Method and device for improving communication efficiency of ultra-high bandwidth private network

Abstract

The invention discloses an ultrahigh-bandwidth private network communication efficiency improving method and device, and the method comprises the steps: obtaining random access requests of a plurality of MTC devices in a to-be-detected region, enabling the MTC devices to access a communication network of a private network, building an access network model of the communication network through employing a network layering algorithm according to the communication relation between the MTC devices, and improving the communication efficiency of the private network. Identifying the traffic of the communication network according to the access network model to obtain a traffic detection result, and performing traffic optimization on the communication network by adopting a data time sensitivity comprehensive scoring algorithm according to the traffic detection result to obtain a traffic scheduling strategy; a federation communication algorithm is adopted to carry out perception prediction on the bandwidth in the flow scheduling strategy to obtain a bandwidth prediction result, the problems of network congestion and uneven bandwidth resource allocation are solved, the data big and small streams in the private network are detected, the link packet loss rate is reduced, and the network communication rate is also improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of private network communication, and particularly relates to a method and device for improving the communication efficiency of a private network with ultra-high bandwidth. Background Art

[0002] With the wide application of 5G technology, more and more industries have begun to consider deploying private networks to meet their specific communication requirements. The private network provides a secure, reliable, and efficient communication environment for enterprises and organizations, etc., and can meet the requirements of large-scale data transmission, low latency, and high reliability. In particular, the private network with ultra-high bandwidth can provide stronger isolation, lower latency, and higher bandwidth control, and is called an important infrastructure for industries such as intelligent manufacturing, smart city, energy management, and transportation. Since there are a large number of user devices in the private network and the requirements for communication quality and performance are relatively high. In order to provide an efficient multi-user access mechanism to ensure that all users can access the network simultaneously and obtain high-quality communication services, therefore, there is an urgent need to provide a method and device for improving the communication efficiency of a private network with ultra-high bandwidth to solve the above-mentioned existing technical problems. Summary of the Invention

[0003] In view of this, the present invention provides a method and device for improving the communication efficiency of a private network with ultra-high bandwidth, which can solve the problems of network congestion and uneven distribution of bandwidth resources, and improve the network communication rate. The following specific technical solutions are adopted to achieve this.

[0004] In the first aspect, the present invention provides a method for improving the communication efficiency of a private network with ultra-high bandwidth, including the following steps: Obtain random access requests of multiple MTC devices in the area to be measured, and connect the MTC devices to the communication network of the private network. Among them, the area to be measured contains x available preambles PA and y MTC devices, and the MTC devices include at least one static distribution device such as a water and electricity meter reading device, a smart home device, and a remote monitoring device; Construct an access network model of the communication network by using a network layering algorithm according to the communication relationship between MTC devices, and identify the traffic size of the communication network according to the access network model to obtain a traffic detection result; Optimize the traffic of the communication network by using a data time sensitivity comprehensive scoring algorithm according to the traffic detection result to obtain a traffic scheduling strategy; Use a federated learning communication algorithm to sense and predict the bandwidth in the traffic scheduling strategy to obtain a bandwidth prediction result.

[0005] As a preference of the above technical solution, constructing an access network model of the communication network by using a network layering algorithm according to the communication relationship between MTC devices includes: Obtain an undirected topology graph of the communication network in the area to be measured according to the communication relationship between MTC devices Among them, node V is each MTC device in the area to be measured, and any two nodes that can communicate with each other are connected to form an edge E; The length of the edge in the network topology is the distance between nodes. Let the nodes and have coordinates , , then the Euclidean distance and between nodes is: That is: (1) Let d be the threshold for judging whether each node can communicate, then the communication relationship between each node is expressed as: (2) Among them, indicates that nodes and can communicate. d is calculated according to the relationship between TA and the physical distance, and the corresponding expression is: (3) Among them, c is the speed of light, SCS represents the subcarrier spacing, represents the number of FFT operation points. The adjacency matrix of the MTC devices accessing the network in the area to be measured is obtained from formula (4) as: (4) Among them, the adjacency matrix in formula (4) can be used to draw the network topology diagram of the access network in the area to be measured, and on the basis of the network topology diagram, the backbone access node set is solved. The mathematical expression of the backbone access node set is: (5) Among them, V is the node set, S is the backbone node set, is the one-hop adjacent node set of node u; When the node is in the initial state in the network topology diagram, the MTC device is in the state of being powered on but not connected. Randomly select a start node and send a handshake message to adjacent nodes, and count the score information in the received handshake information. In the network composed of the start node and adjacent nodes, set the node with the largest score as the initial node and add it to the initial solution S. The score calculation expression is: (6) Among them, is the score of node u, S is the set of backbone nodes, is the set of nodes that change from not being covered to being covered after adding u to S, Represents the frequency value of each node; the initial frequency value of each node is set to 1 and is updated iteratively through continuous local search. , for each node that is not covered , increment by 1. The larger the value of the node, the greater the probability of being selected as a backbone node.

[0006] As an optimization of the above technical solution, the specific process of establishing the initial backbone access network in the area to be measured includes: The first step: Randomly select node v and send handshake messages to its adjacent nodes, count the score information and establish the network topology diagram of the area to be measured; The second step: Compare the scores of node v and its adjacent nodes, and select the node u with the largest score to join the candidate solution S as the backbone access node; The third step: Remove the numbers of node u and its single-hop adjacent nodes from the node set; The fourth step: Repeat the second and third steps among the remaining nodes until all nodes are covered; The fifth step: The network composed of the selected backbone access nodes is the initial backbone network; Among them, according to the satisfaction conditions of the minimum backbone access node set, a mathematical model of the minimum backbone access node set is established. The mathematical model of the minimum backbone access node set is expressed as: (7) (8) (9) (10) Among them, is to solve the minimum backbone access node set, represents the communication relationship between nodes. When reducing the overlapping range of the backbone access node domination area, its constraint condition is nodes or adjacent nodes, and n is the number of nodes.

[0007] As an optimization of the above technical solution, according to the traffic detection results, a data time sensitivity comprehensive scoring algorithm is used to optimize the traffic of the communication link to obtain a traffic scheduling strategy, including: Preset M to represent the total traffic volume in the communication network. Set the large flow to represent the data flow whose traffic volume accounts for a proportion higher than of the total network traffic volume. The definition of the threshold Y is , and at this time, the large flow is the data flow in the private network whose traffic volume is higher than Y. The process of detecting large and small flows in the communication network using sampling and two-level double counting Bloom filter CBF is: (1) Sample the traffic passing through the communication network, and extract a traffic frequency every n. If the number of traffic in the link is large, reduce the sampling to shorten the large flow detection time. If the number of traffic in the link is small, increase the sampling frequency to improve the large flow detection accuracy; (2) Preset as the threshold value, which is used to detect large flows in the communication network, represents the sampling frequency of the traffic in the communication network, and based on the linear relationship, obtain the threshold value for large flow detection by the sampled traffic ; Use the same hash functions , ... . as the structural parameters of the two-level Bloom filter; Set as the length of the counter array in the first-level Bloom filter, and is greater than the power of 2, and set as the number of bits configured for each counter, then ; Set as the length of the counter array in the first-level Bloom filter, then is greater than the power of 2 of M, and set as the number of bits configured for each counter, then , where counter is a counter; (3) Use hash functions to map all the first traffic drawn to the second-level CBF. If all the counter values at the corresponding positions of the first traffic in the second-level CBF are non-zero, the first traffic is the detected large flow, and insert the first traffic into the second-level CBF to obtain the value after adding 1 to each of the counters; (4) If there is a 0 among the counter values at the corresponding position of the second traffic in the second-level Bloom filter, the second traffic is not the detected large flow. At this time, it is necessary to use hash functions to map the second traffic, so that it is mapped to the first-level Bloom filter, and calculate the minimum value of the counters; If the minimum value is equal to , the second traffic belongs to the large flow, perform flow annotation on the second traffic, subtract from each counter value, and map the second traffic to the second-level Bloom filter, and define the values of these counters as ; (5) If the minimum value of the counter is equal to If they are not equal, the corresponding third traffic is small traffic, and the third traffic is inserted into the first-level CBF to obtain the value after adding 1 to each of the counters; (6) Through steps (3) to (5), all the adopted traffic is processed by expansion. The second-level CBF is used to query the unselected traffic. If the values of all the counters are all non-zero, the corresponding fourth traffic belongs to the detected large traffic, and the fourth traffic is added to the second-level CBF to obtain the value after adding 1 to each of the counters. If there is a zero among the values of the counters, no processing is performed; After all the traffic in the ultra-high bandwidth private network communication is detected, the large traffic is marked in step (4), and the unmarked is small traffic.

[0008] As an optimization of the above technical solution, the comprehensive scoring process of data time sensitivity based on multiple indicators includes: Apply the sFlow technology to the RYU controller and the switch based on the OpenFlow protocol to obtain the data time sensitivity information of the network link. The real-time situation of the network link is obtained through the acquisition of the status information of each interface of the switch. Among them, the data time sensitivity information includes at least one of the indicators such as bandwidth utilization rate, traffic volume or packet loss situation. Calculate the maximum remaining bandwidth, data delay and data packet loss of the link on the RYU controller by using the acquired data information; According to the bandwidth information received by the RYU controller, obtain the used bandwidth of each interface of the network monitor The expression is: (11) Among them, and respectively represent the lengths of receiving and sending data within time U; the maximum remaining bandwidth of the link represents the minimum value of the remaining bandwidth among all the nodes on a link. Set as a complete link, then the maximum remaining bandwidth of the link is defined as follows: (12) Among them, and are the total bandwidth and used bandwidth corresponding to node i respectively; Calculate the first timestamp for the switch interface to collect communication data and send communication data, and subtract the second timestamp when the RYU controller receives the data from the first timestamp to obtain the delay from collection to reception of the data. Set the delay as , and set the delay for reverse transmission of data as , assuming that the delay from the switch to the RYU controller to receive data is , and its reverse time delay is , then the final average data time delay U is defined as follows: (13) Based on all the interface data received by the RYU controller to obtain the packet loss rate of each node's data on the link, then the node data packet loss rate is defined as follows: (14) Among them, and respectively represent the lengths of the lost data when the interface receives and sends data, then the packet loss rate of the path Path is defined as follows: (15) Among them, ; Expand and concretize the maximum remaining bandwidth of the link, data time delay, and data transmission packet loss rate to obtain the bandwidth , time delay and packet loss rate of path j, and adopt the weighted average method to obtain the total score of data time sensitivity, which is expressed as follows: (16) Among them, , and are the weights corresponding to the first, second, and third indicators respectively, and , the value range of j is , and z represents the number of indicators.

[0009] As the optimization of the above technical solution, distinguish time-sensitive data and non-time-sensitive data according to the comprehensive score of data time sensitivity, dynamically adjust the algorithm parameters according to the real-time situation, and select the best path according to the distinction of time sensitivity; According to the scheduling requirements of large flows and small flows and use the ant colony algorithm to dynamically adjust the parameters to determine the best path, locate the topological structure of the communication network as a weighted undirected connected graph, and represent it through , where B represents the set of all communication nodes in the communication network, then , and R is the set of all links between two communication nodes; Set to represent the probability that the ant climbs from the time-sensitive data node i of the private network to the non-time-sensitive data node j, then is defined as follows: (17) (18) Among them, is the link from node i to j, is the link the number of pheromones contained in it, is the heuristic factor selected by the node, and represents the influence factor, represents the node set formed from the i-th node to the next node, e is the remaining bandwidth, is the transmission delay, and represents the weight factor, and , , , and The value of depends on the type of data stream at this moment, and time-sensitive data and non-time-sensitive data are distinguished through probability calculation; After each ant's trail search ends, the pheromone needs to be converted into the latest data, and the update of the pheromone is described as follows: (19) Among them, represents the evaporation coefficient of the pheromone, ; represents the pheromone increment, and its value depends on the type of the current data stream. For the definition of the value is as follows: (20) Among them, represents the total amount of time-sensitive data, is the communication network path selection coefficient, and ; Define represents the optimal path set determined after the ant colony conducts several trail searches, , among which, is the number of the maximum paths; According to the small traffic in the time-sensitive data, the communication network path with the shortest delay time is used as the best path. If the current communication data stream belongs to the large flow, then select from the optimal path set network paths with larger remaining bandwidth to carry out the splitting and forwarding of the traffic, , in order to reduce the splitting difficulty and the network packet loss rate, the value range of is: (21) Among them, r is the total number of bytes of the current large flow, is the network path, ; Set path allocation weight coefficients and perform split forwarding according to to complete the communication traffic scheduling of ultra-high bandwidth, the definition of which is as follows: (22).

[0010] As an optimization of the above technical solution, the federated learning communication algorithm is used to sense and predict the bandwidth in the traffic scheduling strategy to obtain the bandwidth prediction result, including: Bandwidth sensing and training: The MTC device conducts training while performing bandwidth sensing; Bandwidth prediction: When the training of the MTC device is completed, the bandwidth is predicted based on the bandwidth sensing data to obtain the size of the data volume that can be uploaded; Compression: According to the scale of the data volume that can be uploaded, the MTC device performs an adaptive Sketch compression operation on the model: Upload: The MTC device uploads the Sketch matrix to the central server; Aggregation: The central server aggregates the Sketch matrices of each MTC device; Update: The central server distributes the aggregated and updated Sketch matrix to each MTC device; Decompression: The MTC device decompresses the downloaded Sketch matrix to restore the model.

[0011] As an optimization of the above technical solution, it is preset that the average bandwidth of the wireless channel allocation bandwidth of the MTC device during a period of time in the training process is , and the corresponding expression is: (23) where is the i-th period, is the bandwidth at time t; According to formula (24), where Z is the data volume size, T is the delay, is the signal-to-noise ratio, p is the transmission power of the MTC device, h is the channel parameter, is the Gaussian white noise variance, it can be seen that the wireless channel signal-to-noise ratio also affects the data volume Z: (24) Model the channel signal-to-noise ratio of the MTC device as a random variable , can be divided into L discrete levels, and each level corresponds to a state of the Markov chain, forming a state space , for time slot t, The channel state can be implemented as , so that represents the probability that the channel transfers from state g to state k within t. The state transition probability matrix of the MTC device can be defined as: (25) During the i-th round of training, after obtaining the data from bandwidth to bandwidth , the predicted bandwidth B is obtained using the prediction model LSTM, and the model parameter size Z is obtained according to the channel signal-to-noise ratio through formula (24); The Iperf tool is used to collect bandwidth data to test the throughput, bandwidth data, delay jitter, and data packet loss at both ends of the network. Among them, the data volume is Transmission Control Protocol (TCP) and User Datagram Protocol (UDP).

[0012] As an optimization of the above technical solution, a Sketch matrix S with a rows and b columns is obtained according to the predicted data volume size Z, where a represents the number of hash functions, and b represents the mapping space size of each hash function; During the compression process, for the model gradient , each element in the a hash function mapping vectors is mapped to different positions in the a rows. According to , the mapping position obtained through the j-th hash function is , where h is the hash function, ; A one-dimensional array q is established for each element in S, and the array is established and the data is stored, so that the data stored at is appended to the end of the array. The corresponding expression is: (26) (26) where represents appending to the end of . When the data compression is completed, further data processing is performed on the array q of each element in S, and the result is stored in S. The processing expression of is: (27) where is a set of one-dimensional gradient data; the mean function is for mean calculation, specifically: , where is the length of; the std function is for standard deviation calculation, specifically: ; The max function calculates the maximum value and returns The biggest data; for Discreteness calculation of ; is the discreteness threshold; if the discreteness is below the threshold, the mean value is used for processing, and if the discreteness is above the threshold, the maximum value is used for processing. The processed data is stored in the jth row of S in the column; According to the linear properties of Sketch, each Sketch is directly added together to obtain , and introduce a one-dimensional counting array count to record the number of times each line of Sketch is superimposed. According to count, the corresponding expression is: (28) in, yes The number of hash functions; MTC equipment receives After decompression, the corresponding expression is: (29).

[0013] In a second aspect, the present invention further provides an ultra-high bandwidth private network communication efficiency improvement device, comprising: An access request acquisition unit is used to acquire random access requests of multiple MTC devices in the area to be tested, and connect the MTC devices to the communication network of the private network, wherein the area to be tested includes x available preamble PAs and y MTC devices, and the MTC devices include at least one statically distributed device selected from water and electricity meter reading, smart home devices, and remote monitoring devices; A communication traffic identification unit, configured to construct an access network model of the communication network using a network hierarchical algorithm according to the communication relationship between the MTC devices, and to identify the traffic size of the communication network according to the access network model to obtain a traffic detection result; A traffic scheduling generation unit, used to optimize the traffic of the communication network by adopting a data time sensitivity comprehensive scoring algorithm according to the traffic detection result to obtain a traffic scheduling strategy; The ultra-high bandwidth prediction unit is used to perceive and predict the bandwidth in the traffic scheduling strategy using a federated learning communication algorithm to obtain a bandwidth prediction result.

[0014] The present invention provides a method and device for improving the communication efficiency of an ultra-high bandwidth private network. The method obtains random access requests of multiple MTC devices in a test area, connects the MTC devices to the communication network of the private network, and uses a network hierarchical algorithm to build an access network model of the communication network according to the communication relationship between the MTC devices. The traffic size of the communication network is identified according to the access network model to obtain a traffic detection result. A data time sensitivity comprehensive scoring algorithm is used according to the traffic detection result to optimize the traffic of the communication network to obtain a traffic scheduling strategy. A federated learning communication algorithm is used to perceive and predict the bandwidth in the traffic scheduling strategy to obtain a bandwidth prediction result, thereby solving the problems of network congestion and uneven allocation of bandwidth resources, detecting the size of data flows in the private network, reducing the link packet loss rate, and improving the network communication rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 A flow chart of a method for improving communication efficiency of an ultra-high bandwidth private network provided by the present invention; Figure 2 A flow chart of the initial backbone access network of the area to be tested provided by the present invention; Figure 3 A flow chart of bandwidth-aware prediction provided by the present invention; Figure 4 This is a structural block diagram of the ultra-high bandwidth private network communication efficiency improvement device provided by the present invention. DETAILED DESCRIPTION

[0017] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0018] See also Figure 1 The present invention provides a method for improving the communication efficiency of an ultra-high bandwidth private network, comprising the following steps: S1: Obtain random access requests from multiple MTC devices in the area to be tested, and connect the MTC devices to the communication network of the private network, wherein the area to be tested includes x available preamble code PAs and y MTC devices, and the MTC devices include at least one statically distributed device among water and electricity meter reading, smart home devices, and remote monitoring devices; S2: Construct an access network model of the communication network using a network layering algorithm according to the communication relationships among MTC devices, and identify the traffic volume of the communication network based on the access network model to obtain a traffic detection result; S3: Optimize the traffic of the communication network using a data time sensitivity comprehensive scoring algorithm according to the traffic detection result to obtain a traffic scheduling strategy; S4: Sense and predict the bandwidth in the traffic scheduling strategy using a federated learning communication algorithm to obtain a bandwidth prediction result.

[0019] In this embodiment, constructing an access network model of the communication network using a network layering algorithm according to the communication relationships among MTC devices includes: Obtain an undirected topology graph of the communication network in the area to be measured according to the communication relationships among MTC devices , where the node V is each MTC device in the area to be measured, and any two nodes that can communicate with each other are connected to form an edge E; The length of the edge in the network topology is the distance between nodes. Let the nodes and have coordinates , , then the Euclidean distance and between nodes is: is: (1) Let d be the threshold for judging whether each node can communicate, then the communication relationship between each node is expressed as: (2) where indicates that nodes and can communicate. d is calculated according to the relationship between TA and the physical distance, and the corresponding expression is: (3) where c is the speed of light, SCS represents the subcarrier spacing, represents the number of FFT operation points. The adjacency matrix of the access network of MTC devices in the area to be measured is obtained from formula (4) as: (4) where the adjacency matrix in formula (4) can be used to construct a network topology graph of the access network in the area to be measured, and based on the network topology graph, the backbone access node set is solved. The mathematical expression of the backbone access node set is: (5) where V is the node set and S is the backbone node set, is the set of one-hop adjacent nodes of node u; When the node is in the initial state in the network topology graph, the MTC device is in the state of being powered on but not connected. It randomly selects a starting node and sends a handshake message to adjacent nodes, and counts the score information in the received handshake messages. In the network composed of the starting node and adjacent nodes, the node with the largest score is set as the initial node and added to the initial solution S. The score calculation expression is: (6) where is the score of node u, S is the set of backbone nodes, is the set of nodes that change from uncovered to covered after adding u to S, represents the frequency value of each node; the initial frequency value of each node is set to 1 and is updated through continuous local search iteration , for each uncovered node in each iteration is incremented by 1. The larger the of the node, the greater the probability of being selected as a backbone node.

[0020] It should be noted that referring to Figure 2 , the specific process of establishing the initial backbone access network in the area to be measured includes: The first step: Randomly select node v and send a handshake message to its adjacent nodes, count the score information and establish the network topology graph of the area to be measured; The second step: Compare the scores of node v and its adjacent nodes, and select the node u with the largest score to join the candidate solution S as the backbone access node; The third step: Remove the numbers of node u and its one-hop adjacent nodes from the node set; The fourth step: Repeat the second step and the third step among the remaining nodes until all nodes are covered; The fifth step: The network composed of the selected backbone access nodes is the initial backbone network; Among them, according to the satisfaction conditions of the minimum backbone access node set, the mathematical model of the minimum backbone access node set is established. The mathematical model of the minimum backbone access node set is expressed as: (7) (8) (9) (10) where is to solve the minimum backbone access node set, Represents the communication relationship between nodes. When reducing the overlapping range of the domination areas of backbone access nodes, its constraint condition is nodes or adjacent nodes, and n is the number of nodes.

[0021] Specifically, since MTC devices initiate access requests to the base station through a contention-based random access method and the number of devices is huge, when a large number of MTC devices suddenly flood into the network and initiate random access requests to the base station, access conflicts will occur due to the competition for limited access resources, and even cause network congestion in the area to be measured. By using the idea of hierarchical access, the MTC devices are grouped for access, and a backbone access network model is established to solve the problem of insufficient access resources. First, when grouping the MTC devices in the area to be measured, it is necessary to make a network topology diagram of the access network in the area to be measured according to the communication range and location of the MTC devices; after obtaining the network topology diagram, use the timing advance TA to group and establish a backbone access network. Considering the tight access resources, the minimum backbone access node set algorithm is removed to further reduce the number of access nodes in the backbone access network and optimize the backbone access network; in order to ensure the communication quality between MTC devices and between MTC devices and the base station, the distance between nodes needs to be considered when establishing the backbone access network model. The entire network needs to be covered with the least number of access nodes, and the number of nodes managed by each access node should not vary too much, and the number of times a node is managed by multiple access nodes should be reduced, otherwise it will cause waste of access resources.

[0022] It should be understood that by obtaining the random access requests of multiple MTC devices in the area to be measured, and connecting the MTC devices to the communication network of the private network, an access network model of the communication network is constructed using a network layering algorithm according to the communication relationship between the MTC devices, and the traffic size of the communication network is identified according to the access network model to obtain a traffic detection result. According to the traffic detection result, a data time sensitivity comprehensive scoring algorithm is used to optimize the traffic of the communication network to obtain a traffic scheduling strategy, and a federated learning communication algorithm is used to sense and predict the bandwidth in the traffic scheduling strategy to obtain a bandwidth prediction result, which solves the problems of network congestion and uneven distribution of bandwidth resources, detects the large and small data flows in the private network, reduces the link packet loss rate, and also improves the network communication rate.

[0023] Optionally, using a data time sensitivity comprehensive scoring algorithm to optimize the traffic of the communication link according to the traffic detection result to obtain a traffic scheduling strategy, including: Preset M to represent the total traffic volume in the communication network, and set the large flow to represent the data flow whose traffic volume accounts for a proportion higher than of the total network traffic. The definition of the threshold Y is , and at this time, the large flow is the data flow in the private network whose traffic volume is higher than Y. The process of detecting the large and small flows in the communication network using a sampling and two-level double counting Bloom filter CBF is as follows: (1) Sample the traffic passing through the communication network, and extract a traffic frequency every n. If the number of traffic in the link is large, reduce the sampling to shorten the large flow detection time. If the number of traffic in the link is small, increase the sampling frequency to improve the large flow detection accuracy; (2) Preset representing the threshold value, which is used to detect large flows in the communication network, representing the sampling frequency of the traffic in the communication network, and obtain the threshold value for large flow detection through the sampled traffic based on a linear relationship ; Use the same hash functions , ... . as the structural parameters of the two-level Bloom filter; Set as the length of the counter array in the first-level Bloom filter, and is greater than the power of 2, and set as the number of bits configured for each counter, then ; Set as the length of the counter array in the first-level Bloom filter, then is greater than the power of 2 of M, and set as the number of bits configured for each counter, then , where counter is a counter; (3) Use hash functions to map all the first traffic drawn to the second-level CBF. If all the counter values at the corresponding positions of the first traffic in the second-level CBF are non-zero, the first traffic is the detected large flow, and insert the first traffic into the second-level CBF to obtain the value after adding 1 to each of the counters; (4) If there is a 0 among the counter values at the corresponding position of the second traffic in the second-level Bloom filter, the second traffic is not the detected large flow. At this time, it is necessary to use hash functions to map the second traffic, so that it is mapped to the first-level Bloom filter, and calculate the minimum value of the counters; If the minimum value is equal to , the second traffic belongs to the large flow, perform flow annotation on the second traffic, subtract from each counter value, and map the second traffic to the second-level Bloom filter, and define the values of these counters as ; (5) If the minimum value of the counter is equal to If they are not equal, the corresponding third traffic is small traffic, and inserting the third traffic into the first-level CBF gives the value after incrementing each of the counters by 1; (6) Perform the processing on all the adopted traffic through steps (3) to (5), and use the second-level CBF to query the traffic that has not been drawn. If the values of all the counters are non-zero, the corresponding fourth traffic belongs to the detected large traffic, and adding the fourth traffic to the second-level CBF gives the value after incrementing each of the counters by 1. If there is a zero among the values of the counters, no processing is taken; After all the traffic in the ultra-high bandwidth private network communication is detected, the large traffic is marked in step (4), and the unmarked is small traffic.

[0024] In this embodiment, the comprehensive scoring process of data time sensitivity based on multiple indicators includes: Apply the sFlow technology to the RYU controller and the switch based on the OpenFlow protocol to obtain the data time sensitivity information of the network link. Obtain the real-time situation of the network link through the acquisition of the status information of each interface of the switch. Among them, the data time sensitivity information includes at least one indicator such as bandwidth utilization rate, traffic volume, or packet loss situation. Calculate the maximum remaining bandwidth, data delay, and data packet loss of the link on the RYU controller using the acquired data information; According to the bandwidth information received by the RYU controller, obtain the used bandwidth of each interface of the network monitor The expression is: (11) Among them, and respectively represent the lengths of receiving and sending data within time U; the maximum remaining bandwidth of the link represents the minimum value of the remaining bandwidth among all nodes on a link. Set as a complete link, then the maximum remaining bandwidth of the link is defined as follows: (12) Among them, and are the total bandwidth and used bandwidth corresponding to node i respectively; Calculate the first timestamp for the switch interface to collect communication data and send communication data, and subtract the second timestamp of the data received by the RYU controller from the first timestamp to obtain the delay from collection to reception of the data. Set the delay as , and set the delay for reverse transmission of data as , assuming the delay from the switch to the RYU controller for receiving data is , and its reverse time delay is , then the final average data time delay U is defined as follows: (13) Based on all the interface data received by the RYU controller to obtain the packet loss rate of the data of each node on the link, then the node data packet loss rate is defined as follows: (14) where and respectively represent the data lengths lost when the interface receives and sends data, then the packet loss rate of the path Path is defined as follows: (15) where ; Expand and visualize the maximum remaining bandwidth, data time delay, and data transmission packet loss rate of the link to obtain the bandwidth , time delay and packet loss rate of path j, and adopt the weighted average method to obtain the total score of the data time sensitivity, which is expressed as follows: (16) where , and are the weights corresponding to the first, second, and third indicators respectively, and , the value range of j is , and z represents the number of indicators.

[0025] It should be noted that time-sensitive data and non-time-sensitive data are distinguished according to the comprehensive score of data time sensitivity, the algorithm parameters are dynamically adjusted according to the real-time situation, and the best path is selected according to the distinction of time sensitivity; According to the scheduling requirements of large flows and small flows and using the dynamic parameter adjustment of the ant colony algorithm to determine the best path, the topological structure of the communication network is located as a weighted undirected connected graph, and is represented by , where B represents the set of all communication nodes in the communication network, then , and R is the set of all links between two communication nodes; Set to represent the probability that the ant climbs from the time-sensitive data node i of the private network to the non-time-sensitive data node j, then is defined as follows: (17) (18) Among them, is the link from node i to j, is the link the number of pheromones contained in, is the heuristic factor selected by the node, and represents the influence factor, represents the node set formed from the i-th node to the next node, e is the remaining bandwidth, is the transmission delay, and represents the weight factor, and , , , and The value of depends on the type of data stream at this moment, and time-sensitive data and non-time-sensitive data are distinguished through probability calculation; After each ant tracing ends, the pheromone needs to be converted into the latest data, and the update of the pheromone is described as follows: (19) Among them, represents the evaporation coefficient of the pheromone, ; represents the pheromone increment, and its value depends on the type of the current data stream. For the definition of the value is as follows: (20) Among them, represents the total amount of time-sensitive data, is the communication network path selection coefficient, and ; Define represents the set of optimal paths determined after the ant colony conducts several tracings, , among which, is the number of the maximum paths; According to the small traffic in the time-sensitive data, the communication network path with the shortest delay time is used as the best path. If the current communication data stream belongs to the large flow, then select from the set of optimal paths network paths with larger remaining bandwidth to expand the splitting and forwarding of the traffic, , in order to reduce the splitting difficulty and the network packet loss rate, the value range of is: (21) Among them, r is the total number of bytes of the current large flow, is the network path, ; Set path allocation weight coefficients , and perform split forwarding according to to complete the communication traffic scheduling for ultra-high bandwidth, The definition of is as follows: (22).

[0026] Specifically, the multi-indicators include the maximum remaining bandwidth of the link, data delay, and data transmission packet loss rate. By performing communication traffic detection, the large flows and small flows of the private network communication network can be distinguished, which helps to reasonably allocate resources, meet the high-bandwidth requirements of large flows, optimize the transmission strategy, improve network performance, and perform fine network scheduling to improve the overall efficiency and service quality. Be brave in the division of time sensitivity, combine large and small flow data to complete scheduling, and avoid the problem of scheduling conflicts caused by the separation of time-sensitive data traffic scheduling and routing, resulting in time slot contention in the scheduling of time-sensitive traffic data and non-time-sensitive traffic data under different traffic conditions.

[0027] Optionally, refer to Figure 3 , and use the federated learning communication algorithm to sense and predict the bandwidth in the traffic scheduling strategy to obtain the bandwidth prediction result, including: Bandwidth sensing and training: The MTC device conducts training while performing bandwidth sensing; Bandwidth prediction: When the training of the MTC device is completed, bandwidth prediction is performed based on the bandwidth-sensed data to obtain the size of the data volume that can be uploaded; Compression: According to the scale of the data volume that can be uploaded, the MTC device performs an adaptive Sketch compression operation on the model: Upload: The MTC device uploads the Sketch matrix to the central server; Aggregation: The central server aggregates the Sketch matrices of each MTC device; Update: The central server distributes the aggregated and updated Sketch matrix to each MTC device; Decompression: The MTC device decompresses the downloaded Sketch matrix to restore the model.

[0028] In this embodiment, it is preset that the average bandwidth of the wireless channel allocation bandwidth of the MTC device during a period of time in the training process is , and the corresponding expression is: (23) Among them, is the i-th period of time, is the bandwidth at time t; According to formula (24), where Z is the data volume size and T is the delay, Let \(SNR\) be the signal-to-noise ratio, \(p\) be the transmission power of the MTC device, and \(h\) be the channel parameter. Let \(\sigma^{2}\) be the variance of Gaussian white noise. It can be seen that the wireless channel signal-to-noise ratio also affects the data volume \(Z\): (24) Model the channel signal-to-noise ratio of the MTC device as a random variable , which can be divided into \(L\) discrete levels. Each level corresponds to a state of the Markov chain, forming a state space . For time slot \(t\), the channel state of can be realized as , such that represents the probability that the channel transfers from state \(g\) to state \(k\) within \(t\). The state transition probability matrix of the MTC device can be defined as: (25) During the \(i\)-th round of training, after obtaining the data from bandwidth to bandwidth , use the prediction model LSTM to obtain the predicted bandwidth \(B\), and obtain the model parameter size \(Z\) according to the channel signal-to-noise ratio through formula (24); Use the Iperf tool to collect bandwidth data to test the throughput, bandwidth data, delay jitter, and data packet loss at both ends of the network. Among them, the data volume is Transmission Control Protocol (TCP) and User Datagram Protocol (UDP).

[0029] It should be noted that according to the predicted data volume size \(Z\), obtain the \(a\times b\) Sketch matrix \(S\), where \(a\) represents the number of hash functions, and \(b\) represents the mapping space size of each hash function; During the compression process, for the model gradient , map each element in the \(a\) hash function mapping vectors to different positions in the \(a\) rows. According to , the mapping position obtained through the \(j\)-th hash function is , where \(h\) is the hash function, ; Establish a one-dimensional array \(q\) for each element in \(S\), and establish the array and store the data, so that the stored at the position is appended to the end of the array. The corresponding expression is: (26) Among them, represents Append at the end. When data compression is completed, further data processing is performed on the array q of each element in S, and the result is stored in S. The processing expression is: (27) Where is a group of one-dimensional gradient data; the mean function is for mean calculation, specifically: , where is the length of; the std function is for standard deviation calculation, specifically: ; the max function is for maximum value calculation, returning the maximum data; is the dispersion calculation of; is the dispersion threshold; if the dispersion calculation is below the threshold, mean processing is used, and if the dispersion is above the threshold, maximum value processing is adopted. The processed data is stored in the j-th row column of S; According to the linear property of Sketch, each Sketch is directly added correspondingly to obtain , and a one-dimensional counting array count is introduced to record the number of times each row of Sketch is superimposed. For calculate the average according to count, and the corresponding expression is: (28) Where is the number of hash functions of; The MTC device performs decompression after receiving , and the corresponding expression is: (29).

[0030] Specifically, since the bandwidth in the limited network is stable, complex wireless networks can be considered. In the current wireless channel frequency division multiplexing, each time slot makes a decision according to the data transmission situation of the current MTC device. When allocating bandwidth, different self-confidence levels are allocated according to the data volume of each time slot. In order to predict the bandwidth, the size of the sub-channel allocated to the wireless channel for each time slot can be predicted. The maximum data volume that can be transmitted by the current channel is obtained by predicting the bandwidth allocated to the MTC device by the prediction network, and then the local model is compressed to improve the communication efficiency of continuous shift learning.

[0031] Refer to Figure 3 , the present invention also provides a device for improving the communication efficiency of a special high-bandwidth network, including: An access request acquisition unit, configured to acquire random access requests of multiple MTC devices in a to-be-tested area, and connect the MTC devices to the communication network of the private network. The to-be-tested area includes x available preambles PA and y MTC devices. The MTC devices include at least one static distribution device such as a water and electricity meter reading device, a smart home device, and a remote monitoring device; A communication traffic identification unit, configured to construct an access network model of the communication network by using a network layering algorithm according to the communication relationship between MTC devices, and identify the traffic size of the communication network according to the access network model to obtain a traffic detection result; A traffic scheduling generation unit, configured to optimize the traffic of the communication network by using a data time sensitivity comprehensive scoring algorithm according to the traffic detection result to obtain a traffic scheduling strategy; An ultra-high bandwidth prediction unit, configured to sense and predict the bandwidth in the traffic scheduling strategy by using a federated learning communication algorithm to obtain a bandwidth prediction result.

[0032] In all the examples shown and described here, any specific value should be construed as merely exemplary and not as a limitation. Therefore, other examples of the exemplary embodiments may have different values.

[0033] It should be noted that like reference numerals and letters denote like items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0034] The above-described embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but it should not be construed as a limitation to the scope of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can be made, and these all belong to the protection scope of the present invention.

Claims

1. A method for improving the communication efficiency of an ultra-high bandwidth private network, characterized in that: The following steps are involved: Obtain random access requests of multiple MTC devices in the area to be tested, and connect the MTC devices to the communication network of the private network, wherein the area to be tested includes x available preamble code PAs and y MTC devices, and the MTC devices include at least one statically distributed device selected from water and electricity meter reading, smart home devices, and remote monitoring devices; According to the communication relationship between MTC devices, a network layering algorithm is used to build an access network model of the communication network, and the traffic size of the communication network is identified according to the access network model to obtain a traffic detection result; According to the traffic detection results, the data time sensitivity comprehensive scoring algorithm is used to optimize the traffic of the communication network to obtain the traffic scheduling strategy; The federated learning communication algorithm is used to perceive and predict the bandwidth in the traffic scheduling strategy to obtain the bandwidth prediction result.

2. The method for improving the communication efficiency of an ultra-high bandwidth private network according to claim 1, characterized in that: According to the communication relationship between MTC devices, a network layering algorithm is used to construct an access network model of the communication network, including: Obtain an undirected topology graph of the communication network in the area to be tested based on the communication relationship between MTC devices , where node V is each MTC device in the area to be tested, and any two nodes that can communicate with each other are connected to form an edge E; The length of the edge in the network topology is the distance between nodes. and The coordinates of , , then the node and The Euclidean distance between for: （1） Let d be the threshold for determining whether each node can communicate, then the communication relationship between each node is It is expressed as: （2） in, Representation Node and Communicable, d is calculated based on the relationship between TA and physical distance, and the corresponding expression is: （3） Where c is the speed of light, SCS represents the subcarrier spacing, represents the number of FFT operation points. The adjacency matrix of the MTC device access network in the test area is obtained by formula (4): （4） Among them, the adjacency matrix in formula (4) can be used to make a network topology diagram of the access network in the area to be tested, and the backbone access node set can be solved based on the network topology diagram. The mathematical expression of the backbone access node set is: （5） Among them, V is the node set, S is the backbone node set, is the set of one-hop adjacent nodes of node u; When the node is in the initial state in the network topology diagram, the MTC device is powered on but not connected. A starting node is randomly selected and a handshake message is sent to the adjacent nodes. The score information in the received handshake message is counted. In the network composed of the starting node and the adjacent nodes, the node with the largest score is set as the initial node and added to the initial solution S. The score calculation expression is: （6） in, is the score of node u, S is the set of backbone nodes, is the set of nodes that change from uncovered to covered after u is added to S. Represents the frequency value of each node; the initial frequency value of each node is set to 1, and is updated through continuous local search iterations , each time the number of nodes not covered is Increase by 1, the node The larger it is, the greater the probability of being selected as a backbone node.

3. The method for improving the communication efficiency of an ultra-high bandwidth private network according to claim 2, characterized in that: The specific process of establishing the initial backbone access network in the area to be tested includes: Step 1: Randomly select node v to send handshake messages to its adjacent nodes, count the score information and establish the network topology map of the area to be tested; Step 2: Compare the scores of node v and its adjacent nodes, and select the node u with the largest score to join the candidate solution S as the backbone access node; Step 3: Remove the numbers of node u and its one-hop adjacent nodes from the node set; Step 4: Repeat steps 2 and 3 for the remaining nodes until all nodes are covered; Step 5: The network formed by the selected backbone access nodes is the initial backbone network; Among them, a mathematical model of the minimum backbone access node set is established according to the conditions that the minimum backbone access node set meets, and the mathematical model of the minimum backbone access node set is expressed as: （7） （8） （9） （10） in, is to find the minimum set of backbone access nodes. It represents the communication relationship between nodes. When reducing the overlapping range of the backbone access node control area, its constraint condition is the node or adjacent node, and n is the number of nodes.

4. The method for improving the communication efficiency of an ultra-high bandwidth private network according to claim 1, characterized in that: According to the traffic detection results, the data time sensitivity comprehensive scoring algorithm is used to optimize the traffic of the communication link to obtain the traffic scheduling strategy, including: The default value M represents the total amount of traffic in the communication network, and the default value M represents the proportion of traffic to the total amount of network traffic that is higher than The data stream, the threshold Y is defined as At this time, the large flow is the data flow with a flow volume higher than Y in the private network. The large and small flow detection process in the communication network using sampling and two-stage double counting Bloom filter CBF is as follows: (1) Sampling the traffic passing through the communication network, and extracting a traffic frequency every n. If the amount of traffic in the link is large, reducing the sampling to shorten the large flow detection time; if the amount of traffic in the link is small, increasing the sampling frequency to improve the large flow detection accuracy; (2) Preset represents the threshold value, which is used to detect large flows in the communication network. Indicates the sampling frequency of traffic in the communication network. Based on the linear relationship, the threshold for large flow detection is obtained by extracting the traffic. ; the same Hash functions , ... . As the structural parameters of the two-stage Bloom filter; set is the length of the counter array in the first-level Bloom filter, and Greater than of the power of 2, and set The number of bits configured for each counter, then ;set up is the length of the counter array in the first-level Bloom filter, then is greater than M to the power of 2, and set The number of bits configured for each counter, then , where counter is a counter; (3) Adoption A hash function maps all the extracted first flows to the secondary CBF. If the first flow is in the corresponding position in the secondary CBF The counter values ​​are all non-zero, the first flow is the detected large flow, and the first flow is inserted into the secondary CBF to obtain The value of each counter after adding 1; (4) If the second flow is at the corresponding position in the secondary Bloom filter If the counter value is 0, the second flow is not the detected large flow, and you need to use A hash function is used to map the second traffic to the first-level Bloom filter, and the The minimum value of the counters; if the minimum value is equal to If they are equal, the second flow belongs to the large flow. Expand the flow label for the second flow and subtract each counter value from , and maps the second flow to the secondary Bloom filter, defining this The value of the counter is ; (5) If the minimum value of the counter is equal to If they are not equal, the corresponding third flow is a small flow. Insert the third flow into the first-level CBF to obtain The value of each counter after adding 1; (6) Process all the traffic through steps (3) to (5), and use the secondary CBF to query the traffic that is not extracted. If If the values ​​of all counters are non-zero, the corresponding fourth flow belongs to the detected large flow. The fourth flow is added to the secondary CBF to obtain The value of each counter after adding 1, if If the value of a counter is zero, no action is taken; After all traffic in the ultra-high broadband private network communication is detected, the ones marked in step (4) are large flows, and the ones without marks are small flows.

5. The method for improving the communication efficiency of an ultra-high bandwidth private network according to claim 4, characterized in that: The comprehensive scoring process of data time sensitivity based on multiple indicators includes: The sFlow technology is applied to the RYU controller and the switch using the OpenFlow protocol to obtain the data time information of the network link. The real-time status of the network link is obtained by obtaining the status information of each interface of the switch. The data time information includes at least one indicator of bandwidth utilization, traffic volume or packet loss. The maximum remaining bandwidth, data delay and data packet loss of the link are calculated on the RYU controller using the obtained data information. According to the bandwidth information received by the RYU controller, the used bandwidth of each interface of the network monitor is obtained The expression is: （11） in, and They represent the length of data received and sent in time U respectively; the maximum remaining bandwidth of a link represents the minimum value of the remaining bandwidth among all nodes on a link. is a complete link, then the maximum remaining bandwidth of the link is The definition is as follows: （12） in, and are the total bandwidth and used bandwidth corresponding to node i respectively; Calculate the first timestamp of the switch interface collecting and sending communication data, and subtract the second timestamp of the RYU controller receiving the data from the first timestamp to get the delay from data collection to data reception. Set the delay as , and set the delay of data transmission in the reverse direction to , assuming that the delay of data sent from the switch to the RYU controller is , and its reverse delay is , then the final average data delay U is defined as follows: （13） Based on all the interface data received by the RYU controller to obtain the packet loss rate of each node data on the link, the node data packet loss rate The definition is as follows: （14） in, and The length of data lost when the interface receives and sends data, respectively, and the data packet loss rate of the path Path The definition is as follows: （15） in, ; The maximum remaining bandwidth of the link, data delay and data transmission packet loss rate are processed concretely to obtain the bandwidth of path j , Delay and packet loss rate , and the weighted average method is used to obtain the total score of data time sensitivity The expression is as follows: （16） in, , and are the weights corresponding to the first, second and third indicators respectively, and , the value of j is , z represents the number of indicators.

6. The method for improving the communication efficiency of an ultra-high bandwidth private network according to claim 5, characterized in that: Also includes: Differentiate between time-sensitive data and non-time-sensitive data based on the comprehensive score of data time sensitivity, dynamically adjust algorithm parameters based on real-time conditions, and select the best path based on the differentiation of time sensitivity; According to the scheduling requirements of large and small flows and the use of dynamic parameter adjustment of the ant colony algorithm to determine the best path, the topological structure of the positioning communication network is a weighted undirected connected graph, and In other words, B represents the set of all communication nodes in the communication network. , R is the set of all links between two communication nodes; set up represents the probability that an ant crawls from a time-sensitive data node i to a non-time-sensitive data node j in a private network, then is defined as follows: （17） （18） in, is the link from node i to j, For Link The number of pheromones contained in The heuristic factor chosen for the node, and represents the impact factor, represents the node set formed from the i-th node to the next node, e is the remaining bandwidth, is the transmission delay, and represents the weight factor, and , , , and The value of depends on the type of data flow at this moment, and time-sensitive data is distinguished from non-time-sensitive data through probability calculation; After each ant's search, the pheromone is converted to the latest data. The description of updating the pheromone is as follows: （19） in, represents the volatility coefficient of pheromone, ; Indicates the pheromone increment, its value depends on the type of current data stream. The values ​​are defined as follows: （20） in, Indicates the total amount of time-sensitive data, is the communication network path selection coefficient, and ; definition It represents the optimal path set determined by the ant colony after several times of searching. ,in, is the maximum number of paths; According to the small flow in the time-sensitive data, the communication network path with the shortest delay time is selected as the best path. If the current communication data flow belongs to a large flow, select the best path from the optimal path set. The remaining bandwidth of the network path is large, and the traffic is split and forwarded. ,In order to reduce the segmentation difficulty and network packet loss rate, The value range of is: （21） Among them, r is the total number of bytes of the current large flow, is the network path, ; Set the path allocation weight coefficient , and follow Expand split forwarding to complete ultra-high bandwidth communication traffic scheduling, is defined as follows: （22）。 7. The method for improving the communication efficiency of an ultra-high bandwidth private network according to claim 1, characterized in that: The federated learning communication algorithm is used to perceive and predict the bandwidth in the traffic scheduling strategy to obtain bandwidth prediction results, including: Bandwidth perception and training: MTC devices are trained and perform bandwidth perception at the same time; Bandwidth prediction: When the MTC device training is completed, the bandwidth is predicted based on the bandwidth perception data to obtain the amount of data that can be uploaded; Compression: Based on the size of the data that can be uploaded, the MTC device performs adaptive Sketch compression on the model: Upload: The MTC device uploads the Sketch matrix to the central server; Aggregation: The central server aggregates the Sketch matrix of each MTC device; Update: The central server sends the aggregated and updated Sketch matrix to each MTC device; Decompression: The MTC device decompresses the downloaded Sketch matrix to restore the model.

8. The method for improving the communication efficiency of an ultra-high bandwidth private network according to claim 7, characterized in that: The average bandwidth of the wireless channel allocation bandwidth of the preset MTC device during the training process is , the corresponding expression is: （23） in, is the i-th period, is the bandwidth at time t; According to formula (24), where Z is the data size and T is the delay, is the signal-to-noise ratio, p is the transmission power of the MTC device, h is the channel parameter, is the Gaussian white noise variance, we can know the wireless channel signal-to-noise ratio It also affects the amount of data Z: （24） Channel signal-to-noise ratio for MTC devices Modeled as a random variable , It can be divided into L discrete levels, each level corresponds to a state of the Markov chain, forming a state space , for time slot t, The channel state can be realized as ,make express The probability of the channel transitioning from state g to state k within t, the state transition probability matrix of the MTC device can be defined as: （25） During the i-th round of training, we get the bandwidth To bandwidth data After that, the prediction model LSTM is used to obtain the prediction bandwidth B, and the model parameter size Z is obtained according to the channel signal-to-noise ratio through formula (24); Use the Iperf tool to collect bandwidth data to test the throughput, bandwidth data, delay jitter, and packet loss at both ends of the network. The data volume is the Transmission Control Protocol TCP and the User Datagram Protocol UDP.

9. The method for improving the communication efficiency of an ultra-high bandwidth private network according to claim 8, characterized in that: Also includes: According to the predicted data size Z, a Sketch matrix S with a rows and b columns is obtained, where a represents the number of hash functions and b represents the size of the mapping space of each hash function; During the compression process, for the model gradient , will map the vector through a hash function To each element in a row, according to The mapping position obtained by the jth hash function is , where h is the hash function, ; For each element in S, create a one-dimensional array q. Creating an Array And store the data so that it is stored in Positional Append to the end of the array, the corresponding expression is: （26） in, Indicates that Add in At the end, when the data compression is completed, further data processing is performed on the array q of each element in S, and the results are stored in S. The processing expression is: （27） in, is a set of one-dimensional gradient data; the mean function is the mean calculation, specifically: ,in for The length of the function; std is the standard deviation calculation, specifically: ; The max function calculates the maximum value and returns The biggest data; for Discreteness calculation of ; is the discreteness threshold; if the discreteness is below the threshold, the mean value is used for processing, and if the discreteness is above the threshold, the maximum value is used for processing. The processed data is stored in the jth row of S in the column; According to the linear properties of Sketch, each Sketch is directly added together to obtain , and introduce a one-dimensional counting array count to record the number of times each line of Sketch is superimposed. According to count, the corresponding expression is: （28） in, yes The number of hash functions; MTC equipment receives After decompression, the corresponding expression is: （29）。 10. An ultra-high bandwidth private network communication efficiency improvement device, characterized in that: The method for improving the communication efficiency of an ultra-high bandwidth private network as claimed in any one of claims 1 to 9 comprises: An access request acquisition unit is used to acquire random access requests of multiple MTC devices in the area to be tested, and connect the MTC devices to the communication network of the private network, wherein the area to be tested includes x available preamble PAs and y MTC devices, and the MTC devices include at least one statically distributed device selected from water and electricity meter reading, smart home devices, and remote monitoring devices; A communication traffic identification unit, configured to construct an access network model of the communication network using a network hierarchical algorithm according to the communication relationship between the MTC devices, and to identify the traffic size of the communication network according to the access network model to obtain a traffic detection result; A traffic scheduling generation unit, used to optimize the traffic of the communication network by adopting a data time sensitivity comprehensive scoring algorithm according to the traffic detection result to obtain a traffic scheduling strategy; The ultra-high bandwidth prediction unit is used to perceive and predict the bandwidth in the traffic scheduling strategy using a federated learning communication algorithm to obtain a bandwidth prediction result.

Images