posterior estimate of the mean of the service characteristics of the reconstruction period and the posterior estimate of the variance Update, expressed as: ; ; So that in the next iteration, the node In the The predicted distribution of business characteristics of a reconstruction cycle is , and Respectively represent nodes In the The predicted values ​​of the mean and variance of the business characteristics of the reconstruction period; among them, the node In the A priori estimate of the mean of the service characteristics of the reconstruction period Prediction , The forecast is obtained by exponential smoothing method and is expressed as ;in, is the preset smoothing index, Indicates The predicted value of the variance of the business characteristics of the reconstruction period.

[0015] In one embodiment, the knowledge plane carries a pre-trained classification model, and outputs the best MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane according to the network characteristics input by the observation plane and the task requirements input by the task plane, including: First, the knowledge aspect is to model the task requirements based on the performance parameters of the MAX protocol in three dimensions: throughput, packet loss rate, and average network delay. This includes using effectiveness requirements, reliability requirements, and time sensitivity requirements to quantify the importance of task requirements, and define effectiveness requirement parameters. , reliability requirement parameters and time-sensitive demand parameters As throughput , Packet loss rate and average network delay The performance parameters of the three dimensions are normalized so that , , , and satisfies ; Then, the normalized performance parameters and task requirement parameters are weighted and summed as the score evaluation index of different MAC protocols. It is expressed as: ; Among them, the subscript h Indicates the number of alternative MAC protocols, the superscript T represents transpose; Finally, the knowledge aspect is based on the pre-trained XGBoost classification model, with network features including network node scale, global network activity, and the mean and variance of global network business features as classification features, and validity requirement parameters , reliability requirement parameters and time-sensitive demand parameters The task requirements are used as evaluation features to form data set features to classify candidate MAC protocols, and the candidate MAC protocol with the highest score is selected as the best MAC protocol that takes into account the network characteristics and task requirements of the next reconstruction cycle, expressed as: .

[0016] A task-driven MAC protocol reconstruction device, the device comprising: The load modeling module is used to consider the multi-domain distributed wireless self-organizing network scenario. The network load model is constructed through multi-dimensional feature partitioning and extraction of network load, Gaussian random process modeling and feature fusion. The protocol reconstruction module is used to build a software-defined distributed subnet node system architecture based on the network load model. The architecture consists of a task plane, a data plane, and a joint control plane. Among them, the task plane is used to cache the communication services generated during the execution of node tasks and cluster tasks in the form of data frames to the data plane to form a queue. The data plane queue contains the service data frames generated by the node tasks, and also contains the control data frames generated by the feature interaction during the MAC protocol reconstruction process and the channel reservation and preemption behavior in the MAC protocol process; The joint control plane includes the observation plane, the knowledge plane and the control plane. The observation plane is used to observe and count the services of each node, obtain the service characteristics of each node in the current reconstruction cycle through maximum likelihood estimation, and obtain the service characteristics of each node in the next reconstruction cycle through Kalman filtering prediction. Then, the service characteristics of all single-hop reachable nodes obtained by broadcast interaction are integrated to obtain the network characteristics of the next reconstruction cycle and pass them to the knowledge plane. The knowledge plane is used to carry the pre-trained classification model and output the optimal MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane based on the network features input by the observation plane and the task requirements input by the task plane. The control plane is used to carry the MAC protocol component library and determine whether the reconstruction strategy output by the knowledge plane is the same as the MAC protocol of the current reconstruction cycle. If they are the same, the reconstruction process is not performed; if they are not the same, the components are searched and the protocol parameters are adjusted to complete the MAC protocol component-level reconstruction and enter the next reconstruction cycle.

[0017] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented: Considering the multi-domain distributed wireless self-organizing network scenario, a network load model is constructed through multi-dimensional feature extraction, Gaussian random process modeling and feature fusion of network load. Based on the network load model, a software-defined distributed subnet node system architecture is constructed, which consists of the task plane, data plane and joint control plane. Among them, the task plane is used to cache the communication services generated during the execution of node tasks and cluster tasks in the form of data frames to the data plane to form a queue. The data plane queue contains the service data frames generated by the node tasks, and also contains the control data frames generated by the feature interaction during the MAC protocol reconstruction process and the channel reservation and preemption behavior in the MAC protocol process; The joint control plane includes the observation plane, the knowledge plane and the control plane. The observation plane is used to observe and count the services of each node, obtain the service characteristics of each node in the current reconstruction cycle through maximum likelihood estimation, and obtain the service characteristics of each node in the next reconstruction cycle through Kalman filtering prediction. Then, the service characteristics of all single-hop reachable nodes obtained by broadcast interaction are integrated to obtain the network characteristics of the next reconstruction cycle and pass them to the knowledge plane. The knowledge plane is used to carry the pre-trained classification model and output the optimal MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane based on the network features input by the observation plane and the task requirements input by the task plane. The control plane is used to carry the MAC protocol component library and determine whether the reconstruction strategy output by the knowledge plane is the same as the MAC protocol of the current reconstruction cycle. If they are the same, the reconstruction process is not performed; if they are not the same, the components are searched and the protocol parameters are adjusted to complete the MAC protocol component-level reconstruction and enter the next reconstruction cycle.

[0018] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps: Considering the multi-domain distributed wireless self-organizing network scenario, a network load model is constructed through multi-dimensional feature extraction, Gaussian random process modeling and feature fusion of network load. Based on the network load model, a software-defined distributed subnet node system architecture is constructed, which consists of the task plane, data plane and joint control plane. Among them, the task plane is used to cache the communication services generated during the execution of node tasks and cluster tasks in the form of data frames to the data plane to form a queue. The data plane queue contains the service data frames generated by the node tasks, and also contains the control data frames generated by the feature interaction during the MAC protocol reconstruction process and the channel reservation and preemption behavior in the MAC protocol process; The joint control plane includes the observation plane, the knowledge plane and the control plane. The observation plane is used to observe and count the services of each node, obtain the service characteristics of each node in the current reconstruction cycle through maximum likelihood estimation, and obtain the service characteristics of each node in the next reconstruction cycle through Kalman filtering prediction. Then, the service characteristics of all single-hop reachable nodes obtained by broadcast interaction are integrated to obtain the network characteristics of the next reconstruction cycle and pass them to the knowledge plane. The knowledge plane is used to carry the pre-trained classification model and output the optimal MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane based on the network features input by the observation plane and the task requirements input by the task plane. The control plane is used to carry the MAC protocol component library and determine whether the reconstruction strategy output by the knowledge plane is the same as the MAC protocol of the current reconstruction cycle. If they are the same, the reconstruction process is not performed; if they are not the same, the components are searched and the protocol parameters are adjusted to complete the MAC protocol component-level reconstruction and enter the next reconstruction cycle.

[0019] The above-mentioned task-driven MAC protocol reconstruction method, device, equipment and medium have the following beneficial effects: 1. Taking into account the differences in business characteristics brought about by the heterogeneity of node tasks, the network load is finely divided into three dimensions of characteristics. Compared with the traditional network load modeling method, the network load model constructed in this application can fully reflect the performance of the MAC protocol in different network environments.

[0020] 2. Build a closed-loop software-defined distributed subnet node system architecture. This architecture provides a complete system design and MAC protocol reconstruction algorithm flow, and provides a solution for distributed nodes to collaborate to complete the reconstruction process. The reconstruction process takes into account different task requirements to solve the problem of balancing protocol performance. It constructs a data set and uses the XGBoost classification model to implement reconstruction decisions for different protocols. Ultimately, it can reconstruct the optimal MAC protocol based on task requirements and taking into account different performance dimensions, thereby improving the adaptability and reliability of the MAC protocol to dynamically changing network environments and task requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram of a multi-domain distributed wireless self-organizing network scenario in an embodiment; Figure 2 A schematic diagram of a software-defined distributed subnet node system architecture in one embodiment; Figure 3 A schematic diagram of the flow of each plane in a software-defined distributed subnet node system architecture in one embodiment; Figure 4 Schematic diagram of performance comparison of 0.6-CSMA protocol under different network characteristic parameters in one embodiment; wherein, Figure 4 (a) is the throughput performance comparison curve of the 0.6-CSMA protocol under different network characteristic parameters. Figure 4 (b) is the packet loss rate performance comparison curve of the 0.6-CSMA protocol under different network characteristic parameters. Figure 4 (c) is the comparison curve of the network average delay performance of the 0.6-CSMA protocol under different network characteristic parameters; Figure 5 Schematic diagram of SOTDMA protocol performance comparison under different network characteristic parameters in one embodiment; wherein: Figure 5 (a) is the throughput performance comparison curve of the SOTDMA protocol under different network characteristic parameters. Figure 5 (b) is the packet loss rate performance comparison curve of the SOTDMA protocol under different network characteristic parameters. Figure 5 (c) is the comparison curve of the network average delay performance of the SOTDMA protocol under different network characteristic parameters; Figure 6 A schematic diagram of a throughput performance curve of a reconstruction protocol under network scale variables in an embodiment; Figure 7 It is a schematic diagram of a packet loss rate performance curve of a reconstruction protocol under network scale variables in one embodiment; Figure 8 A schematic diagram of an average delay performance curve of a reconstruction protocol network under network scale variables in an embodiment; Fig. 9 A schematic diagram of a throughput performance curve of a reconstruction protocol under activity level variables in one embodiment; Fig.10 A schematic diagram of a packet loss rate performance curve of a reconstruction protocol under activity level variables in one embodiment; Fig.11 A schematic diagram of a delay performance curve of a reconstruction protocol under activity level variables in an embodiment; Fig.12 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0023] Multi-domain distributed wireless self-organizing network scenarios such as Figure 1 As shown in the figure, the network adopts a distributed structure, which means that there is no central node for centralized control. The nodes in the network have different functions, and the nodes are interconnected and work together to perform cluster tasks. The mapping of nodes with different functions and applications to communication services is also different. The upper-layer communication services cover heterogeneous modal data such as instructions, text, audio, image and video information, which are reflected in the MAC layer with different data lengths and priorities. Different services also have different requirements for throughput, effectiveness and time sensitivity, which poses a challenge to the existing single fixed MAC protocol.

[0024] Specifically, in order to highlight the performance analysis of the MAC protocol and make the analysis general, the following settings are made for the network system characteristics: (1) All nodes in the network are reachable within one hop and the wireless channel is ideal; (2) All nodes in the network operate in the same frequency band, with a fixed transmission rate and considered as a time slot channel; (3) Data frame collision is the only cause of reception errors; (4) If a data frame collision occurs, all data frames will be destroyed.

[0025] According to the above settings, the transmission rate of the slot channel node is fixed, which means that the MAC protocol transmits data in fixed-length data frames. Then the business data may be split into multiple data frames, which means the occupation of multiple channel time slots. Even if the length is less than one data frame, one time slot is occupied. In order to improve the performance of the competitive MAC protocol, the TXOP (Transmission Opportunity) transmission mechanism is introduced. After the node obtains the transmission opportunity, it is allowed to transmit multiple frames of data, that is, a single channel competition realizes continuous transmission of multiple frames, thereby reducing the number of competitions and improving channel utilization. In this scenario, it is considered that after the node obtains the transmission opportunity through competition, it will complete the transmission of at least one complete data (multiple data frames) to ensure the integrity of the upper-layer business data.

[0026] The heterogeneity of network characteristics and node communication services is reflected in three aspects: network node scale, node activity and service data length. The number of nodes in the network reflects the size of the network node scale, the node activity is reflected in the probability of node communication services, and the service data length is reflected in the occupancy time of the channel time slot.

[0027] Classic models in queuing theory, such as the M / M / 1 (single server queue) model, usually assume that the arrival process is a Poisson process, the service time is exponentially distributed, and the system is in a steady state. However, in actual communication networks, the arrival process may not be Poisson distributed. Especially when the load is high, the arrival of data often shows burstiness and correlation, which causes the model to underestimate queue accumulation and delay, resulting in deviations in model predictions. In a single channel, large file transfers may require longer service times, thereby blocking short data packets of other users and causing greater delay fluctuations. Simple models in queuing theory may not be able to handle this heterogeneous service time distribution, especially when the load intensity is high. This impact will be more significant.

[0028] Based on the above analysis, the three dimensional parameters of network heterogeneity are introduced, which can describe the nature of network load changes in a refined manner and also realize mathematical modeling of the burstiness and correlation of data from different nodes. Let's qualitatively analyze these three dimensions. For example, considering the high-load network situation, the traditional analysis based on queuing theory modeling believes that the reservation-based or polling-based MAC protocol should be used under high-load conditions. In fact, any of the above dimensional parameters will increase the network load. When the network load increases only due to the increase in the number of network nodes, and the node activity and data length remain at a low level, the use of the reservation-based MAC protocol is not the best choice.

[0029] It is worth noting that different MAC protocols have their own unique performance advantages, and there is no MAC protocol that performs best in multiple dimensions. Throughput (effectiveness), packet loss rate (reliability) and latency (fairness, time sensitivity) are the basic indicators for measuring protocol performance. Different application scenarios and tasks have different performance requirements for self-organizing network clusters. It is necessary to change the performance requirements according to different task requirements and achieve a balance between task adaptation and performance through protocol reconstruction.

[0030] In a distributed network scenario, each node makes independent decisions to obtain a reconstructed strategy. Therefore, it is necessary to design a node system architecture that integrates observation, decision-making and control loops. Based on the time-varying network characteristics and task requirements, global information is obtained through observation, distributed fusion and data prediction, the most appropriate MAC protocol is adapted and the protocol parameters are optimized, and the three performance indicators of throughput, packet loss rate and latency in the entire self-organizing network are comprehensively optimized.

[0031] Based on the above analysis, in one embodiment, the present application provides a task-driven MAC protocol reconstruction method, comprising the following steps: Step 1: Considering the multi-domain distributed wireless self-organizing network scenario, a network load model is constructed through multi-dimensional feature partitioning and extraction of network load, Gaussian random process modeling and feature fusion.

[0032] Among them, considering that in complex multi-domain distributed wireless self-organizing network scenarios, different nodes in the network have different functions and corresponding communication service characteristics are also very different. If the service characteristics are simply described by equal-length data packets, it is only applicable to nodes of the same service type. For scenarios with different service characteristics caused by different node functions in heterogeneous networks, the load cannot fully represent the network strength characteristics. Therefore, this application designs a new way to build a network load model, including: Step 1.1: Consider the multi-domain distributed wireless self-organizing network scenario and divide and extract the multi-dimensional characteristics of the network load, including network node scale, node activity level, and node service characteristics.

[0033] Step 1.2: Use a multi-dimensional Gaussian random process with the same dimension to model the business characteristics of each node in the network, and use the weighted fusion algorithm of the multi-dimensional Gaussian component to weightedly fuse the business characteristics of all nodes in the network to obtain the global network business characteristics.

[0034] Specifically, we first consider that the tasks performed by each node in the network are different, and use the same dimension The Gaussian random process models the business characteristics of each node in the network, which can be expressed as ,in, Reason t The set of all sample functions of the business characteristics of a single node at a certain moment, , For the i A sample function, d is the number of sample functions, that is, the number of dimensions of the Gaussian random process is consistent with the number of sample functions, For cycle.

[0035] It is used to describe the random process of the service length (number of data frames) generated by the node changing with time. The normal probability density function is expressed as: ; in, for The expected value of Represents any sample function exist Random variables at time; for The expected value of Represents any sample function In another Random variables at time; is a random variable The variance of is a random variable The variance of represents the expected calculation; is the normalized covariance matrix The determinant of ; Indicates the node generated Business length, Indicates that the task is mapped to d Business characteristic random variables of dimensions; Indicates any d a moment; and Both represent dimensions.

[0036] Further considering that the tasks of each node in the network are different and the service types are relatively fixed, the service characteristics of each node are set to be broadly stable within a limited period of time. At the same time, the service characteristics of the data transmitted during the communication process reflected by different tasks should obey Dimensional Gaussian distribution, so the business characteristics of each node in the network are modeled as The business characteristics of a stationary Gaussian process satisfy the following constraints: ; ; in, for Expected value; is the expected value constant, which is used to indicate that the expected value of a stationary Gaussian process does not change with time; Represents the autocorrelation function at time and The value at for and The expected value of the product of is the time difference, that is ; is the autocorrelation function, which depends only on the time difference .

[0037] Without loss of generality, the business characteristics of each node after modeling are random vectors with business length at any time. Dimensional Gaussian distribution, nodes The business length of each dimension is described as dimensional random vector , the probability density function of its business characteristics is expressed as: ; in, for t Time Node The probability density function of the business characteristics; Representation Node The mean vector of different types of business lengths, Representation Node No. i The mean of the business length, , superscript T represents transpose; is the covariance matrix, which is used to represent the node The correlation between the mean vectors of different types of business lengths, Represents the covariance matrix The determinant of .

[0038] Through each node After using the Gaussian distribution to describe the service characteristics of different nodes, the probability density function set of the service characteristics of each node in the network is expressed as ;in, is the network node size, and .

[0039] In order to obtain the global network service characteristics at a certain moment, it is necessary to integrate the service characteristics of all nodes in the network. The weighted fusion algorithm of Gaussian components considers that the nodes in the network meet the independent and identically distributed conditions. The goal is to fuse the business characteristics of all nodes in the network to obtain a Global network traffic characteristics of Gaussian distribution.

[0040] Assume Node The Gaussian component of the business characteristics is ,node The activity level is , the activity level indicates the node The probability of generating business in a unit time; Perform normalization to obtain the weight of each node’s business characteristics , expressed as: ; in, satisfy and .

[0041] Based on weight Perform weighted fusion on the service features of all nodes in the network to obtain the global network service features ,and conform to Gaussian distribution, that is ; Among them, the mean vector of global network service characteristics With the covariance matrix Respectively expressed as: ; ; in, ,in, express The i means, and ; Superscript T Indicates transpose.

[0042] Step 1.3: Model the network load model based on the scale of network nodes, the global network activity obtained by averaging the activity levels of all nodes, and the global network service characteristics.

[0043] Specifically, according to the scale of network nodes , the global network activity level is obtained by taking the average activity level of all nodes And the mean vector of global network service characteristics Modeling is performed to construct a network load model, where: t The expected network load at a given moment is expressed as: ; in, is the time-varying network load, for Expected value of global network activity , represents the average probability of all nodes in the network generating business per unit time; For Node activity level; For Node No. i The mean length of each business.

[0044] Through the above modeling method, the network load is integrated with the business characteristics of each node, making the way of describing the network load more detailed and able to obtain more accurate network characteristics. Compared with the traditional load-based MAC protocol switching algorithm, this task-feature-based network load model provides more favorable support for the task-driven MAC protocol reconstruction algorithm.

[0045] Step 2: Based on the network load model, build a software-defined distributed subnet node system architecture. Figure 2 As shown in the figure, it consists of a task plane, a data plane, and a joint control plane. Each different functional plane is separated and equipped with different algorithms and protocols, and is implemented through a full-stack software definition. Each node in the network is developed based on this architecture. Each node exists independently and performs different tasks in the network to form a distributed heterogeneous network.

[0046] Furthermore, since all nodes in the network are reachable by a single hop, the feature results of the observation plane fusion are the same, and the knowledge plane is equipped with the same set of pre-trained intelligent algorithm classification models. The reconstruction strategy obtained by inputting into the knowledge plane is also the same, which can ensure the consistency of the reconstruction strategy within the subnet. Through the guidance of the software-defined distributed subnet node system architecture, the closed-loop control problem of node observation, decision-making, and action in a pure distributed network can be solved.

[0047] Among them, the task plane is used to cache the communication services generated during the execution of node tasks and cluster tasks in the form of data frames to the data plane to form a queue. Cluster tasks are upper-layer tasks completed by all nodes in the network, such as silent approach, random reorganization, joint action, etc. Node tasks are specific actions of heterogeneous nodes, such as high-altitude reconnaissance, continuous monitoring, environmental perception, cargo transportation, user access, etc.

[0048] The data plane queue contains the business data frames generated by the node tasks, as well as the control data frames generated by the feature interaction during the MAC protocol reconstruction process and the channel reservation and preemption behaviors in the MAC protocol process.

[0049] The joint control plane includes the observation plane, the knowledge plane and the control plane; the observation plane is used to observe and count the services of each node, obtain the service characteristics of each node in the current reconstruction cycle through maximum likelihood estimation, and obtain the service characteristics of each node in the next reconstruction cycle through Kalman filtering prediction, and then fuse the service characteristics of all single-hop reachable nodes obtained by broadcast interaction to obtain the network characteristics of the next reconstruction cycle and pass them to the knowledge plane. The knowledge plane is used to carry the pre-trained classification model, and output the best MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane based on the network characteristics input by the observation plane and the task requirements input by the task plane. The control plane is used to carry the MAC protocol component library to determine whether the reconstruction strategy output by the knowledge plane is the same as the MAC protocol of the current reconstruction cycle. If they are the same, the reconstruction process will not be executed; if they are not the same, the components are searched and the protocol parameters are adjusted to complete the MAC protocol component-level reconstruction, and the cycle of the next reconstruction cycle is entered.

[0050] Based on the above network system characteristics, consider the time slot channel and assume a reconstruction cycle Include time slots, reconstruction cycle Should be less than the task execution cycle All nodes in the network complete clock synchronization, statistics of network characteristics, observation and estimation of business characteristics within a reconstruction cycle, complete feature interaction and fusion, and decide on the best MAC protocol based on task requirements and network characteristics. The reconstruction process of the MAC protocol is completed within the first few time slots of the next reconstruction cycle. Assume an observation cycle Include time slots, and satisfy Right now The processes of each plane in the software-defined distributed subnet node system architecture are as follows: Figure 3 shown.

[0051] pass Figure 3 As can be seen from the process shown, the node clusters in the network jointly perform a certain type of task. The distributed nodes need to learn the network characteristics of the task cluster in the next reconstruction cycle through three steps: business statistics, observation prediction, and interactive fusion. The obtained network characteristics and current task requirements are passed to the knowledge plane to decide the best MAC protocol for the next reconstruction cycle. In this process, the data plane continues to transmit data according to the MAC protocol of this reconstruction cycle. After obtaining the reconstruction strategy, if it is the same as the protocol of this reconstruction cycle, there is no need to perform the reconstruction process; if it is different, the nodes in the cluster will synchronously execute the protocol reconstruction. After completion, the system clock synchronization is completed through GPS (Global Positioning System) timing, and the next reconstruction cycle begins.

[0052] exist Figure 2In the software-defined distributed subnet node system architecture shown, due to the distributed characteristics of the wireless self-organizing network, it is difficult to centrally obtain the global information of the network through a single node. Therefore, other network characteristics other than the network node scale in the above network load model cannot be directly obtained. In order to obtain accurate business characteristics in the above model, each node needs to perform statistics, fitting and prediction on the parameters of the node through its own observation surface. The algorithm flow executed by the observation surface includes the following steps: Feature estimation part: The observation plane maps tasks and services with the task plane, and observes and counts the nodes in the network in the current observation period. Number of data generated within Observation dataset with data length ; Calculate the node based on the statistical results of the number of times data is generated n The activity level is , and the data length observation data set Use it as an observation sample to estimate and predict business characteristics.

[0053] According to the above network load model construction process, dimensional stationary Gaussian random process for nodes n The business characteristics of the node are modeled, and in order to simplify the calculation complexity of the observation surface, the node n Business characteristics The dimensional Gaussian distribution is reduced to a one-dimensional Gaussian distribution description, and then the node is obtained by maximum likelihood estimation based on the prior information that the data frame length variable at the sampling time obeys the Gaussian distribution. n Mean of business characteristics and standard deviation Specifically, let , for the dataset , its likelihood function is the product of the probability density functions of all observations, expressed as: ; in, Represents the observation period Neidi i The length of the data generated, and .

[0054] Taking the logarithm of the likelihood function, we get the log-likelihood function , expressed as: .

[0055] To find the maximum log-likelihood function and ,right About and Find the partial derivative and set it equal to zero, expressed as: ; ; Solved and Respectively expressed as: ; ; Based on the above solution, we get and , determine the node n The business characteristics in the current reconstruction cycle are subject to ;in, , is the network node scale. However, the result obtained is the service characteristics of this reconstruction cycle, which cannot directly provide guidance for the control behavior of the next reconstruction cycle. Therefore, based on the estimation results of the previous reconstruction cycle and the current reconstruction cycle, the Kalman filter method is used to predict the service characteristics of the next reconstruction cycle.

[0056] Feature prediction part: After estimating the service features of each node in the current reconstruction cycle, in order to ensure the rigor of the reconstruction logic, the service features of the next reconstruction cycle are predicted to guide the protocol reconstruction strategy. Considering the autocorrelation of heterogeneous node communication services, each node first uses the Kalman filter algorithm for prediction, and then interacts and integrates the prediction results with the features of other nodes in the network to obtain the global network features.

[0057] Specifically, set the Nodes within the reconstruction cycle The real business characteristics of , through observation and maximum likelihood estimation, the business characteristics are obtained ;in, For Node In the The true mean of the business characteristics of the reconstruction cycle; For Node In the The observed value of the service characteristics of the reconstruction period represents the average data length; For Node In the The variance of the service characteristics of a reconstruction period reflects the degree of discreteness of the data frame.

[0058] Among them, setting The change of obeys the random walk process, and the state equation is expressed as: ; in, is the process noise variance, estimated by the variance of the mean change of historical data, expressed as , is the number of reconstructions.

[0059] And, set The true mean The observation equation for the noisy measurement is: ; Among them, process noise With measurement noise They are independent of each other and are all zero-mean Gaussian white noise.

[0060] The goal of Kalman filtering is to and and its corresponding variance and , prediction node In the The business characteristics of the reconstruction cycle are subject to , and then The prediction results of the reconstruction cycle are weightedly integrated with the service characteristics of all single-hop reachable nodes obtained by broadcast interaction to obtain the Finally, the network node scale, the global network activity obtained by averaging the activity of all nodes in the network, and the global network service characteristics are combined into the first The network features of the reconstruction cycle are transferred to the knowledge surface.

[0061] Specifically, the Kalman filter estimates the state through two steps of prediction and update. First, the prediction step obtains the first The prior estimates of the mean and variance of the business characteristics of the reconstruction period are expressed as follows: ; ; in, For the A priori estimate of the mean of the business characteristics of the reconstruction period, For the The posterior estimate of the mean of the business characteristics of the reconstruction period; For the The variance of the business characteristics of the reconstruction cycle A priori estimate of For the A posterior estimate of the variance of the business characteristics of the reconstruction period; is the process noise variance.

[0062] Then, the updated Kalman gain is obtained based on the update step of the posterior estimate , the expression is: ; According to the Kalman gain Respectively The posterior estimate of the mean of the service characteristics of the reconstruction period and the posterior estimate of the variance Update, expressed as: ; ; So that in the next iteration, the node In the The predicted distribution of business characteristics of a reconstruction cycle is , and Respectively represent nodes In the The predicted values ​​of the mean and variance of the business characteristics of the reconstruction period; among them, the node In the A priori estimate of the mean of the service characteristics of the reconstruction period Prediction , The forecast is obtained by exponential smoothing method and is expressed as ;in, is the preset smoothing index, Indicates The predicted value of the variance of the business characteristics of the reconstruction period.

[0063] It is worth noting that in the first two reconstruction cycles when the network is initially established, the Kalman filter prediction process is not performed, and the new nodes added to the network at the end of the reconstruction cycle do not participate in the algorithm flow of the observation surface, and the observation surface only performs observation, maximum likelihood estimation and feature interactive fusion processes in the first two reconstruction cycles. By introducing the Kalman filter algorithm, this application can not only predict the business characteristics of each node in the next reconstruction cycle, but also eliminate the observation and estimation errors caused by the randomness of the node-generated data to a certain extent, making the node task mapping to the characteristics of the communication service more accurate.

[0064] Figure 2 The algorithm flow executed in the knowledge aspect includes: various ad hoc MAC protocols have different performance under different network characteristics. The specific performance evaluation indicators can be divided into throughput, packet loss rate and average network delay, reflecting the advantages and disadvantages of a certain MAC protocol in three different dimensions. It is defined as the number of data frames successfully transmitted per unit time, which can be expressed as: ; in, Represents the observation period The number of data frames successfully transmitted and received within Represents the observation period Total number of time slots.

[0065] Packet loss rate It is defined as the ratio of the sum of data frames damaged during transmission and discarded due to failure to transmit in time to all data frames generated per unit time, which can be expressed as: ; in, Represents the observation period The number of data frames damaged during transmission at the internal node, Represents the observation period The number of data frames that are discarded due to failure to transmit in time by the internal node or accumulated in the queue at the end of the cycle. Represents the observation period The number of data frames generated by all nodes in the network.

[0066] Average network delay It is defined as the average time that all data frames generated in the network wait in the queue. Due to the TXOP mechanism, the delay statistics method can be simplified. It is only necessary to count the waiting time of each node in the send suspension state and the number of times the send process is activated, and the average delay of all nodes in the network is averaged to obtain the average network delay, which can be expressed as: ; in, The observation period Internal Node Send the waiting time of the pending state. is the number of activation sending processes. h There are alternative MAC protocols. The performance of different protocols under different network characteristics constitutes a vector, which can be expressed as , , .

[0067] In order to achieve unbiased evaluation of the three performance indicators, they are normalized. The definitions of throughput and packet loss rate have been normalized, so only the average network delay needs to be calculated. Normalize. Using the local information normalization method, the average delay performance of the candidate MAC protocol network obtained under different network characteristics is normalized to obtain: ; in, The normalized average network delay can be obtained as .

[0068] First, the knowledge aspect is to model the task requirements based on the performance parameters of the MAX protocol in three dimensions: throughput, packet loss rate, and average network delay. This includes using effectiveness requirements, reliability requirements, and time sensitivity requirements to quantify the importance of task requirements, and define effectiveness requirement parameters. , reliability requirement parameters and time-sensitive demand parameters As throughput , Packet loss rate and average network delay The weight of .

[0069] Among them, the performance parameters of the three dimensions are normalized so that , , , and satisfies .

[0070] Then, the normalized performance parameters and task requirement parameters are weighted and summed as the score evaluation index of different MAC protocols. It is expressed as: ; Among them, the subscript h Indicates the number of alternative MAC protocols, the superscript T Indicates transpose.

[0071] Finally, the knowledge aspect is based on the pre-trained XGBoost classification model, with network features including network node scale, global network activity, and the mean and variance of global network business features as classification features, and validity requirement parameters , reliability requirement parameters and time-sensitive demand parameters The task requirements are used as evaluation features to form data set features to classify candidate MAC protocols, and the candidate MAC protocol with the highest score is selected as the best MAC protocol that takes into account the network characteristics and task requirements of the next reconstruction cycle, expressed as: .

[0072] Among them, the dataset feature settings of the XGBoost classification model are shown in Table 1.

[0073] Table 1 Dataset feature settings:

[0074] The scale of network nodes and task requirements are randomly generated within the value range and constraints. At the same time, in order to fully eliminate the probability error of data generated within the cycle and the randomness of node transmission conflicts, each different network feature is reconstructed with a step length of 200 times, and the average of throughput, packet loss rate and average network delay is obtained as a complete data.

[0075] XGBoost (eXtreme Gradient Boosting, XGB for short) is an efficient implementation of the GBDT (gradient boosting decision tree) algorithm, which optimizes algorithm efficiency, model performance and scalability. XGBoost introduces regularization terms to prevent overfitting, and uses efficient tree splitting algorithms and column sampling techniques to perform well on large-scale data sets. The idea of ​​the algorithm is to continuously add trees and continuously perform feature splitting to grow a tree. Each time a tree is added, a new function is learned to fit the residual of the last prediction.

[0076] The objective function of XGBoost consists of two parts: training loss and regularization term. The objective function is defined as follows: ; in, is the training loss term, represents the loss function; is the regularization term, representing the complexity of the tree.

[0077] Since the application scenario is to deal with multi-classification problems and hope to obtain the probability estimate of each category, it is more appropriate to use the logistic regression loss function as the loss function. If there are more noise or outliers in the data set, the logistic regression loss function can also provide better robustness. The expression of the logistic loss function is: ; in It is i Samples Since XGBoost is an additive model, the prediction score is the cumulative sum of the scores of each tree, expressed as: ; in represents the sum of the trees, For the The prediction function of the tree.

[0078] Assume The trained tree model after iterations is ,sample The prediction result can be expressed as: ; in For the front The prediction result of the tree. Substituting into the objective function, we get: .

[0079] The XGBoost algorithm is used to map the task requirements and protocol performance, and the XGB classification model is obtained through simulation and construction of data sets. As a knowledge-based algorithm model, it is installed on each distributed node, providing strategic support for the intelligent reconstruction of distributed nodes in scenarios with different network characteristics and task requirements.

[0080] Furthermore, the method proposed in this article optimizes the performance in three dimensions: throughput, packet loss rate, and average network delay. It only focuses on the channel access mechanism of different MAC protocols and ignores the consideration of existing MAC protocols for performance dimensions such as priority, security, and low energy consumption. The functions of the protocol optimized for other dimensions can be added on this basis.

[0081] Figure 2 The control plane shown is aimed at the application scenario of distributed wireless ad hoc network. Based on the existing MAC protocol, different access transmission mechanisms are summarized, and two types of transmission mechanisms are obtained, namely contention type and scheduling type MAC protocols, as well as a MAC protocol component library composed of six different types of MAC protocols, as shown in Table 2.

[0082] Table 2 MAC protocol component library:

[0083] Based on the setting of the time slot channel above, and the data frame length is equal, the protocol will compete and occupy the channel only at the beginning of the time slot, and will occupy a complete time slot when the data volume is less than the frame length. At the same time, following the TXOP transmission mechanism, when the competitive protocol and the polling protocol obtain the transmission opportunity, they will continue to occupy the channel until the entire data is sent, ensuring the integrity of the data.

[0084] As the simplest MAC protocol, the S-ALOHA protocol directly broadcasts as long as communication services are generated. It is applicable when there are few nodes, low activity, and short service average. However, since the protocol mechanism lacks the ability to monitor the channel occupancy status, any time slot may be occupied by data frames from other nodes, causing conflicts and resulting in the destruction of data frames. This is particularly evident in the transmission of long data, and its reliability and throughput are difficult to guarantee.

[0085] The CSMA (Carrier Sense Multiple Access) protocol adds the function of carrier sense on the basis of the S-ALOHA protocol. After the service is generated, it will first join the queue. Only after the channel is sensed to be idle will the data of this node be sent. The CSMA protocol provides a guarantee for the reliability of data transmission, but it is still possible that when the channel is idle, multiple nodes occupy the same time slot at the same time, causing conflicts. Therefore, this transmission mechanism can only guarantee a certain performance under the condition of low network activity.

[0086] P -Adhere to the CSMA protocol as a variant, after finding the channel idle, the probability Send data to This mechanism essentially weakens the activity of the network by reducing the probability of competition, and its performance is limitedly improved compared with the 1-persistent CSMA protocol.

[0087] The 802.11 DCF protocol and the fixed backoff window CSMA / CA (Carrier Sense Multiple Access / Collision Avoidance) protocol add the channel reservation function on the basis of CSMA, which can solve the problem of hidden / exposed terminals, but this consideration is not necessary in the scenario where the network is reachable in one hop. The essential difference in its mechanism lies in the non-persistent backoff algorithm. After each conflict, the node will adjust the contention window and reselect the backoff duration. In the 802.11 standard, the DCF (Distributed Coordination Function) protocol considers dynamic window backoff and adopts a binary backoff algorithm. The node will retransmit according to the number of retransmissions. Adjust the contention window size, specifically expressed as: ; in For the The contention window size for the retransmission. is the minimum window value, is the maximum window value; , is the maximum number of retransmissions. After determining the contention window size, the node randomly selects an integer value within this range as the backoff duration. , waiting Similarly, a fixed backoff window means that the range of the backoff duration is fixed. After a conflict occurs, the contention window size is not changed, and a new random integer value is selected as the backoff duration. .

[0088] The polling protocol implements node access scheduling by querying each node in turn or passing tokens. It is commonly used in wired networks and wireless networks with central nodes. In the distributed node scenario, the polling protocol design is slightly different, but the polling transmission mechanism is fully retained. Counting the number of network nodes when reconstructing the protocol , from node 1 to node The token is passed in order, and the channel is occupied in turn for transmission. The TXOP mechanism is also followed during transmission. After the node completes the transmission of complete data, it passes the token to the next node.

[0089] SOTDMA is a MAC protocol designed for dynamic, distributed wireless communication environments. Its transmission mechanism is similar to the traditional TDMA (Time Division Multiple Access) protocol, but it can update the time slot allocation scheme according to the dynamic changes in the topology to achieve efficient use of the channel. In this scenario, the update period of the protocol time slot allocation is set to the reconstruction period to reflect the performance characteristics of the transmission mechanism. The difference between SOTDMA and polling scheduling protocols is that the nodes of TDMA take turns to send only one data frame of data, and the transmission of complete data is achieved through multiple cycles.

[0090] During the specific implementation, in order to verify the guiding significance of the network load model constructed in this application for the reconstruction of the MAC protocol, a simulation experiment was further carried out to verify the network load modeling method based on the Poisson distribution of the queuing theory model, and the performance of the competition protocol and the scheduling protocol were simulated respectively, and the impact of the two models on the performance of the MAC protocol was obtained.

[0091] The competition protocol uses p-CSMA protocol for simulation, and the probability ,The simulation time is set with a step of 5000 time slots per cycle, and the throughput, packet loss rate and average network delay performance of 10 cycles are counted.,Three networks with different characteristics are set, and the network characteristic parameters are shown in Table 3.

[0092] Table 3 Network characteristic parameter settings for competition protocol simulation:

[0093] Among them, the characteristic variance It is randomly generated during the simulation process to reflect the error of model parameter estimation and simulate the heterogeneity of business data generated by network nodes. The variance range of global network business characteristics is set to The network load is calculated by the above parameters: , , then the expectation of the Poisson distribution is 0.8, and the probability density function is: ; The above formula represents that 0.8 data frames arrive at the channel in a unit time (time slot). Although the above three network characteristic parameters are different, they have the same load.

[0094] Figure 4The throughput, packet loss rate and average network delay performance curves of the contention-type 0.6-CSMA protocol in three different network characteristics are given respectively. Although the three networks have the same network load, it can be seen from the simulation results that there are large differences in the performance of the protocol. Under the same activity level, network 1 has a large number of nodes and a short service data length, with a lower delay, but poor throughput and packet loss performance; network 3 has a small number of nodes and a long service data length, with good throughput and packet loss performance, but a longer delay; the performance of network 2 is between the two. The reason is that the increase in the number of nodes leads to an increase in the probability of conflict when accessing the channel, and the competition for time slots leads to the conflict and loss of data frames. The transmission of short frames also increases the number of times network nodes compete for channels, and the conflict situation is therefore aggravated, but it also brings more opportunities for nodes to access the channel, which reduces the average waiting delay, and vice versa.

[0095] The scheduling protocol uses the SOTDMA protocol for simulation. Similarly, three networks with different characteristics are set up, and the fused characteristic parameters are shown in Table 4.

[0096] Table 4 Scheduling protocol simulation network characteristic parameter settings:

[0097] Similarly, the variance range of the global network service characteristics is set to The network load can be approximated by calculating the above parameters. , , then the expectation of the Poisson distribution is 1, and the probability density function , the three networks with different characteristics still have the same load.

[0098] Figure 5The throughput, packet loss rate and average network delay performance curves of the scheduling SOTDMA protocol in three networks with different characteristic parameters are given respectively. The three networks also have the same network load characteristics, and the protocol performance differences in the simulation results are still obvious. Network 1 has many nodes, but the nodes are not active, and the probability of generating business is extremely low. If business is generated, long data will be accumulated; Network 3 has fewer nodes and shorter business frame length, but the network nodes are active, and the probability of generating business per unit time is relatively high; the characteristics of network 2 are between the two. Based on the access mechanism analysis of the scheduling class, the performance of the SOTDMA protocol in network 3 should be better than that of network 1, which is also verified in the simulation results. The throughput and packet loss rate performance of network 2 and network 3 gradually improve with the increase of simulation time, showing a trend of convergence to the performance of network 3. This is because at the beginning of the simulation, the activity level of network 3 is extremely low, resulting in large differences in the business queues of each node. As time goes by, the randomness of generating business is gradually eliminated, and the differences in the business queues of each node tend to be flat. Due to the inherent characteristics of SOTDMA, the delay performance is highly positively correlated with the number of network nodes.

[0099] Through the simulation of contention protocols and scheduling protocols in networks with different characteristics, the performance of different protocols under the same load conditions was obtained.

[0100] from Figure 4 and Figure 5 It can be seen that the same network load of the three networks represents the same load parameters of the Poisson distribution. This application splits the network load into three dimensions of characteristics. The characteristics are different but the network load is the same, which shows significant differences in performance. This means that the traditional load modeling method based on Poisson distribution has certain limitations when analyzing the performance of the MAC protocol, which is reflected in the insufficient characterization of network characteristics and the inability to fully describe the burstiness and autocorrelation of the business. In some studies, the parameter of network load is used as a standard for protocol reconstruction, which is actually not perfect. Different protocols have essential differences in access mechanisms and are highly sensitive to different network characteristics. Therefore, it is necessary to fine-tune the modeling of characteristic parameters that affect protocol performance in different network environments.

[0101] Simulations have proven that the network load model based on task characteristics constructed in this application is effective and sufficient in analyzing MAC protocol performance.

[0102] To simulate and verify the data set construction and XGBoost algorithm training results, the data set parameters and examples are set as shown in Table 5. The three types of indicators in the complete data set, namely network features, task requirements and labels, are retained as training sets to train the XGBoost classification model. Among the 30,000 data generated, 70% of the data set is divided as a training set and the other 30% as a test set, and it is ensured that the test set and the training set have no overlap and the distribution is consistent. After training and grid search method to complete parameter adjustment, the model parameters with the best classification performance are shown in Table 6. The parameters not indicated in the table are the default values.

[0103] Table 5 Dataset parameters and examples:

[0104] Table 6 XGBoost classification model parameter settings:

[0105] After setting the above parameters and completing the training, the accuracy of the classification model is evaluated. The 5-fold cross validation is used to reduce the randomness of the evaluation results, and the classification accuracy results of the training set and the test set are obtained, as shown in Table 7.

[0106] Table 7 Classification accuracy results of XGBoost classification model:

[0107] Based on the trained XGBoost classification model, the performance of the reconstructed protocol under different network characteristics is verified. The network scale, activity level and business feature mean are used as variables to simulate the effect of protocol reconstruction. The simulation parameter settings for protocol reconstruction verification are shown in Table 8, where is the reconstruction step size.

[0108] Table 8 Reconstruction simulation parameter settings:

[0109] Figures 6 to 8 The performance curves of the reconstruction protocol throughput, packet loss rate and average network delay under network scale variables are given respectively. The initial value of the network node scale is set to , the range is The performance curve of the reconstruction protocol is represented by a dotted line, and the solid line represents the theoretical performance of different protocols. Figures 6 to 8 It is not difficult to see that as the number of network nodes continues to increase, it has undergone four reconstruction processes, namely , , and The MAC protocol was reconstructed when the task was completed. The protocol types include p-persistent CSMA, fixed window backoff CSMA / CA and SOTDMA. Each protocol is the best MAC protocol under the set network characteristics and demand parameters. According to the setting of the task demand parameters, the network has a high demand for throughput. Figure 6 From the throughput performance curve shown, we can see that the throughput of the reconstruction protocol has always remained close to the curve of the best MAC protocol. The packet loss rate and delay performance of the reconstruction protocol are also kept in a good range.

[0110] Figures 9 to 11 The performance curves of the reconstruction protocol throughput, packet loss rate and average network delay under the activity level variables are given respectively. The initial value of the activity level is set to , the range is .from Figures 9 to 11 It is not difficult to see that as the number of network nodes continues to increase, it has undergone two reconstruction processes, one in and The MAC protocol is reconfigured when the task is time-sensitive. The protocol types include p-persistent CSMA and SOTDMA. Since the task has high requirements for time sensitivity, p-persistent CSMA can provide nodes with the opportunity to fully compete for the channel, and the corresponding waiting delay is minimal, but the throughput and packet loss rate performance are not the best among all reconfigurable protocols.

[0111] It can be seen from the above simulation experiment results that the task-driven MAC protocol reconstruction method proposed in this application can take into account the performance of different dimensions of the protocol and the requirements of different tasks, and select the best MAC protocol under conditional constraints.

[0112] In one embodiment, a task-driven MAC protocol reconstruction device is provided, comprising: The load modeling module is used to consider the multi-domain distributed wireless self-organizing network scenario. The network load model is constructed through multi-dimensional feature partitioning and extraction of network load, Gaussian random process modeling and feature fusion. The protocol reconstruction module is used to build a software-defined distributed subnet node system architecture based on the network load model. The architecture consists of a task plane, a data plane, and a joint control plane. Among them, the task plane is used to cache the communication services generated during the execution of node tasks and cluster tasks in the form of data frames to the data plane to form a queue. The data plane queue contains the service data frames generated by the node tasks, and also contains the control data frames generated by the feature interaction during the MAC protocol reconstruction process and the channel reservation and preemption behavior in the MAC protocol process; The joint control plane includes the observation plane, the knowledge plane and the control plane. The observation plane is used to observe and count the services of each node, obtain the service characteristics of each node in the current reconstruction cycle through maximum likelihood estimation, and obtain the service characteristics of each node in the next reconstruction cycle through Kalman filtering prediction. Then, the service characteristics of all single-hop reachable nodes obtained by broadcast interaction are integrated to obtain the network characteristics of the next reconstruction cycle and pass them to the knowledge plane. The knowledge plane is used to carry the pre-trained classification model and output the optimal MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane based on the network features input by the observation plane and the task requirements input by the task plane. The control plane is used to carry the MAC protocol component library and determine whether the reconstruction strategy output by the knowledge plane is the same as the MAC protocol of the current reconstruction cycle. If they are the same, the reconstruction process is not performed; if they are not the same, the components are searched and the protocol parameters are adjusted to complete the MAC protocol component-level reconstruction and enter the next reconstruction cycle.

[0113] For the specific limitations of the task-driven MAC protocol reconstruction device, please refer to the limitations of the task-driven MAC protocol reconstruction method above, which will not be repeated here. Each module in the above-mentioned task-driven MAC protocol reconstruction device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0114] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.12 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a task-driven MAC protocol reconstruction method is implemented.

[0115] Those skilled in the art will understand that Fig.12The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0116] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented: Considering the multi-domain distributed wireless self-organizing network scenario, a network load model is constructed through multi-dimensional feature extraction, Gaussian random process modeling and feature fusion of network load. Based on the network load model, a software-defined distributed subnet node system architecture is constructed, which consists of the task plane, data plane and joint control plane. Among them, the task plane is used to cache the communication services generated during the execution of node tasks and cluster tasks in the form of data frames to the data plane to form a queue. The data plane queue contains the service data frames generated by the node tasks, and also contains the control data frames generated by the feature interaction during the MAC protocol reconstruction process and the channel reservation and preemption behavior in the MAC protocol process; The joint control plane includes the observation plane, the knowledge plane and the control plane. The observation plane is used to observe and count the services of each node, obtain the service characteristics of each node in the current reconstruction cycle through maximum likelihood estimation, and obtain the service characteristics of each node in the next reconstruction cycle through Kalman filtering prediction. Then, the service characteristics of all single-hop reachable nodes obtained by broadcast interaction are integrated to obtain the network characteristics of the next reconstruction cycle and pass them to the knowledge plane. The knowledge plane is used to carry the pre-trained classification model and output the optimal MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane based on the network features input by the observation plane and the task requirements input by the task plane. The control plane is used to carry the MAC protocol component library and determine whether the reconstruction strategy output by the knowledge plane is the same as the MAC protocol of the current reconstruction cycle. If they are the same, the reconstruction process is not performed; if they are not the same, the components are searched and the protocol parameters are adjusted to complete the MAC protocol component-level reconstruction and enter the next reconstruction cycle.

[0117] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented: Considering the multi-domain distributed wireless self-organizing network scenario, a network load model is constructed through multi-dimensional feature extraction, Gaussian random process modeling and feature fusion of network load. Based on the network load model, a software-defined distributed subnet node system architecture is constructed, which consists of the task plane, data plane and joint control plane. Among them, the task plane is used to cache the communication services generated during the execution of node tasks and cluster tasks in the form of data frames to the data plane to form a queue. The data plane queue contains the service data frames generated by the node tasks, and also contains the control data frames generated by the feature interaction during the MAC protocol reconstruction process and the channel reservation and preemption behavior in the MAC protocol process; The joint control plane includes the observation plane, the knowledge plane and the control plane. The observation plane is used to observe and count the services of each node, obtain the service characteristics of each node in the current reconstruction cycle through maximum likelihood estimation, and obtain the service characteristics of each node in the next reconstruction cycle through Kalman filtering prediction. Then, the service characteristics of all single-hop reachable nodes obtained by broadcast interaction are integrated to obtain the network characteristics of the next reconstruction cycle and pass them to the knowledge plane. The knowledge plane is used to carry the pre-trained classification model and output the optimal MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane based on the network features input by the observation plane and the task requirements input by the task plane. The control plane is used to carry the MAC protocol component library and determine whether the reconstruction strategy output by the knowledge plane is the same as the MAC protocol of the current reconstruction cycle. If they are the same, the reconstruction process is not performed; if they are not the same, the components are searched and the protocol parameters are adjusted to complete the MAC protocol component-level reconstruction and enter the next reconstruction cycle.

[0118] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0119] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0120] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several modifications and improvements may be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A task-driven MAC protocol reconstruction method, characterized in that: The method comprises: Considering the multi-domain distributed wireless self-organizing network scenario, a network load model is constructed through multi-dimensional feature extraction, Gaussian random process modeling and feature fusion of network load. Based on the network load model, a software-defined distributed subnet node system architecture is constructed, which consists of a task plane, a data plane, and a joint control plane; The task plane is used to cache the communication services generated during the execution of node tasks and cluster tasks in the form of data frames to form a queue on the data plane. The data plane queue includes the service data frames generated by the node tasks, and also includes the control data frames generated by the feature interaction during the MAC protocol reconstruction process and the channel reservation and preemption behavior in the MAC protocol process. The joint control plane includes an observation plane, a knowledge plane and a control plane; wherein the observation plane is used to observe and count the services of each node, obtain the service characteristics of each node in the current reconstruction period by maximum likelihood estimation, and obtain the service characteristics of each node in the next reconstruction period by Kalman filtering prediction, and then perform feature fusion on the service characteristics of all single-hop reachable nodes obtained by broadcast interaction to obtain the network characteristics of the next reconstruction period and pass them to the knowledge plane; The knowledge plane is used to carry the pre-trained classification model, and output the best MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane according to the network characteristics input by the observation plane and the task requirements input by the task plane; The control plane is used to carry the MAC protocol component library to determine whether the reconstruction strategy output by the knowledge plane is the same as the MAC protocol of the current reconstruction cycle. If they are the same, the reconstruction process is not performed; if they are not the same, the components are searched and the protocol parameters are adjusted to complete the MAC protocol component-level reconstruction, and enter the cycle of the next reconstruction cycle.

2. The method according to claim 1, characterized in that Considering the multi-domain distributed wireless self-organizing network scenario, the network load model is constructed through multi-dimensional feature partitioning and extraction of network load, Gaussian random process modeling and feature fusion, including: Considering the multi-domain distributed wireless self-organizing network scenario, the multi-dimensional characteristics of the network load are divided and extracted, including the network node scale, node activity level and node business characteristics; The service characteristics of each node in the network are modeled using a multi-dimensional Gaussian random process with the same dimension, and the service characteristics of all nodes in the network are weighted and fused using a weighted fusion algorithm of multi-dimensional Gaussian components to obtain the global network service characteristics. The network load model is constructed based on the scale of network nodes, the global network activity obtained by averaging the activity levels of all nodes, and the global network service characteristics.

3. The method according to claim 2, characterized in that A multi-dimensional Gaussian random process with the same dimension is used to model the service characteristics of each node in the network, including: Considering that each node in the network performs different tasks, we use The Gaussian random process models the business characteristics of each node in the network, which can be expressed as ,in, Reason t The set of all sample functions of the business characteristics of a single node at a certain moment, , For the i A sample function, d is the number of sample functions, that is, the number of dimensions of the Gaussian random process is consistent with the number of sample functions, For the cycle; It is used to describe the random process in which the length of the business generated by the node changes over time. The normal probability density function is expressed as: ； in, for The expected value of Represents any sample function exist Random variables at time; for The expected value of Represents any sample function In another Random variables at time; is a random variable The variance of is a random variable The variance of represents the expected calculation; is the normalized covariance matrix The determinant of ; Indicates the node generated Business length, Indicates that the task is mapped to d Business characteristic random variables of dimensions; Indicates any d a moment; and Both represent dimensions; Further considering that the tasks of each node in the network are different and the business types are relatively fixed, the business characteristics of each node are set to be wide-sense stationary within a limited period of time, and the business characteristics of each node in the network are modeled as The business characteristics of a stationary Gaussian process satisfy the following constraints: ； ； in, for Expected value; is the expected value constant, which is used to indicate that the expected value of a stationary Gaussian process does not change with time; Represents the autocorrelation function at time and The value at for and The expected value of the product of is the time difference, that is ; is the autocorrelation function, which depends only on the time difference .

4. The method according to claim 3, characterized in that The business characteristics of each node after modeling are random vectors with business length at any time. Dimensional Gaussian distribution, nodes The business length of each dimension is described as dimensional random vector , the probability density function of its business characteristics is expressed as: ； in, for t Time Node The probability density function of the business characteristics; Representation Node The mean vector of different types of business lengths, Representation Node No. i The mean of the business length, , superscript T represents transpose; is the covariance matrix, which is used to represent the node The correlation between the mean vectors of different types of business lengths, Represents the covariance matrix The determinant of ; Through each node After using the Gaussian distribution to describe the service characteristics of different nodes, the probability density function set of the service characteristics of each node in the network is expressed as ;in, is the network node size, and .

5. The method according to claim 4, characterized in that Using the weighted fusion algorithm of multi-dimensional Gaussian components, the service characteristics of all nodes in the network are weighted and fused to obtain the global network service characteristics; use The weighted fusion algorithm of Gaussian components considers that the nodes in the network meet the independent and identically distributed conditions. The goal is to fuse the business characteristics of all nodes in the network to obtain a Global network service characteristics of Gaussian distribution; Assume Node The Gaussian component of the business characteristics is ,node The activity level is , the activity level indicates the node The probability of generating business in a unit time; Perform normalization to obtain the weight of each node’s business characteristics , expressed as: ； in, satisfy and ; Based on weight Perform weighted fusion on the service features of all nodes in the network to obtain the global network service features ,and conform to Gaussian distribution, that is ; Among them, the mean vector of global network service characteristics With the covariance matrix Respectively expressed as: ； ； in, ,in, express The i means, and ; Superscript T Indicates transpose.

6. The method according to claim 5, characterized in that The network load model is constructed based on the scale of network nodes, the global network activity obtained by taking the average of the activity levels of all nodes, and the global network service characteristics, and includes: According to the scale of network nodes , the global network activity level is obtained by taking the average activity level of all nodes And the mean vector of global network service characteristics Modeling is performed to construct a network load model, where: t The expected network load at a given moment is expressed as: ； in, is the time-varying network load, for Expected value of global network activity , represents the average probability of all nodes in the network generating business per unit time; For Node activity level; For Node No. i The mean length of each business.

7. The method according to claim 1, characterized in that The observation plane observes and counts the services of each node, and obtains the service characteristics of each node in the current reconstruction period through maximum likelihood estimation, including: The observation plane maps tasks and services with the task plane to observe and count the nodes in the network in the current observation period. Number of data generated within Observation dataset with data length ; Among them, the reconstruction cycle Include time slots, observation period Include time slots, and satisfy Right now ; Calculate the nodes based on the statistical results of the number of times data is generated n The activity level is , and the data length observation data set Use it as an observation sample to estimate and predict business characteristics; use dimensional stationary Gaussian random process for nodes n The business characteristics of the node are modeled and the n Business characteristics The dimensional Gaussian distribution is reduced to a one-dimensional Gaussian distribution description, and then the node is obtained by maximum likelihood estimation based on the prior information that the data frame length variable at the sampling time obeys the Gaussian distribution. n Mean of business characteristics and standard deviation Specifically, let , for the dataset , its likelihood function is the product of the probability density functions of all observations, expressed as: ； in, Represents the observation period Neidi i The length of the data generated, and ; Taking the logarithm of the likelihood function, we get the log-likelihood function , and About and Find the partial derivative and set it equal to zero, and the solution is and Respectively expressed as: ； ； Based on the above solution, we get and , determine the node n The business characteristics in the current reconstruction cycle are subject to ;in, , is the network node size.

8. The method according to claim 7, characterized in that The observation plane predicts the service features of each node in the next reconstruction cycle through Kalman filtering, and then fuses the service features of all single-hop reachable nodes obtained through broadcast interaction to obtain the network features of the next reconstruction cycle and pass them to the knowledge plane, including: Set Nodes within the reconstruction cycle The real business characteristics of , through observation and maximum likelihood estimation, the business characteristics are obtained ;in, For Node In the The true mean of the business characteristics of the reconstruction period, and set The change of follows a random walk process; For Node In the The observed values ​​of the business characteristics of the reconstruction period represent the average data length and set The true mean Noisy measurement of For Node In the The variance of the service characteristics of the reconstruction period reflects the degree of discreteness of the data frame; The goal of Kalman filtering is to and and its corresponding variance and , prediction node In the The business characteristics of the reconstruction cycle are subject to , and then The prediction results of the reconstruction cycle are weightedly integrated with the service characteristics of all single-hop reachable nodes obtained by broadcast interaction to obtain the Finally, the network node scale, the global network activity obtained by averaging the activity of all nodes in the network, and the global network service characteristics are combined into the first The network features of the reconstruction cycle are transmitted to the knowledge plane; Among them, the Kalman filter prediction process is not performed in the first two reconstruction cycles when the network is initially established, the new nodes added to the network at the end of the reconstruction cycle do not participate in the algorithm process of the observation surface, and the observation surface only performs the process of observation, maximum likelihood estimation and feature interactive fusion in the first two reconstruction cycles.

9. The method according to claim 8, characterized in that The Kalman filter estimates the state through two steps of prediction and update. First, the prediction step obtains the The prior estimates of the mean and variance of the business characteristics of the reconstruction period are expressed as follows: ； ； in, For the A priori estimate of the mean of the business characteristics of the reconstruction period, For the The posterior estimate of the mean of the business characteristics of the reconstruction period; For the The variance of the business characteristics of the reconstruction cycle A priori estimate of For the A posterior estimate of the variance of the business characteristics of the reconstruction period; is the process noise variance; Then, the updated Kalman gain is obtained based on the update step of the posterior estimate , the expression is: ； According to the Kalman gain Respectively The posterior estimate of the mean of the service characteristics of the reconstruction period and the posterior estimate of the variance Update, expressed as: ； ； So that in the next iteration, the node In the The predicted distribution of the business characteristics of the reconstruction cycle is , and Respectively represent nodes In the The predicted values ​​of the mean and variance of the business characteristics of the reconstruction period; among them, the node In the A priori estimate of the mean of the service characteristics of the reconstruction period Prediction , The forecast is obtained by exponential smoothing method and is expressed as ;in, is a preset smoothing index, Indicates The predicted value of the variance of the business characteristics of the reconstruction period.

10. The method according to claim 1, characterized in that The knowledge plane carries a pre-trained classification model, and outputs the best MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane according to the network features input by the observation plane and the task requirements input by the task plane, including: First, the knowledge aspect is to model the task requirements based on the performance parameters of the MAX protocol in three dimensions: throughput, packet loss rate, and average network delay. This includes using effectiveness requirements, reliability requirements, and time sensitivity requirements to quantify the importance of task requirements and define effectiveness requirement parameters. , reliability requirement parameters and time-sensitive demand parameters Throughput , Packet loss rate and average network delay The performance parameters of the three dimensions are normalized so that , , , and satisfies ; Then, the normalized performance parameters and task requirement parameters are weighted and summed as the score evaluation index of different MAC protocols. It is expressed as: ； Among them, the subscript h Indicates the number of alternative MAC protocols, the superscript T represents transpose; Finally, the knowledge aspect is based on the pre-trained XGBoost classification model, with network features including network node scale, global network activity, and the mean and variance of global network business features as classification features, and validity requirement parameters as validation parameters. , reliability requirement parameters and time-sensitive demand parameters The task requirements are used as evaluation features to form data set features to classify candidate MAC protocols, and the candidate MAC protocol with the highest score is selected as the best MAC protocol that takes into account the network characteristics and task requirements of the next reconstruction cycle, expressed as: 。 11. A task-driven MAC protocol reconstruction device, characterized in that: The device comprises: The load modeling module is used to consider the multi-domain distributed wireless self-organizing network scenario. The network load model is constructed through multi-dimensional feature partitioning and extraction of network load, Gaussian random process modeling and feature fusion. A protocol reconstruction module, used to construct a software-defined distributed subnet node system architecture based on the network load model, the architecture consisting of a task plane, a data plane and a joint control plane; The task plane is used to cache the communication services generated during the execution of node tasks and cluster tasks in the form of data frames to form a queue on the data plane. The data plane queue includes the service data frames generated by the node tasks, and also includes the control data frames generated by the feature interaction during the MAC protocol reconstruction process and the channel reservation and preemption behavior in the MAC protocol process. The joint control plane includes an observation plane, a knowledge plane and a control plane; wherein the observation plane is used to observe and count the services of each node, obtain the service characteristics of each node in the current reconstruction period by maximum likelihood estimation, and obtain the service characteristics of each node in the next reconstruction period by Kalman filtering prediction, and then perform feature fusion on the service characteristics of all single-hop reachable nodes obtained by broadcast interaction to obtain the network characteristics of the next reconstruction period and pass them to the knowledge plane; The knowledge plane is used to carry the pre-trained classification model, and output the best MAC protocol reconstruction strategy for the next reconstruction cycle to the control plane according to the network characteristics input by the observation plane and the task requirements input by the task plane; The control plane is used to carry the MAC protocol component library to determine whether the reconstruction strategy output by the knowledge plane is the same as the MAC protocol of the current reconstruction cycle. If they are the same, the reconstruction process is not performed; if they are not the same, the components are searched and the protocol parameters are adjusted to complete the MAC protocol component-level reconstruction, and enter the cycle of the next reconstruction cycle.

12. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 10 are implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 10 are implemented.

Images