Data transmission method based on TSN edge computing system

Abstract

The invention relates to the technical field of data transmission, and discloses a data transmission method based on a TSN edge computing system, and the method comprises the steps: carrying out the multi-dimensional feature analysis of data entering the TSN edge computing system through employing a machine learning algorithm, obtaining an analysis result, carrying out the preliminary classification of data through employing a DBSCAN algorithm based on the analysis result according to a multi-dimensional feature vector, and carrying out the data classification through employing a DBSCAN algorithm. According to the complete technical system constructed by the invention, from data feature analysis and classification, network state prediction, resource allocation optimization to transmission path intelligent selection and dynamic adjustment, all links are closely coordinated and supplement each other, potential network fluctuations and faults are effectively dealt with, the stability and reliability of key data transmission are ensured, and the system is suitable for large-scale popularization and application. According to the whole system, through dynamic and self-adaptive resource management and data transmission strategies, the resource utilization efficiency, the data transmission performance and the capability of dealing with network risks are comprehensively improved, and continuous optimization of the system can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of data transmission, and more specifically to a data transmission method based on a TSN edge computing system. Background Art

[0002] With the development of Industry 4.0 and the Internet of Things, edge computing systems have gradually become a key architecture for processing real-time data. As a new generation of Ethernet standard, time-sensitive network TSN provides a guarantee for low-latency and deterministic data transmission.

[0003] In the existing TSN edge computing system data transmission technology, network resource allocation is not accurate and flexible enough, data transmission path planning mostly relies on fixed strategies, and it is difficult to cope with complex and changeable network environments. The transmission result feedback mechanism is also not perfect enough and cannot effectively achieve continuous optimization of the system. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a data transmission method based on a TSN edge computing system to solve the problems existing in the above-mentioned background technology.

[0005] The present invention provides the following technical solutions:

[0006] A data transmission method based on a TSN edge computing system comprises the following steps:

[0007] S1: Use machine learning algorithms to perform multi-dimensional feature analysis on the data entering the TSN edge computing system to obtain analysis results. Based on the analysis results, use the DBSCAN algorithm to preliminarily divide the data categories according to the multi-dimensional feature vectors. When new data flows in, dynamically update the cluster center and boundaries according to the distance between the new data and the category center and the surrounding density, and add or reduce categories in a timely manner so that the categories can evolve adaptively with the data features. Dynamically divide the data into multiple categories, establish a priority evaluation model, and assign dynamically changing priorities to each category, thereby obtaining data category priorities;

[0008] S2: In the TSN edge computing system, a time series prediction model is introduced to predict the network traffic, node load, and bandwidth change trend in the future, and obtain the prediction results. Based on the data category priority in step S1 and the prediction results, a dynamic programming algorithm is used to carry out network resource allocation to obtain the network resource allocation results.

[0009] S3: Deploy intelligent sensing nodes in the TSN edge computing system. The nodes are equipped with a feedback mechanism based on link quality change monitoring. First, monitor the link packet loss rate, delay, bandwidth utilization and other indicators at the initial sampling frequency, adjust the sampling frequency according to the fluctuation of link quality indicators, collect link quality data at different times, analyze the spatial correlation coefficient and time correlation coefficient of link quality data at different times, and use weighted edges to represent links when constructing the network topology. The weight integrates the current quality data with the spatial correlation coefficient and the time correlation coefficient to give the link a comprehensive weight.

[0010] S4: Using reinforcement learning algorithm, intelligently select the optimal transmission path according to the data priority and current network status. During the transmission process, if the sensing node detects that the link status has deteriorated or a failure has occurred, the path reselection mechanism is immediately triggered. The reinforcement learning algorithm quickly determines the new optimal path according to the new network status to ensure the continuity and efficiency of data transmission.

[0011] Preferably, in S1, a machine learning algorithm is used to perform multi-dimensional feature analysis on the data entering the TSN edge computing system to obtain analysis results. The analysis steps are as follows:

[0012] S11: Set the data point set entering the TSN edge computing system to be X = {x1, x2, ..., x n}, where each data point x i Represents a two-dimensional vector, x i1 and x i2 Represents the data point x i The eigenvalues ​​in two different dimensions, x i =(x i1 , x i2 ), i = 1, 2, ..., n, n is the total number of data points;

[0013] S12: For two-dimensional data point x i =(x i1 , x i2 ) is standardized so that the two-dimensional data point x i Convert to a distribution with mean 0 and variance 1:

[0014]

[0015] In the formula, and Represents the data point x i The mean in two different dimensions, and Represents the data point x i variance in two different dimensions;

[0016] S13: Calculate the standardized x i1b and x i2b , the calculation formula is:

[0017]

[0018]

[0019] In the formula, s1 and s2 are and The square root of is the standard deviation;

[0020] S14: Reference the second data point x j =(x j1 , x j2 ), indicating x j The values ​​in two dimensions for two data points x i =(x i1 , x i2 ) and x j =(x j1 , x j2 ), the Euclidean distance formula is used to calculate the distance between two data points. The Euclidean distance formula is as follows:

[0021]

[0022] In the formula, d(x i , x j ) represents the data point x i and x j The distance between

[0023] S15: Traverse each data point x in the data set i , the number of data points in the neighborhood with ∈ as radius is calculated by the density formula, ∈ represents the neighborhood range of the data point, and the density formula is as follows:

[0024] N ∈ (x i )={x j ∈X|d(x i ,x j )≤∈};

[0025] Where N ∈ (x i ) represents the data point x i The set of data points in the ∈-neighborhood of;

[0026] S16: Statistical data point set N ∈ (x i ) to get the number of elements in |N ∈ (x i )|;

[0027] S17: Set the integer parameter to Min, and |N ∈ (x i )|Compare with:

[0028] When |N ∈ (x i )|≥Min, then x i Marked as core points;

[0029] When |N ∈ (x i )|<Min, then x i are marked as boundary points;

[0030] S18: If the data point x i If it is neither a core point nor a boundary point, it is marked as a noise point;

[0031] S19: Starting from any core point that has not been visited, put it into a new cluster C k , initially k = 1, k represents the total number of clusters, and then all core points in all neighborhoods and boundary points falling in the neighborhood of these core points are added to cluster C k In this way, a cluster is formed;

[0032] S20: Repeat steps S11 to S19 until all core points have been visited, and finally all data points are divided into different clusters, completing the preliminary data category division.

[0033] Preferably, in S1, when new data flows in, the cluster center and boundary are dynamically updated according to the distance between the new data and the category center and the surrounding density. The updating steps are as follows:

[0034] When a new data point x new When inflow;

[0035] Set the zth cluster C z The center μ z =(μ z1 , μ z2 ), z = 1, 2, ..., k, for μ z1 and μ z2 The calculation formula is as follows:

[0036]

[0037] In the formula, |C z | represents cluster C z The number of data points in z1 and μ z2 Represents cluster Cz The coordinate values ​​of the center in the first and second dimensions;

[0038] Calculate the new data point x by the density calculation formula in step S15 new The number of data points in the neighborhood with ∈ as radius is obtained. ∈ (x new )|;

[0039] When |N ∈ (x new )|≥Min, then x new Marked as core points;

[0040] When |N ∈ (x new )|<Min, then x new are marked as boundary points;

[0041] Set a priority function P(C z ) to form each cluster C z Assign priority and calculate cluster C z The proportion of the number of data points to the total number of data points is P1(C z ), the calculation formula is as follows:

[0042]

[0043] Set each cluster C z The number of times new data flows in per unit time is denoted as f z , calculate the cluster C z The data update frequency accounts for the proportion of all cluster data update frequencies P2(C z ), the calculation formula is as follows:

[0044]

[0045] In the formula, f k Represents cluster C z The number of new data flows per unit time, used to measure the clustering C z Data update speed;

[0046] For the priority function P(C z ) is calculated, and the calculation formula is as follows:

[0047] P(C z )=w1P1(C z )+w2P2(C z );

[0048] In the formula, w1 and w2 represent the corresponding 1(C z ) and P2(Cz )’s weight, and w1+w2=1.

[0049] Preferably, regarding S2, in the TSN edge computing system, a time series prediction model is introduced to predict the network traffic, node load, and bandwidth change trend in the future, and obtain the prediction results. The prediction process is as follows:

[0050] Assume that the traffic time series is T(t), the node load time series is L(Δt), and the bandwidth time series is B(t), where t represents the time step. The moving average prediction model is used to predict the traffic change trend. The traffic prediction formula is as follows:

[0051]

[0052] Where kz represents the predicted time step index, kz = 1, 2, ..., nu, nu represents the prediction step, mu represents the moving average window size, T cy (t+kz) represents the predicted flow value, and io represents the summation index variable, which is used in the summation operation. It is used to traverse the time steps from t-mu+1 to t to obtain the flow data T(io) corresponding to these time steps;

[0053] The node load value is predicted, and the node load value prediction formula is as follows:

[0054]

[0055] Where, L cy (t+kz) represents the predicted node load value, and L(io) represents the summation of io. The time steps from t-mu+1 to t are traversed in the figure, and the node load data corresponding to these time steps;

[0056] The bandwidth value is predicted. The bandwidth value prediction formula is as follows:

[0057]

[0058] In the formula, B(io) represents the summation operation. The bandwidth data corresponding to these time steps are traversed from t-mu+1 to t, B cy (t+kz) represents the predicted bandwidth value;

[0059] Set the high priority data transmission interruption risk function to R h (t), the network congestion risk function is R c (t), for R h (t) is calculated using the following formula:

[0060]

[0061] Where fp h represents the empirical function, P h (t) indicates the current transmission rate and bandwidth occupancy of high-priority data;

[0062] R c (t) is calculated using the following formula:

[0063] R c (t) = fp h (T cy (t+kz), L cy (t+kz), B cy (t+kz));

[0064] Set the bandwidth required for high priority data to BW h (t), the cache resource is CH h (t), the redundant bandwidth ratio is α h , the redundant cache ratio is β h ;

[0065] For high-priority data resources, the allocation formula is as follows:

[0066] When R h (t)>TU h And R c (t)>CH h (t), then:

[0067] BW h (t) = BW req (t)+α h ×BW ectra (t);

[0068] When R h (t)≤TU h And R c (t)≤CH h (t), then:

[0069] BW h (t) = BW req (t);

[0070] In the formula, TU h Indicates the interruption risk threshold of high priority data, BW req (t) represents the basic bandwidth requirement of the current high-priority data, BW ectra (t) represents the additional bandwidth resources, α h Indicates the redundant bandwidth ratio of high priority data;

[0071] When Rh (t)>TU h And R c (t)>CH h (t), then:

[0072] CH h (t) = CH req (t)+β h ×CH ectra (t);

[0073] When R h (t)≤TU h And R c (t)≤CH h (t), then:

[0074] CH h (t) = CH req (t);

[0075] In the formula, CH req (t) represents the basic cache requirement, CH ectra (t) represents additional cache resources, β h Indicates the redundant cache ratio of high-priority data;

[0076] For medium priority data resources, set the required bandwidth to BW. m (t), the dynamic adjustment factor is γ m , the resource allocation is AW m (t), the allocation formula for medium priority data resources is as follows:

[0077] When R h (t)>CH m ,but:

[0078] BW m (t) = AW m (t)+γ m ×ΔAW m (t);

[0079] When R h (t)≤CH m ,but:

[0080] BW m (t) = AW m (t);

[0081] Where, ΔAW m (t) represents the resource adjustment amount, CH m Indicates the impact limit of the network congestion risk threshold on medium priority data;

[0082] For low priority data, set its occupied bandwidth to BW l (t), the low priority data resource allocation formula is as follows:

[0083] When R h (t)>TU h And R c (t)>CW l (t), then:

[0084] BW l (t) = 0;

[0085] When R h (t)≤TU h And R c (t)≤CW l (t), then:

[0086] BW l (t) = BW l (t);

[0087] In the formula, CW l (t) represents the impact limit of the network congestion risk threshold on low-priority data;

[0088] In the TSN edge computing system, when faced with resource constraints and high risks, low-priority data is reasonably transferred to ensure efficient and stable allocation of resources across the entire network.

[0089] Preferably, in the TSN edge computing system, when faced with resource constraints and high risk, low-priority data is reasonably transferred to ensure efficient and stable allocation of the entire network resources. The transfer process is as follows:

[0090] Set the current path of low priority data to P l (t);

[0091] The set of alternate transmission paths is {P s1 , P s2 ,…,P sv}, sv represents the total number of paths, P is represents the is-th backup transmission path, is = 1, 2, ..., sv, and the transfer threshold is TU tq ;

[0092] When R c (t)>TU tq And R h (t)>TU tq , and after satisfying the following condition 1, then:

[0093] PE new (t) = P is ;

[0094] Where, PE new (t) represents the new transmission path of low priority data after the routing strategy is selected;

[0095] Condition 1 is as follows:

[0096]

[0097] In the formula, Cost(P is ) represents the backup transmission path P is Cost (P js ) represents the measurement of the backup transmission path P js The cost function, P js represents the set of alternate transmission paths {P s1 , P s2 ,…,P sv} any alternate route, js represents the index variable;

[0098] When R c (t)≤TU tq And R h (t)≤TU tq , and after condition 2 is met, then:

[0099] PE new (t) = P l (t);

[0100] Where P l (t) represents the transmission path used by low-priority data before the routing strategy is selected;

[0101] Condition 2 is as follows:

[0102]

[0103] Cost(P js ) is calculated, and the calculation formula is as follows:

[0104]

[0105] In the formula, DR js Indicates the backup path P js Delay, LR js Indicates the backup path P js Load condition, BR js Indicates the backup path P js bandwidth availability, wn1, wn2 and wn3 represent the bandwidth availability of DR js , L.R. js and BR js The weight coefficient of .

[0106] Preferably, in S3, the sampling frequency is adjusted according to the fluctuation of the link quality indicator, and the adjustment steps are as follows:

[0107] Set the initial sampling frequency of the intelligent sensing node to fa int ,The link packet loss rate is LPR, where 0≤LPR≤1, the link delay is DPR, and the link bandwidth utilization rate is BUR, where 0≤BUR≤1;

[0108] The sampling frequency formula is as follows:

[0109]

[0110] In the formula, fa new represents the new sampling frequency adjusted according to the fluctuation of the link quality indicator, αz represents the adjustment coefficient, αz is a real number greater than 0, β LPR , β DPR and β BUR denote the weight coefficients corresponding to LPR, DPR and BUR, respectively, and β LPR +β DPR +β BUR =1, LPR avg 、DPR avg and BUR avg They represent the historical averages of LPR, DPR and BUR respectively. max 、DPR max and BUR max They represent the historical maximum values ​​of LPR, DPR and BUR respectively.

[0111] Preferably, in S3, the time series change trend correlation of the link quality data at different times is analyzed, and the analysis steps are as follows:

[0112] S31: The packet loss rate data sequence collected by the intelligent sensing node is determined as:

[0113] {LPR1, LPR2, ..., LPR ea};

[0114] Set the delayed data sequence to {DPR1, DPR2, ..., DPR ea}, the bandwidth utilization data sequence is {BUR1, BUR2, …, BUR ea}, where ea is the number of data points collected;

[0115] Assume that the current link is link A, and its adjacent link set is {BA1, BA2, ..., BA mr ,}, where BA ivc represents the ivcth adjacent link, ivc = 1, 2, ..., mr;

[0116] S32: For link A and a certain adjacent link BA ivz A quality indicator XUQ, where XUQ represents any one of LPR, DPR and BUR, and the measured values ​​at tn time points are:

[0117] QUR A ={XUQ A1 ,XUQ A2 ,…,XUQ Atn};

[0118]

[0119] In the formula, Indicates the adjacent link BA ivz The measured value of the quality indicator XUQ at tn time points, XUQ A represents the measurement value sequence of the quality indicator XUQ of link A at tn time points, Indicates the adjacent link BA ivz The measurement value sequence of the quality indicator XUQ at tn time points, XUQ Atn represents the measured value of the quality indicator XUQ of link A at tn time points;

[0120] S33: Calculate the spatial correlation coefficient CHE, and the calculation formula is as follows:

[0121]

[0122] In the formula, XUQ Ajxb represents the measured value of the quality indicator XUQ of link A at the jxbth time point, represents the average value of the quality indicator XUQ of link A, Indicates the adjacent link BA ivz The measured value of the quality indicator XUQ at the jxbth time point, Indicates the adjacent link BA ivz The average value of the quality indicator XUQ measurement value;

[0123] S34: Obtain the link packet loss rate spatial correlation coefficient CHE respectively through the calculation formulas of steps S32 to S33 LPR , delay spatial correlation coefficient CHE DPR and bandwidth utilization spatial correlation coefficient CHE BUR ;

[0124] S35: Calculate the comprehensive spatial correlation coefficient CHE zz , the calculation formula is as follows:

[0125]

[0126] Wherein, wju, wju2 and wju3 represent the corresponding CHE LPR CHE DPR and CHE BUR The weight coefficient of .

[0127] Preferably, in S3, the time correlation coefficient of the quality data of the link at different times is analyzed, and the analysis steps are as follows:

[0128] S36: Set the time series data of the link quality indicator YBB to:

[0129] {YBB1, YBB2, ..., YBB ncb};

[0130] Where ncb represents the total number of link quality indicator time series data points, YBB represents any one of LPR, DPR and BUR, and the lag order is ksb, ksb<ncb;

[0131] S37: Calculate the time correlation coefficient ACF(ksb), the calculation formula is as follows:

[0132]

[0133] Where, YBB tn represents the measured value of the link quality indicator YBB at the tnth time point, Indicates the average value of the link quality indicator YBB measurement, YBB tn+ksb represents the measured value of the link quality indicator YBB at the tn+ksb-th time point;

[0134] S38: The calculation formulas of the above steps S36 to S37 are respectively obtained:

[0135] Link packet loss rate time correlation coefficient ACF LPR (ksb), delay time correlation coefficient ACF DPR (ksb) and bandwidth utilization time correlation coefficient ACF BUR (ksb);

[0136] S39: Calculate the comprehensive time correlation coefficient ACF tr , the calculation formula is as follows:

[0137]

[0138] Where wxx DPR 、wxx DPR and wxx BUR Respectively represent the corresponding ACF LPR(ksb), ACF DPR (ksb) and ACF BUR The weight coefficient of (ksb);

[0139] S40: Assign a comprehensive weight to the link. The assignment process is as follows:

[0140] Set the packet loss rate of the current link to LPR tnew , the delay is DPR tnew , bandwidth utilization BUR tnew , calculate the link weight WLL, the calculation formula is as follows:

[0141]

[0142] Where wll LPR 、wll DPR and wll BUR They are the weight coefficients of LPR, DPR and BUR in link calculation respectively.

[0143] Preferably, in S4, the optimal transmission path is intelligently selected according to the data priority and the current network status. The selection process is as follows:

[0144] Through the intelligent sensing node to continuously monitor the link status, for any link loe, calculate the time window TYY kjz Packet loss rate LPR loe , the calculation formula is as follows:

[0145]

[0146] In the formula, NNQ lost Indicates the number of lost packets, NNQ tatal Indicates the total number of packets sent;

[0147] For any link loe, calculate the delay change rate ΔDPR loe , the calculation formula is as follows:

[0148]

[0149] In the formula, DPR curre Indicates the average delay at the current moment, DPR prev represents the average delay of the previous moment;

[0150] For any link loe, calculate the bandwidth utilization change rate ΔBUR loe , the calculation formula is as follows:

[0151] ΔBUR loe =BUR curre -BURprev ;

[0152] Where BUR curre Indicates the current bandwidth utilization, BUR prev represents the bandwidth utilization rate at the previous moment; the fault judgment function of any link is set to FGG loe , if any one of the conditions 3 is met, then: FGG loe =1, it is determined that any link loe fails, triggering the path reselection mechanism;

[0153] Condition 3 is as follows:

[0154] LPR loe >LPR thred ;

[0155] ΔDPR loe >ΔDPR thred ;

[0156] ΔBUR loe >ΔBUR thred ;

[0157] If any one of the conditions 4 is met, then:

[0158] FGG loe =0, it is determined that no link loe fails;

[0159] Condition 4 is as follows:

[0160] LPR loe ≤LPR thred ;

[0161] ΔDPR loe ≤ΔDPR thred ;

[0162] ΔBUR loe ≤ΔBUR thred ;

[0163] Among them, LPR thred , ΔDPR thred and ΔBUR thred They represent the packet loss rate fault threshold, delay change rate threshold, and bandwidth utilization change rate threshold respectively.

[0164] Preferably, in S4, the path reselection mechanism is triggered, and the triggering steps are as follows:

[0165] S41: Set the reward function to RNV, and calculate the reward function RNV. The calculation formula is as follows:

[0166]

[0167] In the formula, RYC delay represents the reward function related to transmission delay, DPR bw represents the reward function related to link bandwidth utilization, PC prooc represents the reward function related to data priority, RYX reliby represents the reward function related to link reliability, wsa1, wsa2, wsa3 and wsa4 correspond to RYC respectively. delay 、DPR bw 、PC prooc and RYX reliby The weight coefficient of

[0168] S42: RYC delay The calculation process is as follows:

[0169] When delay≤dela ytrld ,but:

[0170] RYC delay =kbiq1×(delay trld -delay);

[0171] When delay>delay trld , then RYC delay =0;

[0172] Among them, delay represents the actual data transmission delay time, delay trld represents the transmission delay threshold, kbiq1 represents the proportional coefficient of transmission delay;

[0173] S43: DPR bw The calculation formula is as follows:

[0174] DPR bw =kbiq2×bw utizon ;

[0175] In the formula, kbiq2 represents the proportional coefficient of bandwidth utilization, bw utizon Indicates the bandwidth utilization of the current link;

[0176] S44: For PC prooc The calculation formula is as follows:

[0177] PC prooc =base+kbiq3×prooc pri ;

[0178] In the formula, base represents the basic reward value set according to data of different priorities, kbiq3 represents the coefficient for adjusting the influence of transmission performance on the reward, and proocpri Indicators related to data transmission performance;

[0179] S45: RYX reliby The calculation formula is as follows:

[0180] RYX reliby =kbiq4×(1-faire)×qucbbzu;

[0181] Where kbiq4 represents the proportional coefficient of link reliability, which is used to adjust the magnitude of link reliability reward, faire represents the historical failure frequency of the link, ranging from 0 to 1, and qucbbzu represents the current quality status indicator of the link, ranging from 0 to 1;

[0182] S46: When any link loe fails and the path reselection mechanism is triggered, the reward function RNV is adjusted and a new reward function RNV is calculated. new , the calculation process is as follows:

[0183]

[0184] In the formula, RYC newdelay represents the adjusted reward function related to transmission delay, RYX newreliby represents the adjusted reward function related to link reliability, wsa new1 and wsa new4 Corresponding to RYC newdelay and RYX newreliby Weight coefficient;

[0185] S47: RYC newdelay The calculation process is as follows:

[0186] When delay≤delay newtrld ,but:

[0187] RYC newdelay =kbiq new1 ×(delay newtrld -delay);

[0188] When delay>delay newtrld , then RYC newdelay =0;

[0189] Where delay newtrld Indicates the adjusted transmission delay threshold, kbi qnew1 represents the proportionality factor of the adjusted transmission delay;

[0190] S48: RYX newrelibyThe calculation formula is as follows:

[0191] RYX newreliby =kbiq new4 ×(1-faire)×qucbbzu;

[0192] In the formula, kbi qnew4 The proportionality factor representing the adjusted link reliability;

[0193] S49: Set that there are nkaa candidate paths;

[0194] For each candidate path set {HxJJ1, HxJJ2, …, HxJJ iaa}, where iaa = 1, 2, ..., nkaa. When the path reselection mechanism is triggered, the new reward function value RNV corresponding to each candidate path is calculated through steps S46 to S48. niccq ;

[0195] Where inccq = 1, 2, ..., nkaa;

[0196] S50: Compare the new reward values ​​of all candidate paths, and obtain the candidate path with the largest reward value. The comparison formula is as follows:

[0197]

[0198] In the formula, HxJJ znew Represents the new optimal transmission path, HxJJ nkaa In the process of path reselection, it refers to the set of all candidate paths {HxJJ1, HxJJ2, …, HxJJ iaa}, the candidate path with the largest reward value, max{RNV1, RNV2, …, RNV nakk ,} represents the reward value set {RNV1, RNV2, …, RNV nakk ,}, where represents the conditional qualifier, used to accurately describe HxJJ znew is the path that satisfies the specific reward value condition, RNV nkaa Indicates the reward value corresponding to the nkaa-th candidate path;

[0199] S51: Determine the new optimal transmission path HxJJ znew After that, data transmission is switched from the faulty path to the optimal transmission path HxJJ znew superior.

[0200] Technical effects and advantages of the present invention:

[0201] In the TSN edge computing system, the present invention comprehensively uses a variety of advanced technical means to significantly improve the system's data processing and transmission efficiency and resource management capabilities. It uses a machine learning algorithm to perform multi-dimensional data feature analysis and combines the DBSCAN algorithm to achieve accurate adaptive division of data categories and dynamic priority evaluation. At the same time, it introduces a time series prediction model to grasp network status changes in advance, so that resource allocation can be closely aligned with actual data needs and network dynamic changes. This not only improves resource utilization efficiency and avoids idle and wasted resources, but also ensures that high-priority data is always adequately protected, effectively reducing data transmission delays, losses, and interruption risks, enhancing the system's processing capabilities and service quality for various types of data, and improving the overall stability and reliability of the system, thereby providing users with a better, more stable, and efficient network service experience.

[0202] The deployment of intelligent sensing nodes of the present invention and the establishment of a feedback mechanism based on link quality change monitoring, combined with in-depth analysis of the spatiotemporal correlation of link quality data and the construction of a weighted network topology map, provide the system with accurate network status perception capabilities. On this basis, the reinforcement learning algorithm intelligently selects the optimal transmission path based on data priority and real-time network status, and quickly triggers the path reselection mechanism when the link is abnormal, greatly improving the success rate and continuity of data transmission. This method effectively reduces transmission interruptions and retransmission times caused by link problems, reduces transmission delays, optimizes the utilization efficiency of network resources, and enables the system to flexibly respond to complex and changeable network environments, significantly enhancing the adaptability and robustness of the system, meeting the needs for efficient and reliable data transmission in diversified business scenarios, and improving the performance of the system under different network conditions.

[0203] The present invention organically integrates core factors such as data transmission delay, bandwidth utilization, data priority satisfaction and link reliability by comprehensively considering the reward function of multiple key data transmission objectives, providing scientific and comprehensive guidance for the path selection behavior of reinforcement learning agents. By reasonably setting the weight coefficients of various factors, the various objectives are intelligently and accurately weighed in the path selection process to pursue the maximum comprehensive reward. This innovative mechanism enables the system to further optimize the bandwidth utilization of the overall network while ensuring the quality of high-priority data transmission, effectively reduce transmission delays, and improve the overall performance and efficiency of the system. At the same time, the ability to flexibly adjust the weight coefficient according to actual application scenarios and needs gives the system greater adaptability and flexibility, enabling it to better adapt to the unique needs of different business scenarios, effectively enhancing the competitiveness and practicality of the TSN edge computing system, and laying a solid foundation for its wide application in complex and changeable network application fields.

[0204] The complete technical system constructed by the present invention closely cooperates and complements each other in all links, from data feature analysis and classification, network status prediction, resource allocation optimization to intelligent selection and dynamic adjustment of transmission paths. This highly integrated and intelligent design enables the TSN edge computing system to effectively respond to potential network fluctuations and failures by reserving sufficient current resources and redundant resources based on risk assessment when facing high-priority data transmission needs, and ensure the stability and reliability of key data transmission. For medium-priority data, resources are reasonably allocated and dynamically adjusted during resource idle periods based on traffic and real-time requirements, achieving a balance between efficient resource utilization and timeliness of data transmission. Low-priority data makes full use of network idle resource gaps for transmission, and gives priority to releasing resources when resources are tight, further optimizing resource allocation. Through this dynamic and adaptive resource management and data transmission strategy, the entire system comprehensively improves resource utilization efficiency, data transmission performance, and the ability to cope with network risks, and can achieve continuous optimization of the system, showing excellent performance and high adaptability in complex network environments, providing strong technical support for the realization of efficient, intelligent, and reliable edge computing network services, with significant technical advantages and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0205] Figure 1 This is a flowchart of the steps of the data transmission method based on the TSN edge computing system of the present invention. DETAILED DESCRIPTION

[0206] The present invention is further described below in conjunction with specific embodiments. However, people familiar with the art should understand that the detailed description given here in conjunction with the drawings is for better explanation, and the structure of the present invention necessarily exceeds these limited embodiments. For some equivalent replacement schemes or common means, they are no longer described in detail herein, but still belong to the scope of protection of the present application.

[0207] Figure 1 The best embodiment of the present invention is shown below in conjunction with the attached Figure 1 The present invention is further described.

[0208] A data transmission method based on a TSN edge computing system comprises the following steps:

[0209] S1: Use machine learning algorithms to perform multi-dimensional feature analysis on the data entering the TSN edge computing system to obtain analysis results. Based on the analysis results, use the DBSCAN algorithm to preliminarily divide the data categories according to the multi-dimensional feature vectors. When new data flows in, dynamically update the cluster center and boundaries according to the distance between the new data and the category center and the surrounding density, and add or reduce categories in a timely manner so that the categories can evolve adaptively with the data features. Dynamically divide the data into multiple categories, establish a priority evaluation model, and assign dynamically changing priorities to each category, thereby obtaining data category priorities;

[0210] S2: In the TSN edge computing system, a time series prediction model is introduced to predict the network traffic, node load, and bandwidth change trend in the future, and obtain the prediction results. Based on the data category priority in step S1 and the prediction results, a dynamic programming algorithm is used to carry out network resource allocation to obtain the network resource allocation results.

[0211] S3: Deploy intelligent sensing nodes in the TSN edge computing system. The nodes are equipped with a feedback mechanism based on link quality change monitoring. First, monitor the link packet loss rate, delay, bandwidth utilization and other indicators at the initial sampling frequency, adjust the sampling frequency according to the fluctuation of link quality indicators, collect link quality data at different times, analyze the spatial correlation coefficient and time correlation coefficient of link quality data at different times, and use weighted edges to represent links when constructing the network topology. The weight integrates the current quality data with the spatial correlation coefficient and the time correlation coefficient to give the link a comprehensive weight.

[0212] S4: Using reinforcement learning algorithm, intelligently select the optimal transmission path according to the data priority and current network status. During the transmission process, if the sensing node detects that the link status has deteriorated or a failure has occurred, the path reselection mechanism is immediately triggered. The reinforcement learning algorithm quickly determines the new optimal path according to the new network status to ensure the continuity and efficiency of data transmission.

[0213] As a further step of the present invention, in S1, a machine learning algorithm is used to perform multi-dimensional feature analysis on the data entering the TSN edge computing system to obtain analysis results. The analysis steps are as follows:

[0214] S11: Set the data point set entering the TSN edge computing system to be X = {x1, x2, ..., x n}, where each data point x i Represents a two-dimensional vector, x i1 and x i2 Represents the data point x i The eigenvalues ​​in two different dimensions, x i =(x i1 , x i2), i = 1, 2, ..., n, n is the total number of data points;

[0215] S12: For two-dimensional data point x i =(x i1 , x i2 ) is standardized so that the two-dimensional data point x i Convert to a distribution with mean 0 and variance 1:

[0216]

[0217] In the formula, and Represents the data point x i The mean in two different dimensions, and Represents the data point x i variance in two different dimensions;

[0218] S13: Calculate the standardized x i1b and x i2b , the calculation formula is:

[0219]

[0220] In the formula, s1 and s2 are and The square root of is the standard deviation;

[0221] S14: Reference the second data point x j =(x j1 , x j2 ), indicating x j The values ​​in two dimensions for two data points x i =(x i1 , x i2 ) and x j =(x j1 , x j2 ), the Euclidean distance formula is used to calculate the distance between two data points. The Euclidean distance formula is as follows:

[0222]

[0223] In the formula, d(x i , x j ) represents the data point x i and x j The distance between

[0224] S15: Traverse each data point x in the data set i, the number of data points in the neighborhood with ∈ as radius is calculated by the density formula, ∈ represents the neighborhood range of the data point, and the density formula is as follows:

[0225] N ∈ (x i )={x j ∈X|d(x i ,x j )≤∈};

[0226] Where N ∈ (x i ) represents the data point x i The set of data points in the ∈-neighborhood of;

[0227] S16: Statistical data point set N ∈ (x i ) to get the number of elements in |N ∈ (x i )|;

[0228] S17: Set the integer parameter to Min, and |N ∈ (x i )|Compare with:

[0229] When |N ∈ (x i )|≥Min, then x i Marked as core points;

[0230] When |N ∈ (x i )|<Min, then x i are marked as boundary points;

[0231] S18: If the data point x i If it is neither a core point nor a boundary point, it is marked as a noise point;

[0232] S19: Starting from any core point that has not been visited, put it into a new cluster C k , initially k = 1, k represents the total number of clusters, and then all core points in all neighborhoods and boundary points falling in the neighborhood of these core points are added to cluster C k In this way, a cluster is formed;

[0233] S20: Repeat steps S11 to S19 until all core points have been visited, and finally all data points are divided into different clusters, completing the preliminary data category division.

[0234] As a further aspect of the present invention, in S1, when new data flows in, the cluster center and the boundary are dynamically updated according to the distance between the new data and the category center and the surrounding density, and the updating steps are as follows:

[0235] When a new data point x new When inflow;

[0236] Set the zth cluster C z The center μ z =(μ z1 , μ z2 ), z = 1, 2, ..., k, for μ z1 and μ z2 The calculation formula is as follows:

[0237]

[0238] In the formula, |C z | represents cluster C z The number of data points in z1 and μ z2 Represents cluster C z The coordinate values ​​of the center in the first and second dimensions;

[0239] Calculate the new data point x by the density calculation formula in step S15 new The number of data points in the neighborhood with ∈ as radius is obtained. ∈ (x new )|;

[0240] When |N ∈ (x new )|≥Min, then x new Marked as core points;

[0241] When |N ∈ (x new )|<Min, then x new are marked as boundary points;

[0242] Set a priority function P(C z ) to form each cluster C z Assign priority and calculate cluster C z The proportion of the number of data points to the total number of data points is P1(C z ), the calculation formula is as follows:

[0243]

[0244] Set each cluster C z The number of times new data flows in per unit time is denoted as f z , calculate the cluster C zThe data update frequency accounts for the proportion of all cluster data update frequencies P2(C z ), the calculation formula is as follows:

[0245]

[0246] In the formula, f k Represents cluster C z The number of new data flows per unit time, used to measure the clustering C z Data update speed;

[0247] For the priority function P(C z ) is calculated, and the calculation formula is as follows:

[0248] P(C z )=w1P1(C z )+w2P2(C z );

[0249] In the formula, w1 and w2 represent the corresponding 1(C z ) and P2(C z )’s weight, and w1+w2=1.

[0250] As a further aspect of the present invention, regarding S2, in the TSN edge computing system, a time series prediction model is introduced to predict the network traffic, node load, and bandwidth change trend in the future, and obtain the prediction results. The prediction process is as follows:

[0251] Assume that the traffic time series is T(t), the node load time series is L(Δt), and the bandwidth time series is B(t), where t represents the time step. The moving average prediction model is used to predict the traffic change trend. The traffic prediction formula is as follows:

[0252]

[0253] Where kz represents the predicted time step index, kz = 1, 2, ..., nu, nu represents the prediction step, mu represents the moving average window size, T cy (t+kz) represents the predicted flow value, and io represents the summation index variable, which is used in the summation operation. It is used to traverse the time steps from t-mu+1 to t to obtain the flow data T(io) corresponding to these time steps;

[0254] The node load value is predicted, and the node load value prediction formula is as follows:

[0255]

[0256] Where, L cy(t+kz) represents the predicted node load value, and L(io) represents the summation of io. The time steps from t-mu+1 to t are traversed in the figure, and the node load data corresponding to these time steps;

[0257] The bandwidth value is predicted. The bandwidth value prediction formula is as follows:

[0258]

[0259] In the formula, B(io) represents the summation operation. The bandwidth data corresponding to these time steps are traversed from t-mu+1 to t, B cy (t+kz) represents the predicted bandwidth value;

[0260] Set the high priority data transmission interruption risk function to R h (t), the network congestion risk function is R c (t), for R h (t) is calculated using the following formula:

[0261]

[0262] Where fp h represents the empirical function, P h (t) indicates the current transmission rate and bandwidth occupancy of high-priority data;

[0263] R c (t) is calculated using the following formula:

[0264] R c (t) = fp h (T cy (t+kz), L cy (t+kz), B cy (t+kz));

[0265] Set the bandwidth required for high priority data to BW h (t), the cache resource is CH h (t), the redundant bandwidth ratio is α h , the redundant cache ratio is β h ;

[0266] For high-priority data resources, the allocation formula is as follows:

[0267] When R h (t)>TU h And R c (t)>CH h (t), then:

[0268] BWh (t) = BW req (t)+α h ×BW ectra (t);

[0269] When R h (t)≤TU h And R c (t)≤CH h (t), then:

[0270] BW h (t) = BW req (t);

[0271] In the formula, TU h Indicates the interruption risk threshold of high priority data, BW req (t) represents the basic bandwidth requirement of the current high-priority data, BW ectra (t) represents the additional bandwidth resources, α h Indicates the redundant bandwidth ratio of high priority data;

[0272] When R h (t)>TU h And R c (t)>CH h (t), then:

[0273] CH h (t) = CH req (t)+β h ×CH ectra (t);

[0274] When R h (t)≤TU h And R c (t)≤CH h (t), then:

[0275] CH h (t) = CH req (t);

[0276] In the formula, CH req (t) represents the basic cache requirement, CH ectra (t) represents additional cache resources, β h Indicates the redundant cache ratio of high-priority data;

[0277] For medium priority data resources, set the required bandwidth to BW. m (t), the dynamic adjustment factor is γ m , the resource allocation is AW m (t), the allocation formula for medium priority data resources is as follows:

[0278] When R h (t)>CH m ,but:

[0279] BW m (t) = AW m (t)+γ m ×ΔAW m (t);

[0280] When R h (t)≤CH m ,but:

[0281] BW m (t) = AW m (t);

[0282] Where, ΔAW m (t) represents the resource adjustment amount, CH m Indicates the impact limit of the network congestion risk threshold on medium priority data;

[0283] For low priority data, set its occupied bandwidth to BW l (t), the low priority data resource allocation formula is as follows:

[0284] When R h (t)>TU h And R c (t)>CW l (t), then:

[0285] BW l (t) = 0;

[0286] When R h (t)≤TU h And R c (t)≤CW l (t), then:

[0287] BW l (t) = BW l (t);

[0288] In the formula, CW l (t) represents the impact limit of the network congestion risk threshold on low-priority data;

[0289] In the TSN edge computing system, when faced with resource constraints and high risks, low-priority data is reasonably transferred to ensure efficient and stable allocation of resources across the entire network.

[0290] As a further feature of the present invention, in the TSN edge computing system, when faced with a situation of resource shortage and high risk, low-priority data is reasonably transferred to ensure efficient and stable allocation of the entire network resources. The transfer process is as follows:

[0291] Set the current path of low priority data to P l (t);

[0292] The set of alternate transmission paths is {P s1 , P s2 ,…,P sv}, sv represents the total number of paths, P is represents the is-th backup transmission path, is = 1, 2, ..., sv, and the transfer threshold is TU tq ;

[0293] When R c (t)>TU tq And R h (t)>TU tq , and after satisfying the following condition 1, then:

[0294] PE new (t) = P is ;

[0295] Where, PE new (t) represents the new transmission path of low priority data after the routing strategy is selected;

[0296] Condition 1 is as follows:

[0297]

[0298] In the formula, Cost(P is ) represents the backup transmission path P is Cost (P js ) represents the measurement of the backup transmission path P js The cost function, P js represents the set of alternate transmission paths {P s1 , P s2 ,…,P sv} any alternate route, js represents the index variable;

[0299] When R c (t)≤TU tq And R h (t)≤TU tq , and after condition 2 is met, then:

[0300] PE new (t) = P l (t);

[0301] Where P l (t) represents the transmission path used by low-priority data before the routing strategy is selected;

[0302] Condition 2 is as follows:

[0303]

[0304] Cost(P js ) is calculated, and the calculation formula is as follows:

[0305]

[0306] In the formula, DR js Indicates the backup path P js Delay, LR js Indicates the backup path P js Load condition, BR js Indicates the backup path P js bandwidth availability, wn1, wn2 and wn3 represent the bandwidth availability of DR js , L.R. js and BR js The weight coefficient of .

[0307] As a further step of the present invention, in S3, the sampling frequency is adjusted according to the fluctuation of the link quality indicator, and the adjustment steps are as follows:

[0308] Set the initial sampling frequency of the intelligent sensing node to fa int ,The link packet loss rate is LPR, where 0≤LPR≤1, the link delay is DPR, and the link bandwidth utilization rate is BUR, where 0≤BUR≤1;

[0309] The sampling frequency formula is as follows:

[0310]

[0311] In the formula, fa new represents the new sampling frequency adjusted according to the fluctuation of the link quality indicator, αz represents the adjustment coefficient, αz is a real number greater than 0, β LPR , β DPR and β BUR denote the weight coefficients corresponding to LPR, DPR and BUR, respectively, and β LPR +β DPR +β BUR =1, LPR avg 、DPR avg and BUR avg They represent the historical averages of LPR, DPR and BUR respectively. max 、DPRmax and BUR max They represent the historical maximum values ​​of LPR, DPR and BUR respectively.

[0312] As a further step of the present invention, regarding S3, analyzing the correlation of time series change trends of link quality data at different times, the analysis steps are as follows:

[0313] S31: The packet loss rate data sequence collected by the intelligent sensing node is determined as:

[0314] {LPR1, LPR2, ..., LPR ea};

[0315] Set the delayed data sequence to {DPR1, DPR2, ..., DPR ea}, the bandwidth utilization data sequence is {BUR1, BUR2, …, BUR ea}, where ea is the number of data points collected;

[0316] Assume that the current link is link A, and its adjacent link set is {BA1, BA2, ..., BA mr ,}, where BA ivc represents the ivcth adjacent link, ivc = 1, 2, ..., mr;

[0317] S32: For link A and a certain adjacent link BA ivz A quality indicator XUQ, where XUQ represents any one of LPR, DPR and BUR, and the measured values ​​at tn time points are:

[0318] QUR A ={XUQ A1 ,XUQ A2 ,…,XUQ Atn};

[0319]

[0320] In the formula, Indicates the adjacent link BA ivz The measured value of the quality indicator XUQ at tn time points, XUQ A represents the measurement value sequence of the quality indicator XUQ of link A at tn time points, Indicates the adjacent link BA ivz The measurement value sequence of the quality indicator XUQ at tn time points, XUQ Atn represents the measured value of the quality indicator XUQ of link A at tn time points;

[0321] S33: Calculate the spatial correlation coefficient CHE, and the calculation formula is as follows:

[0322]

[0323] In the formula, XUQ Ajxb represents the measured value of the quality indicator XUQ of link A at the jxbth time point, represents the average value of the quality indicator XUQ of link A, Indicates the adjacent link BA ivz The measured value of the quality indicator XUQ at the jxbth time point, Indicates the adjacent link BA ivz The average value of the quality indicator XUQ measurement value;

[0324] S34: Obtain the link packet loss rate spatial correlation coefficient CHE respectively through the calculation formulas of steps S32 to S33 LPR , delay spatial correlation coefficient CHE DPR and bandwidth utilization spatial correlation coefficient CHE BUR ;

[0325] S35: Calculate the comprehensive spatial correlation coefficient CHE zz , the calculation formula is as follows:

[0326]

[0327] Wherein, wju, wju2 and wju3 represent the corresponding CHE LPR CHE DPR and CHE BUR The weight coefficient of .

[0328] As a further step of the present invention, regarding S3, analyzing the time correlation coefficient of the link quality data at different times, the analysis steps are as follows:

[0329] S36: Set the time series data of the link quality indicator YBB to:

[0330] {YBB1, YBB2, ..., YBB ncb};

[0331] Where ncb represents the total number of link quality indicator time series data points, YBB represents any one of LPR, DPR and BUR, and the lag order is ksb, ksb<ncb;

[0332] S37: Calculate the time correlation coefficient ACF(ksb), the calculation formula is as follows:

[0333]

[0334] Where, YBB tnrepresents the measured value of the link quality indicator YBB at the tnth time point, Indicates the average value of the link quality indicator YBB measurement, YBB tn+ksb represents the measured value of the link quality indicator YBB at the tn+ksb-th time point;

[0335] S38: The calculation formulas of the above steps S36 to S37 are respectively obtained:

[0336] Link packet loss rate time correlation coefficient ACF LPR (ksb), delay time correlation coefficient ACF DPR (ksb) and bandwidth utilization time correlation coefficient ACF BUR (ksb);

[0337] S39: Calculate the comprehensive time correlation coefficient ACF tr , the calculation formula is as follows:

[0338]

[0339] Where wxx DPR 、wxx DPR and wxx BUR Respectively represent the corresponding ACF LPR (ksb), ACF DPR (ksb) and ACF BUR The weight coefficient of (ksb);

[0340] S40: Assign a comprehensive weight to the link. The assignment process is as follows:

[0341] Set the packet loss rate of the current link to LPR tnew , the delay is DPR tnew , bandwidth utilization BUR tnew , calculate the link weight WLL, the calculation formula is as follows:

[0342]

[0343] Where wll LPR 、wll DPR and wll BUR They are the weight coefficients of LPR, DPR and BUR in link calculation respectively.

[0344] As a further feature of the present invention, in S4, the optimal transmission path is intelligently selected according to the priority of the data and the current network status, and the selection process is as follows:

[0345] Through the intelligent sensing node to continuously monitor the link status, for any link loe, calculate the time window TYYkjz Packet loss rate LPR loe , the calculation formula is as follows:

[0346]

[0347] In the formula, NNQ lost Indicates the number of lost packets, NNQ tatal Indicates the total number of packets sent;

[0348] For any link loe, calculate the delay change rate ΔDPR loe , the calculation formula is as follows:

[0349]

[0350] In the formula, DPR curre Indicates the average delay at the current moment, DPR prev represents the average delay of the previous moment;

[0351] For any link loe, calculate the bandwidth utilization change rate ΔBUR loe , the calculation formula is as follows:

[0352] ΔBUR loe =BUR curre -BUR prev ;

[0353] Where BUR curre Indicates the current bandwidth utilization, BUR prev Indicates the bandwidth utilization at the previous moment;

[0354] Assume the fault judgment function of any link is FGG loe , if any one of the conditions 3 is met, then:

[0355] FGG loe =1, it is determined that any link loe fails, triggering the path reselection mechanism;

[0356] Condition 3 is as follows:

[0357] LPR loe >LPR thred ;

[0358] ΔDPR loe >ΔDPR thred ;

[0359] ΔBUR loe >ΔBUR thred ;

[0360] If any one of the conditions 4 is met, then:

[0361] FGG loe =0, it is determined that no link loe fails;

[0362] Condition 4 is as follows:

[0363] LPR loe ≤LPR thred ;

[0364] ΔDPR loe ≤ΔDPR thred ;

[0365] ΔBUR loe ≤ΔBUR thred ;

[0366] Among them, LPR thred , ΔDPR thred and ΔBUR thred They represent the packet loss rate fault threshold, delay change rate threshold, and bandwidth utilization change rate threshold respectively.

[0367] As a further feature of the present invention, regarding triggering the path reselection mechanism in S4, the triggering steps are as follows:

[0368] S41: Set the reward function to RNV, and calculate the reward function RNV. The calculation formula is as follows:

[0369]

[0370] In the formula, RYC delay represents the reward function related to transmission delay, DPR bw represents the reward function related to link bandwidth utilization, PC prooc represents the reward function related to data priority, RYX reliby represents the reward function related to link reliability, wsa1, wsa2, wsa3 and wsa4 correspond to RYC respectively. delay 、DPR bw 、PC prooc and RYX reliby The weight coefficient of

[0371] S42: RYC delay The calculation process is as follows:

[0372] When delay≤delay trld ,but:

[0373] RYC delay =kbiq1×(delay trld -delay);

[0374] When delay>delay trld , then RYC delay =0;

[0375] Among them, delay represents the actual data transmission delay time, delay trld represents the transmission delay threshold, kbiq1 represents the proportional coefficient of transmission delay;

[0376] S43: DPR bw The calculation formula is as follows:

[0377] DPR bw =kbiq2×bw utizon ;

[0378] In the formula, kbiq2 represents the proportional coefficient of bandwidth utilization, bw utizon Indicates the bandwidth utilization of the current link;

[0379] S44: For PC prooc The calculation formula is as follows:

[0380] PC prooc =base+kbiq3×prooc pri ;

[0381] In the formula, base represents the basic reward value set according to data of different priorities, kbiq3 represents the coefficient for adjusting the influence of transmission performance on the reward, and prooc pri Indicators related to data transmission performance;

[0382] S45: RYX reliby The calculation formula is as follows:

[0383] RYX reliby =kbiq4×(1-faire)×qucbbzu;

[0384] Where kbiq4 represents the proportional coefficient of link reliability, which is used to adjust the magnitude of link reliability reward, faire represents the historical failure frequency of the link, ranging from 0 to 1, and qucbbzu represents the current quality status indicator of the link, ranging from 0 to 1;

[0385] S46: When any link loe fails and the path reselection mechanism is triggered, the reward function RNV is adjusted and a new reward function RNV is calculated. new , the calculation process is as follows:

[0386]

[0387] In the formula, RYCnewdelay represents the adjusted reward function related to transmission delay, RYX newreliby represents the adjusted reward function related to link reliability, wsa new1 and wsa new4 Corresponding to RYC newdelay and RYX newreliby Weight coefficient;

[0388] S47: RYC newdelay The calculation process is as follows:

[0389] When delay≤delay newtrld ,but:

[0390] RYC newdelay =kbiq new1 ×(delay newtrld -delay);

[0391] When delay>delay newtrld , then RYC newdelay =0;

[0392] Where delay newtrld Indicates the adjusted transmission delay threshold, kbiq new1 represents the proportionality factor of the adjusted transmission delay;

[0393] S48: RYX newreliby The calculation formula is as follows:

[0394] RYX newreliby =kbiq new4 ×(1-faire)×qucbbzu;

[0395] In the formula, kbiq new4 The proportionality factor representing the adjusted link reliability;

[0396] S49: Set that there are nkaa candidate paths;

[0397] For each candidate path set {HxJJ1, HxJJ2, …, HxJJ iaa}, where iaa = 1, 2, ..., nkaa. When the path reselection mechanism is triggered, the new reward function value RNV corresponding to each candidate path is calculated through steps S46 to S48. niccq ;

[0398] Where inccq = 1, 2, ..., nkaa;

[0399] S50: Compare the new reward values ​​of all candidate paths, and obtain the candidate path with the largest reward value. The comparison formula is as follows:

[0400]

[0401] In the formula, HxJJ znew Represents the new optimal transmission path, HxJJ nkaa In the process of path reselection, it refers to the set of all candidate paths {HxJJ1, HxJJ2, …, HxJJ iaa}, the candidate path with the largest reward value, max{RNV1, RNV2, …, RNV nakk ,} represents the reward value set {RNV1, RNV2, …, RNV nakk ,}, where represents the conditional qualifier, used to accurately describe HxJJ znew is the path that satisfies the specific reward value condition, RNV nkaa Indicates the reward value corresponding to the nkaa-th candidate path;

[0402] S51: Determine the new optimal transmission path HxJJ znew After that, data transmission is switched from the faulty path to the optimal transmission path HxJJ znew superior.

[0403] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A data transmission method based on a TSN edge computing system, characterized in that: The following steps are involved: S1: Use machine learning algorithms to perform multi-dimensional feature analysis on the data entering the TSN edge computing system to obtain analysis results. Based on the analysis results, use the DBSCAN algorithm to preliminarily divide the data categories according to the multi-dimensional feature vectors. When new data flows in, dynamically update the cluster center and boundaries according to the distance between the new data and the category center and the surrounding density, and add or reduce categories in a timely manner so that the categories can evolve adaptively with the data features. Dynamically divide the data into multiple categories, establish a priority evaluation model, and assign dynamically changing priorities to each category, thereby obtaining data category priorities; S2: In the TSN edge computing system, a time series prediction model is introduced to predict the network traffic, node load, and bandwidth change trend in the future, and obtain the prediction results. Based on the data category priority in step S1 and the prediction results, a dynamic programming algorithm is used to carry out network resource allocation to obtain the network resource allocation results. S3: Deploy intelligent sensing nodes in the TSN edge computing system. The nodes are equipped with a feedback mechanism based on link quality change monitoring. First, monitor the link packet loss rate, delay, bandwidth utilization and other indicators at the initial sampling frequency, adjust the sampling frequency according to the fluctuation of link quality indicators, collect link quality data at different times, analyze the spatial correlation coefficient and time correlation coefficient of link quality data at different times, and use weighted edges to represent links when constructing the network topology. The weight integrates the current quality data with the spatial correlation coefficient and the time correlation coefficient to give the link a comprehensive weight. S4: Using reinforcement learning algorithm, intelligently select the optimal transmission path according to the data priority and current network status. During the transmission process, if the sensing node detects that the link status has deteriorated or a failure has occurred, the path reselection mechanism is immediately triggered. The reinforcement learning algorithm quickly determines the new optimal path according to the new network status to ensure the continuity and efficiency of data transmission.

2. According to a data transmission method based on a TSN edge computing system according to claim 1, it is characterized in that: In S1, a machine learning algorithm is used to perform multi-dimensional feature analysis on the data entering the TSN edge computing system to obtain the analysis results. The analysis steps are as follows: S11: Set the data point set entering the TSN edge computing system to be X = {x1, x2, ..., x n }, where each data point x i Represents a two-dimensional vector, x i1 and x i2 Represents the data point x i The eigenvalues ​​in two different dimensions, x i =(x i1 , x i2 ), i = 1, 2, ..., n, n is the total number of data points; S12: For two-dimensional data point x i =(x i1 , x i2 ) is standardized so that the two-dimensional data point x i Convert to a distribution with mean 0 and variance 1: In the formula, and Represents the data point x i The mean in two different dimensions, and Represents the data point x i variance in two different dimensions; S13: Calculate the standardized x i1b and x i2b , the calculation formula is: In the formula, s1 and s2 are and The square root of is the standard deviation; S14: Reference the second data point x j =(x j1 , x j2 ), indicating x j The values ​​in two dimensions for two data points x i =(x i1 , x i2 ) and x j =(x j1 , x j2 ), the Euclidean distance formula is used to calculate the distance between two data points. The Euclidean distance formula is as follows: In the formula, d(x i， x j ) represents the data point x i and x j The distance between S15: Traverse each data point x in the data set i , the number of data points in the neighborhood with ∈ as radius is calculated by the density formula, ∈ represents the neighborhood range of the data point, and the density formula is as follows: N ∈ (x i )={x j ∈X|d(x i ,x j )≤∈}; Where N ∈ (x i ) represents the data point x i The set of data points in the ∈-neighborhood of; S16: Statistical data point set N ∈ (x i ) to get the number of elements in |N ∈ (x i )|; S17: Set the integer parameter to Min, and |N ∈ (x i )|Compare with: When |N ∈ (x i )|≥Min, then x i Marked as core points; When |N ∈ (x i )|<Min, then x i are marked as boundary points; S18: If the data point x i If it is neither a core point nor a boundary point, it is marked as a noise point; S19: Starting from any core point that has not been visited, put it into a new cluster C k , initially k = 1, k represents the total number of clusters, and then all core points in all neighborhoods and boundary points falling in the neighborhood of these core points are added to cluster C k In this way, a cluster is formed; S20: Repeat steps S11 to S19 until all core points have been visited, and finally all data points are divided into different clusters, completing the preliminary data category division.

3. According to a data transmission method based on a TSN edge computing system according to claim 2, it is characterized in that: Regarding S1, when new data flows in, the cluster center and boundary are dynamically updated according to the distance between the new data and the category center and the surrounding density. The updating steps are as follows: When a new data point x new When inflow; Set the zth cluster C z The center μ z =(μ z1 , μ z2 ), z = 1, 2, ..., k, for μ z1 and μ z2 The calculation formula is as follows: In the formula, |C z | represents cluster C z The number of data points in z1 and μ z2 Represents cluster C z The coordinate values ​​of the center in the first and second dimensions; Calculate the new data point x by the density calculation formula in step S15 new The number of data points in the neighborhood with ∈ as radius is obtained. ∈ (x new )|; When |N ∈ (x new )|≥Min, then x new Marked as core points; When |N ∈ (x new )|<Min, then x new are marked as boundary points; Set a priority function P(C z ) to form each cluster C z Assign priority and calculate cluster C z The proportion of the number of data points to the total number of data points is P1(C z ), the calculation formula is as follows: Set each cluster C z The number of times new data flows in per unit time is denoted as f z , calculate the cluster C z The data update frequency accounts for the proportion of all cluster data update frequencies P2(C z ), the calculation formula is as follows: In the formula, f k Represents cluster C z The number of new data flows per unit time, used to measure the clustering C z Data update speed; For the priority function P(C z ) is calculated, and the calculation formula is as follows: P(C z )=w1P1(C z )+w2P2(C z ); In the formula, w1 and w2 represent the corresponding 1(C z ) and P2(C z )’s weight, and w1+w2=1.

4. According to a data transmission method based on a TSN edge computing system according to claim 1, it is characterized in that: Regarding S2, in the TSN edge computing system, a time series prediction model is introduced to predict the network traffic, node load, and bandwidth change trend in the future, and obtain the prediction results. The prediction process is as follows: Assume that the traffic time series is T(t), the node load time series is L(Δt), and the bandwidth time series is B(t), where t represents the time step. The moving average prediction model is used to predict the traffic change trend. The traffic prediction formula is as follows: Where kz represents the predicted time step index, kz = 1, 2, ..., nu, nu represents the prediction step, mu represents the moving average window size, T cy (t+kz) represents the predicted flow value, and io represents the summation index variable, which is used in the summation operation. It is used to traverse the time steps from t-mu+1 to t to obtain the flow data T(io) corresponding to these time steps; The node load value is predicted, and the node load value prediction formula is as follows: Where, L cy (t+kz) represents the predicted node load value, and L(io) represents the summation of io. The time steps from t-mu+1 to t are traversed in the figure, and the node load data corresponding to these time steps; The bandwidth value is predicted. The bandwidth value prediction formula is as follows: In the formula, B(io) represents the summation operation. The bandwidth data corresponding to these time steps are traversed from t-mu+1 to t, B cy (t+kz) represents the predicted bandwidth value; Set the high priority data transmission interruption risk function to R h (t), the network congestion risk function is R c (t), for R h (t) is calculated using the following formula: Where fp h represents the empirical function, P h (t) indicates the current transmission rate and bandwidth occupancy of high-priority data; R c (t) is calculated using the following formula: R c (t)=fp h (T cy (t+kz),L cy (t+kz),B cy (t+kz)); Set the bandwidth required for high priority data to BW h (t), the cache resource is CH h (t), the redundant bandwidth ratio is α h , the redundant cache ratio is β h ; For high-priority data resources, the allocation formula is as follows: When R h (t)>TU h And R c (t)>CH h (t), then: BW h (t)=BW req (t)+α h ×BW ectra (t); When R h (t)≤TU h And R c (t)≤CH h (t), then: BW h (t)=BW req (t); In the formula, TU h Indicates the interruption risk threshold of high priority data, BW req (t) represents the basic bandwidth requirement of the current high-priority data, BW ectra (t) represents the additional bandwidth resources, α h Indicates the redundant bandwidth ratio of high priority data; When R h (t)>TU h And R c (t)>CH h (t), then: CH h (t)=CH req (t)+β h ×CH ectra (t); When R h (t)≤TU h And R c (t)≤CH h (t), then: CH h (t)=CH req (t); In the formula, CH req (t) represents the basic cache requirement, CH ectra (t) represents additional cache resources, β h Indicates the redundant cache ratio of high-priority data; For medium priority data resources, set the required bandwidth to BW. m (t), the dynamic adjustment factor is γ m , the resource allocation is AW m (t), the allocation formula for medium priority data resources is as follows: When R h (t)>CH m ,but: BW m (t)=AW m (t)+γ m ×ΔAW m (t); When R h (t)≤CH m ,but: BW m (t)=AW m (t); Where, ΔAW m (t) represents the resource adjustment amount, CH m Indicates the impact limit of the network congestion risk threshold on medium priority data; For low priority data, set its occupied bandwidth to BW l (t), the low priority data resource allocation formula is as follows: When R h (t)>TU h And R c (t)>CW l (t), then: BW l (t)=0; When R h (t)≤TU h And R c (t)≤CW l (t), then: BW l (t)=BW l (t); In the formula, CW l (t) represents the impact limit of the network congestion risk threshold on low-priority data; In the TSN edge computing system, when faced with resource constraints and high risks, low-priority data is reasonably transferred to ensure efficient and stable allocation of resources across the entire network.

5. A data transmission method based on a TSN edge computing system according to claim 4, characterized in that: In the TSN edge computing system, when faced with resource constraints and high risks, low-priority data is reasonably transferred to ensure efficient and stable allocation of resources across the entire network. The transfer process is as follows: Set the current path of low priority data to P l (t); The set of alternate transmission paths is {P s1 , P s2 ,…,P sv }, sv represents the total number of paths, P is represents the is-th backup transmission path, is = 1, 2, ..., sv, and the transfer threshold is TU tq ; When R c (t)>TU tq And R h (t)>TU tq , and after satisfying the following condition 1, then: ON new (t)=P is ; Where, PE new (t) represents the new transmission path of low priority data after the routing strategy is selected; Condition 1 is as follows: In the formula, Cost(P is ) represents the backup transmission path P is Cost (P js ) represents the measure of the backup transmission path P js The cost function, P js represents the set of alternate transmission paths {P s1 , P s2 ,…,P sv } any alternate route, js represents the index variable; When R c (t)≤TU tq And R h (t)≤TU tq , and after condition 2 is met, then: ON new (t)=P l (t); Where P l (t) represents the transmission path used by low-priority data before the routing strategy is selected; Condition 2 is as follows: Cost(P js ) is calculated, and the calculation formula is as follows: In the formula, DR js Indicates the backup path P js Delay, LR js Indicates the backup path P js Load condition, BR js Indicates the backup path P js bandwidth availability, wn1, wn2 and wn3 represent the bandwidth availability of DR js , L.R. js and BR js The weight coefficient of .

6. A data transmission method based on a TSN edge computing system according to claim 1, characterized in that: Regarding S3, the sampling frequency is adjusted according to the fluctuation of the link quality indicator. The adjustment steps are as follows: Set the initial sampling frequency of the intelligent sensing node to fa int ,The link packet loss rate is LPR, where 0≤LPR≤1, the link delay is DPR, and the link bandwidth utilization rate is BUR, where 0≤BUR≤1; The sampling frequency formula is as follows: In the formula, fa new represents the new sampling frequency adjusted according to the fluctuation of the link quality indicator, αz represents the adjustment coefficient, αz is a real number greater than 0, β LPR , β DPR and β BUR They represent the weight coefficients corresponding to LPR, DPR and BUR, and β LPR +β DPR +β BUR =1, LPR avg 、DPR avg and BUR avg They represent the historical averages of LPR, DPR and BUR respectively. max 、DPR max and BUR max They represent the historical maximum values ​​of LPR, DPR and BUR respectively.

7. A data transmission method based on a TSN edge computing system according to claim 6, characterized in that: Regarding S3, the analysis steps for analyzing the correlation of time series change trends of link quality data at different times are as follows: S31: The packet loss rate data sequence collected by the intelligent sensing node is determined as: {LPR1,LPR2,…,LPR ea }; Set the delayed data sequence to {DPR1, DPR2, ..., DPR ea }, the bandwidth utilization data sequence is {BUR1, BUR2, ..., BUR ea }, where ea is the number of data points collected; Assume that the current link is link A, and its adjacent link set is {BA1, BA2, ..., BA mr ,}, where BA ivc represents the ivcth adjacent link, ivc = 1, 2, ..., mr; S32: For link A and a certain adjacent link BA ivz A quality indicator XUQ, where XUQ represents any one of LPR, DPR and BUR, and the measured values ​​at tn time points are: HONOUR A ={HQ A1 ,HQ A2 ,…,HQ Atn }; In the formula, Indicates the adjacent link BA ivz The measured value of the quality indicator XUQ at tn time points, XUQ A represents the measurement value sequence of the quality indicator XUQ of link A at tn time points, Indicates the adjacent link BA ivz The measurement value sequence of the quality indicator XUQ at tn time points, XUQ Atn represents the measured value of the quality indicator XUQ of link A at tn time points; S33: Calculate the spatial correlation coefficient CHE, and the calculation formula is as follows: In the formula, XUQ Ajxb represents the measured value of the quality indicator XUQ of link A at the jxbth time point, represents the average value of the quality indicator XUQ of link A, Indicates the adjacent link BA ivz The measured value of the quality indicator XUQ at the jxbth time point, Indicates the adjacent link BA ivz The average value of the quality indicator XUQ measurement value; S34: Obtain the link packet loss rate spatial correlation coefficient CHE respectively through the calculation formulas of steps S32 to S33 LPR , delay spatial correlation coefficient CHE DPR and bandwidth utilization spatial correlation coefficient CHE BUR ; S35: Calculate the comprehensive spatial correlation coefficient CHE zz , the calculation formula is as follows: Wherein, wju, wju2 and wju3 represent the corresponding CHE LPR CHE DPR and CHE BUR The weight coefficient of .

8. A data transmission method based on a TSN edge computing system according to claim 7, characterized in that: Regarding S3, the time correlation coefficient of the link quality data at different times is analyzed. The analysis steps are as follows: S36: Set the time series data of the link quality indicator YBB to: {YBB1,YBB2,…,YBB ncb }; Where ncb represents the total number of link quality indicator time series data points, YBB represents any one of LPR, DPR and BUR, and the lag order is ksb, ksb<ncb; S37: Calculate the time correlation coefficient ACF(ksb), the calculation formula is as follows: Where, YBB tn represents the measured value of the link quality indicator YBB at the tnth time point, Indicates the average value of the link quality indicator YBB measurement, YBB tn+ksb represents the measured value of the link quality indicator YBB at the tn+ksb-th time point; S38: The calculation formulas of the above steps S36 to S37 are respectively obtained: Link packet loss rate time correlation coefficient ACF LPR (ksb), delay time correlation coefficient ACF DPR (ksb) and bandwidth utilization time correlation coefficient ACF BUR (ksb); S39: Calculate the comprehensive time correlation coefficient ACF tr , the calculation formula is as follows: Where wxx DPR 、wxx DPR and wxx BUR Respectively represent the corresponding ACF LPR (ksb), ACF DPR (ksb) and ACF BUR The weight coefficient of (ksb); S40: Assign a comprehensive weight to the link. The assignment process is as follows: Set the packet loss rate of the current link to LPR tnew , the delay is DPR tnew , bandwidth utilization BUR tnew , calculate the link weight WLL, the calculation formula is as follows: Where wll LPR 、wll DPR and wll BUR They are the weight coefficients of LPR, DPR and BUR in link calculation respectively.

9. A data transmission method based on a TSN edge computing system according to claim 8, characterized in that: In S4, the optimal transmission path is intelligently selected based on the data priority and current network status. The selection process is as follows: Through the intelligent sensing node to continuously monitor the link status, for any link loe, calculate the time window TYY kjz Packet loss rate LPR loe , the calculation formula is as follows: In the formula, NNQ lost Indicates the number of lost packets, NNQ tatal Indicates the total number of packets sent; For any link loe, calculate the delay change rate ΔDPR loe , the calculation formula is as follows: In the formula, DPR curre Indicates the average delay at the current time, DPR prev represents the average delay of the previous moment; For any link loe, calculate the bandwidth utilization change rate ΔBUR loe , the calculation formula is as follows: ΔBUR loe =BUR curre -BUR prev ; Where BUR curre Indicates the current bandwidth utilization, BUR prev Indicates the bandwidth utilization at the previous moment; Assume the fault judgment function of any link is FGG loe , if any one of the conditions 3 is met, then: FGG loe =1, it is determined that any link loe fails, triggering the path reselection mechanism; Condition 3 is as follows: LPR loe >LPR thred ; ΔDPR loe >ΔDPR thred ; ΔBUR loe >ΔBUR thred ; If any one of the conditions 4 is met, then: FGG loe =0, it is determined that no link loe fails; Condition 4 is as follows: LPR loe ≤LPR thred ; ΔDPR loe ≤ΔDPR thred ; ΔBUR loe ≤ΔBUR thred ; Among them, LPR thred , ΔDPR thred and ΔBUR thred They represent the packet loss rate fault threshold, delay change rate threshold, and bandwidth utilization change rate threshold respectively.

10. A data transmission method based on a TSN edge computing system according to claim 9, characterized in that: Regarding triggering the path reselection mechanism in S4, the triggering steps are as follows: S41: Set the reward function to RNV, and calculate the reward function RNV. The calculation formula is as follows: In the formula, RYC delay represents the reward function related to transmission delay, DPR bw represents the reward function related to link bandwidth utilization, PC prooc represents the reward function related to data priority, RYX reliby represents the reward function related to link reliability, wsa1, wsa2, wsa3 and wsa4 correspond to RYC respectively. delay 、DPR bw 、PC prooc and RYX reliby The weight coefficient of S42: RYC delay The calculation process is as follows: When delay≤delay trld ,but: RYC delay =kbiq1×(delay trld -delay); When delay>delay trld , then RYC delay =0; Among them, delay represents the actual data transmission delay time, delay trld represents the transmission delay threshold, kbiq1 represents the proportional coefficient of transmission delay; S43: DPR bw The calculation formula is as follows: DPR bw =kbiq2×bw utizon ; In the formula, kbiq2 represents the proportional coefficient of bandwidth utilization, bw utizon Indicates the bandwidth utilization of the current link; S44: For PC prooc The calculation formula is as follows: PC prooc =base+kbiq3×prooc pri ; In the formula, base represents the basic reward value set according to data of different priorities, kbiq3 represents the coefficient for adjusting the influence of transmission performance on the reward, and prooc pri Indicators related to data transmission performance; S45: RYX reliby The calculation formula is as follows: RYX reliby =kbiq4×(1-do)×qucbbzu; Where kbiq4 represents the proportional coefficient of link reliability, which is used to adjust the magnitude of link reliability reward, faire represents the historical failure frequency of the link, ranging from 0 to 1, and qucbbzu represents the current quality status indicator of the link, ranging from 0 to 1; S46: When any link loe fails and the path reselection mechanism is triggered, the reward function RNV is adjusted and a new reward function RNV is calculated. new , the calculation process is as follows: In the formula, RYC newdelay represents the adjusted reward function related to transmission delay, RYX newreliby represents the adjusted reward function related to link reliability, wsa new1 and wsa new4 Corresponding to RYC newdelay and RYX newreliby Weight coefficient; S47: RYC newdelay The calculation process is as follows: When delay≤delay newtrld ,but: RYC newdelay =kbiq new1 ×(delay newtrld -delay); When delay>delay newtrld , then RYC newdelay =0; Where delay newtrld Indicates the adjusted transmission delay threshold, kbiq new1 represents the proportionality factor of the adjusted transmission delay; S48: RYX newreliby The calculation formula is as follows: RYX newreliby =kbiq new4 ×(1-do)×qucbbzu; In the formula, kbiq new4 The proportionality factor representing the adjusted link reliability; S49: Set that there are nkaa candidate paths; For each candidate path set {HxJJ1, HxJJ2, …, HxJJ iaa }, where iaa = 1, 2, ..., nkaa. When the path reselection mechanism is triggered, the new reward function value RNV corresponding to each candidate path is calculated through steps S46 to S48. niccq ; Where inccq = 1, 2, ..., nkaa; S50: Compare the new reward values ​​of all candidate paths, and obtain the candidate path with the largest reward value. The comparison formula is as follows: In the formula, HxJJ znew Represents the new optimal transmission path, HxJJ nkaa In the process of path reselection, it refers to the set of all candidate paths {HxJJ1, HxJJ2, …, HxJJ iaa }, the candidate path with the largest reward value, max{RNV1, RNV2, …, RNV nakk ,} represents the reward value set {RNV1, RNV2, …, RNV nakk ,}, where represents the conditional qualifier, used to accurately describe HxJJ znew is the path that satisfies the specific reward value condition, RNV nkaa Indicates the reward value corresponding to the nkaa-th candidate path; S51: Determine the new optimal transmission path HxJJ znew After that, data transmission is switched from the faulty path to the optimal transmission path HxJJ znew superior.

Images