A cross-domain end-to-end data transmission pressure recognition method and device

Abstract

The present invention relates to the field of communication technology, and specifically discloses a cross-domain end-to-end data transmission pressure recognition method and device. The method includes segmenting the historical business flow based on the given business data transmission demand, determining the fixed time slot business flow, performing a one-dimensional discrete wavelet transform, obtaining the original scale coefficient, using a multi-fractal wavelet model and an optimized decomposition series, performing coefficient decomposition, obtaining an optimized decomposition coefficient sequence, determining the transmission pressure characteristic value according to the average transmission delay and the rated transmission delay, splicing the fixed time slot business flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value, obtaining the transmission pressure characteristic signal sequence, and then dividing it according to proportion to obtain a training set and a test set, using the training set to train the LSTM neural network model, and then transferring the test set into the trained LSTM neural network model to obtain the prediction result. The purpose is to ensure the security and efficiency of end-to-end data transmission through pressure transmission evaluation.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and in particular to a method and device for recognizing cross-domain end-to-end data transmission pressure. Background Art

[0002] In the new power system where the proportion of new energy is gradually increasing, the high proportion of new energy access and large-scale application of power electronic equipment make the system more complex, and the operation of the power grid faces challenges such as increased uncertainty in power and electricity balance and decreased system inertia. Power companies need to take into account the development of high-proportion new energy and the safe and stable operation of the power system, and accelerate the digital upgrade of the power system to effectively support the integrated development of source, grid, load and storage. Therefore, it is necessary to build a stronger, more flexible, and multi-service-carrying power communication network to achieve comprehensive perception of new energy and flexible loads, and effectively support production operations, operation management, and operational control on the basis of "observable, measurable, adjustable, controllable, and traceable", laying a solid foundation for the safe and stable operation of the power system.

[0003] With the rapid development of automation and communication technologies, the business traffic of the power backbone communication transmission network has shown explosive growth, and various problems such as network congestion have emerged. Summary of the invention

[0004] In view of the above problems, an object of the present invention is to provide a cross-domain end-to-end data transmission pressure recognition method.

[0005] The second object of the present invention is to provide a cross-domain end-to-end data transmission pressure recognition device.

[0006] The first technical solution adopted by the present invention is: S1, based on a given service data transmission requirement, dividing the historical service flow and determining a fixed time slot service flow;

[0007] S2, according to the fixed time slot service flow, perform one-dimensional discrete wavelet transform to obtain the original scale coefficient;

[0008] S3, according to the original scale coefficient, using a multi-fractal wavelet model and an optimized decomposition series, performing coefficient decomposition to obtain an optimized decomposition coefficient sequence;

[0009] S4, determining the transmission pressure characteristic value according to the relationship between the average transmission delay of each data packet in the time slot corresponding to the fixed time slot service flow and the rated transmission delay;

[0010] S5, splicing the fixed time slot service flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value to obtain a transmission pressure characteristic signal sequence, and then dividing it according to proportion to obtain a training set and a test set of a neural network;

[0011] S6. After training the LSTM neural network model using the training set, the test set is passed into the trained LSTM neural network model to obtain a prediction result.

[0012] Optionally, the S1 includes:

[0013] The historical business flow is:

[0014] X=[x(1),x(2)…x(T)]

[0015] The service flow of the fixed time slot is:

[0016] X t =[x(t),x(t+1)…x(t+l-1)]

[0017] Where X is a transmission pressure data stream with a type characteristic containing T data, l is the length of the fixed time slot service flow, and x(t) is the data packet successfully sent in the tth time slot.

[0018] Optionally, the calculation formula of the original scale coefficient is:

[0019] u 0,k =∫x(t)φ 0,k (t)dt

[0020] φ 0,k (t) = ε(tk) - ε(tk-1)

[0021] In the formula, u 0,k is the original scale factor, φ 0,k (t) is the scaling function, and ε(tk) is the step function shifted by k units.

[0022] Optionally, the S3 includes:

[0023] S31, select Haar wavelet, perform multi-fractal wavelet model decomposition on the original scale coefficients, and obtain an optimized coefficient set. The process is as follows:

[0024]

[0025] In the formula, u j,2k is the 2kth decomposition coefficient in the j-level decomposition; A j-1,k The random variable decomposed at level m-1 approximately follows Gaussian distribution, H is the Hurst parameter of the historical business flow, p is the self-similar variable of the random variable, and var is the variance;

[0026] S32. According to the multi-fractal wavelet model and the optimal decomposition level, the optimized decomposition coefficient sequence is determined as follows:

[0027]

[0028] Where M0 is the optimal decomposition level.

[0029] Optionally, the transmission pressure characteristic value is calculated as follows:

[0030]

[0031] Where P t is the transmission pressure evaluation value, The rated transmission delay of the service flow in a fixed time slot determined by the service communication quality requirement, S t is the average delay of all data packets transmitted in a fixed time slot;

[0032] The average delay calculation formula for all data packets transmitted within a fixed time slot is as follows:

[0033]

[0034] Where s(t) is the actual transmission delay of the successfully sent data packet at time t.

[0035] Optionally, the S5 includes:

[0036] S51, concatenating the fixed time slot service flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value, as shown in the following formula:

[0037] Y t =[X t ,U t ,P t ]

[0038] Where Y t For X t The transmission pressure characteristic signal sequence;

[0039] S52: Determine a modeling data set based on the transmission pressure characteristic signal sequence:

[0040] Y=[Y1,Y2,...,Y T-l ];

[0041] S53, dividing the modeling data set into a training set and a test set, as shown in the following formula:

[0042]

[0043] In the formula, σ is the division ratio, E is the training set, and C is the test set.

[0044] Optionally, the S6 includes:

[0045] S61, inputting the training set into the LSTM neural network model to perform feature extraction to obtain temporal abstract features;

[0046] S62, based on the time series abstract features, with the goal of minimizing the loss function, using the root mean square error formula to perform error calculation to obtain the trained LSTM neural network model;

[0047] S63: Input the test set into the trained LSTM neural network model to obtain a prediction result.

[0048] The second technical solution adopted by the present invention is: a cross-domain end-to-end data transmission pressure recognition device, comprising: a service flow segmentation module, used to segment the historical service flow based on the given service data transmission demand, and determine the fixed time slot service flow;

[0049] A wavelet transform module, used for performing a one-dimensional discrete wavelet transform according to the fixed time slot service flow to obtain an original scale coefficient;

[0050] A coefficient decomposition module is used to perform coefficient decomposition according to the original scale coefficients by using a multi-fractal wavelet model and an optimized decomposition series to obtain an optimized decomposition coefficient sequence;

[0051] A characteristic value determination module, used to determine a transmission pressure characteristic value according to an average transmission delay and a rated transmission delay of each data packet in a time slot corresponding to the fixed time slot service flow;

[0052] A signal division module, used for splicing the fixed time slot service flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value to obtain a transmission pressure characteristic signal sequence and then dividing it according to proportion to obtain a training set and a test set of a neural network;

[0053] The model training module uses the training set to train the LSTM neural network model, and then transfers the test set to the trained LSTM neural network model to obtain a prediction result.

[0054] Optionally, the calculation formula of the original scale coefficient is:

[0055] u 0，k =∫x(t)φ 0,k (t)dt

[0056] φ 0,k (t) = ε(tk) - ε(tk-1)

[0057] In the formula, u 0,k is the original scale factor, φ 0,k (t) is the scaling function, and ε(tk) is the step function shifted by k units.

[0058] Optionally, the coefficient decomposition module includes:

[0059] The wavelet decomposition submodule is used to select Haar wavelet, perform multi-fractal wavelet model decomposition on the original scale coefficients, and obtain an optimized coefficient set. The process is as follows:

[0060]

[0061]

[0062] In the formula, u j,2k is the 2kth decomposition coefficient in the j-level decomposition; A j-1,k The random variable decomposed at level m-1 approximately follows Gaussian distribution, H is the Hurst parameter of the data stream of historical information, p is the self-similar variable of the random variable, and var is the variance;

[0063] The series decomposition submodule is used to determine the optimized decomposition coefficient sequence according to the multi-fractal wavelet model and the optimal decomposition series, as shown in the following formula:

[0064]

[0065] Where M0 is the optimal decomposition level.

[0066] Beneficial effects of the above technical solution:

[0067] The present invention is based on a given business data transmission demand, divides the historical business flow, determines the fixed time slot business flow, performs a one-dimensional discrete wavelet transform according to the fixed time slot business flow data, obtains the original scale coefficient, and performs coefficient decomposition according to the original scale coefficient using a multi-fractal wavelet model and an optimized decomposition series to obtain an optimized decomposition coefficient sequence, determines the transmission pressure characteristic value according to the relationship between the average transmission delay of each data packet in the corresponding time slot of the fixed time slot business flow and the rated transmission delay, splices the fixed time slot business flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value, and divides the transmission pressure characteristic signal sequence according to the proportion to obtain the training set and the test set of the neural network, and after the LSTM neural network model is trained by the training set, the test set is transferred into the trained LSTM neural network model to obtain the prediction result. The transmission pressure cognitive model of the present invention makes a more accurate judgment on the end-to-end data transmission pressure, is conducive to the development of efficient communication technology on the edge side of distributed power access, and has important guiding significance for the security and efficiency of end-to-end data transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 A flow chart of a cross-domain end-to-end data transmission pressure recognition method provided by the present invention;

[0069] Figure 2 A schematic diagram of the structure of a cross-domain end-to-end data transmission pressure recognition device provided by the present invention;

[0070] Figure 3 A schematic diagram of a cross-domain end-to-end data transmission pressure recognition scenario provided by the present invention;

[0071] Figure 4 It is a schematic diagram of the scaling function and wavelet function of Haar wavelet used in the model decomposition process;

[0072] Figure 5 Schematic diagram of the binary tree of Haar wavelet scaling coefficients during the model decomposition process. DETAILED DESCRIPTION

[0073] The following detailed description of the embodiments of the present invention is further described in detail in conjunction with the accompanying drawings and examples. The detailed description of the following embodiments and the accompanying drawings are used to exemplarily illustrate the principles of the present invention, but cannot be used to limit the scope of the present invention, that is, the present invention is not limited to the preferred embodiments described, and the scope of the present invention is defined by the claims.

[0074] In the description of the present invention, it should be noted that, unless otherwise specified, “plurality” means two or more than two; the terms “first”, “second”, etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance; for ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0075] Embodiment 1

[0076] An embodiment of the present invention provides a cross-domain end-to-end data transmission pressure recognition method, such as Figure 1 As shown, its application scenarios are as follows Figure 3 As shown,

[0077] S1. Based on the given business data transmission requirements, the historical business flow is segmented to determine the fixed time slot business flow;

[0078] The historical business flow is:

[0079] X=[x(1),x(2)…x(T)]

[0080] The service flow of the fixed time slot is:

[0081] X t =[x(t),x(t+1)…x(t+l-1)]

[0082] Where X is a transmission pressure data stream with a type characteristic containing T data, l is the length of the fixed time slot service flow, and x(t) is the data packet successfully sent in the tth time slot.

[0083] S2. Perform one-dimensional discrete wavelet transform on the fixed time slot service flow data to obtain the original scale coefficient.

[0084] The original set of scaling coefficients is as follows:

[0085]

[0086] The calculation formula of the original scale coefficient is:

[0087] u 0,k =∫x(t)φ 0,k (t)dt

[0088] φ 0,k (t) = ε(tk) - ε(tk-1)

[0089] In the formula, u 0,k is the original scale factor, φ 0,k (t) is the scaling function, and ε(tk) is the step function shifted by k units.

[0090] S3, according to the original scale coefficient, using a multi-fractal wavelet model and an optimized decomposition series, performing coefficient decomposition to obtain an optimized decomposition coefficient sequence;

[0091] The schematic diagram of the scaling function and wavelet function of Haar wavelet used in the model decomposition process is as follows Figure 4 As shown;

[0092] The schematic diagram of the binary tree of Haar wavelet scaling coefficients during the model decomposition process is as follows: Figure 5 shown.

[0093] S31, select Haar wavelet, perform multi-fractal wavelet model decomposition on the original scale coefficients, and obtain an optimized coefficient set. The process is as follows:

[0094]

[0095] In the formula, u j,2k is the 2kth decomposition coefficient in the j-level decomposition; A j-1,k The random variable decomposed at level m-1 approximately follows Gaussian distribution, H is the Hurst parameter of the historical business flow, p is the self-similar variable of the random variable, and var is the variance;

[0096] S32. According to the multi-fractal wavelet model and the optimal decomposition level, the optimized decomposition coefficient sequence is determined as follows:

[0097]

[0098] Where M0 is the optimal decomposition level.

[0099] S4, determining the transmission pressure characteristic value according to the relationship between the average transmission delay of each data packet in the time slot corresponding to the fixed time slot service flow and the rated transmission delay;

[0100] The transmission pressure characteristic value is calculated as follows:

[0101]

[0102] Where P t is the transmission pressure evaluation value, The rated transmission delay of the service flow in a fixed time slot determined by the service communication quality requirement, S t is the average delay of all data packets transmitted in a fixed time slot, and when Time t =1, that is, P t ∈[0,1];

[0103] The average delay calculation formula for all data packets transmitted within a fixed time slot is as follows:

[0104]

[0105] Where s(t) is the actual transmission delay of the successfully sent data packet at time t.

[0106] S5, splicing the fixed time slot service flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value to obtain a transmission pressure characteristic signal sequence, and then dividing it according to proportion to obtain a training set and a test set of a neural network;

[0107] S51, concatenating the fixed time slot service flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value, as shown in the following formula:

[0108] Y t =[X t ,U t ,P t ]

[0109] Where Y t For X t The transmission pressure characteristic signal sequence;

[0110] S52: Determine a modeling data set based on the transmission pressure characteristic signal sequence:

[0111] Y=[Y1,Y2,...,Y T-l ];

[0112] S53, dividing the modeling data set into a training set and a test set, as shown in the following formula:

[0113]

[0114] In the formula, σ is the division ratio, E is the training set, and C is the test set.

[0115] S6. After training the LSTM neural network model using the training set, the test set is passed into the trained LSTM neural network model to obtain a prediction result.

[0116] S61, inputting the training set into the LSTM neural network model to perform feature extraction to obtain temporal abstract features;

[0117] S62, based on the time series abstract features, with the goal of minimizing the loss function, using the root mean square error formula to perform error calculation to obtain the trained LSTM neural network model;

[0118] S63: Input the test set into the trained LSTM neural network model to obtain a prediction result.

[0119] The training set is used for training, and the transmission pressure characteristic signal sequence is passed into the optimal LSTM neural network to calculate the time series abstract features. The model structure determination of the network involves the adjustment of many parameters. The root mean square error formula is used to calculate the error. The goal is to minimize the loss function. The transmission pressure characteristic signal sequence in the test set is input into the trained optimal LSTM neural network to obtain the prediction result.

[0120] The prediction result is a transmission pressure assessment value, which can be expressed in percentage or percent.

[0121] The parameters include the number of input data In_Number of the input layer of the LSTM model structure, the number of hidden layers Hide_Number, the number of output data Out_Number of the output layer, and the initial learning rate InitialLearnRate, the maximum number of training times MaxEpochs, the learning rate drop factor LearnRateDropFactor, the iterative update method Update_method of the weight parameters, etc. for adjusting the preferred LSTM model parameters.

[0122] This embodiment is a cross-domain end-to-end data transmission pressure recognition method. According to the given business data transmission requirements, the historical business flow is segmented, the fixed time slot business flow is determined, and a one-dimensional discrete wavelet transform is performed on it to obtain the original scale coefficient. The coefficient is decomposed using a multi-fractal wavelet model and an optimized decomposition series to obtain an optimized decomposition coefficient sequence, the average transmission delay and the rated transmission delay of each data packet in the corresponding time slot of the fixed time slot business flow are calculated, and a transmission pressure characteristic model is constructed. The fixed time slot business flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value are spliced ​​to obtain a transmission pressure characteristic signal sequence, a training set and a test set of a neural network are constructed, and an LSTM neural network model and hyperparameters are determined. The transmission pressure characteristic signal sequence in the training set is passed into the preferred LSTM, the neural network is trained, and the signal sequence in the test set is passed into the determined neural network to obtain a transmission pressure prediction result. The transmission pressure recognition model described in the present invention makes a more accurate judgment on the end-to-end data transmission pressure, which is conducive to the development of efficient communication technology on the edge side of distributed power access, and has important guiding significance for the reasonable planning and deployment of network resources of the power communication network.

[0123] Embodiment 2

[0124] Another embodiment of the present invention provides a cross-domain end-to-end data transmission pressure recognition device 200, such as Figure 2 As shown, it includes: a service flow segmentation module 201, which is used to segment the historical service flow based on the given service data transmission requirements and determine the fixed time slot service flow;

[0125] The wavelet transform module 202 is used to perform a one-dimensional discrete wavelet transform according to the fixed time slot service flow to obtain an original scale coefficient;

[0126] The coefficient decomposition module 203 is used to perform coefficient decomposition according to the original scale coefficients by using a multi-fractal wavelet model and an optimized decomposition series to obtain an optimized decomposition coefficient sequence;

[0127] A characteristic value determination module 204, configured to determine a transmission pressure characteristic value according to an average transmission delay and a rated transmission delay of each data packet in a time slot corresponding to the fixed time slot service flow;

[0128] The signal division module 205 is used to splice the fixed time slot service flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value to obtain a transmission pressure characteristic signal sequence and then divide it according to proportion to obtain a training set and a test set of a neural network;

[0129] The model training module 206 uses the training set to train the LSTM neural network model, and then passes the test set into the trained LSTM neural network model to obtain a prediction result.

[0130] Furthermore, the calculation formula of the original scale coefficient is:

[0131] u 0,k =∫x(t)φ 0,k (t)dt

[0132] φ 0,k (t) = ε(tk) - ε(tk-1)

[0133] In the formula, u 0,k is the original scale factor, φ 0,k (t) is the scaling function, and ε(tk) is the step function shifted by k units.

[0134] Furthermore, the coefficient decomposition module 203 includes:

[0135] The wavelet decomposition submodule 2031 is used to select Haar wavelet and perform multi-fractal wavelet model decomposition on the original scale coefficients to obtain an optimized coefficient set. The process is as follows:

[0136]

[0137] In the formula, u j,2k is the 2kth decomposition coefficient in the j-level decomposition; A j-1,k The random variable decomposed at level m-1 approximately follows Gaussian distribution, H is the Hurst parameter of the data stream of historical information, p is the self-similar variable of the random variable, and var is the variance;

[0138] The series decomposition submodule 2023 is used to determine the optimized decomposition coefficient sequence according to the multi-fractal wavelet model and the optimal decomposition series, as shown in the following formula:

[0139]

[0140] Where M0 is the optimal decomposition level.

[0141] This embodiment is a cross-domain end-to-end data transmission pressure recognition device, including a service flow segmentation module, which is used to segment the historical service flow based on the given service data transmission demand and determine the fixed time slot service flow;

[0142] The wavelet transform module is used to perform a one-dimensional discrete wavelet transform according to the fixed time slot service flow to obtain the original scale coefficient; the coefficient decomposition module is used to perform coefficient decomposition according to the original scale coefficient using a multi-fractal wavelet model and an optimized decomposition series to obtain an optimized decomposition coefficient sequence; the characteristic value determination module is used to determine the transmission pressure characteristic value according to the average transmission delay and the rated transmission delay of each data packet in the time slot corresponding to the fixed time slot service flow; the signal division module is used to splice the fixed time slot service flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value, and then divide the transmission pressure characteristic signal sequence according to the proportion to obtain the training set and test set of the neural network; the model training module uses the training set to train the LSTM neural network model, and then transfers the test set into the trained LSTM neural network model to obtain the prediction result. The transmission pressure cognitive model described in the present invention makes a more accurate judgment on the end-to-end data transmission pressure, which is conducive to the development of efficient communication technology on the edge side of distributed power access, and has important guiding significance for the security and efficiency of end-to-end data transmission.

[0143] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A cross-domain end-to-end data transmission pressure recognition method, characterized in that: include: S1. Based on the given business data transmission requirements, the historical business flow is segmented to determine the fixed time slot business flow; S2. Perform a one-dimensional discrete wavelet transform according to the fixed time slot service flow to obtain an original scale coefficient; S3, according to the original scale coefficient, using a multi-fractal wavelet model and an optimized decomposition series, performing coefficient decomposition to obtain an optimized decomposition coefficient sequence; S4, determining a transmission pressure characteristic value according to an average transmission delay and a rated transmission delay of each data packet in a time slot corresponding to the fixed time slot service flow; S5, splicing the fixed time slot service flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value to obtain a transmission pressure characteristic signal sequence, and then dividing it according to proportion to obtain a training set and a test set of a neural network; S6. After training the LSTM neural network model using the training set, the test set is passed into the trained LSTM neural network model to obtain a prediction result; The S3 includes: S31, select Haar wavelet, perform multi-fractal wavelet model decomposition on the original scale coefficients, and obtain an optimized coefficient set. The process is as follows: In the formula, u j,2k is the 2kth decomposition coefficient in the j-level decomposition; A j-1,k The random variable decomposed at level m-1 approximately follows Gaussian distribution, H is the Hurst parameter of the historical business flow, p is the self-similar variable of the random variable, and var is the variance; S32. According to the multi-fractal wavelet model and the optimal decomposition level, the optimized decomposition coefficient sequence is determined as follows: Where M0 is the optimal decomposition level.

2. The cross-domain end-to-end data transmission pressure recognition method according to claim 1 is characterized in that: The S1 includes: The historical business flow is: X=[x(1),x(2)…x(T)] The service flow of the fixed time slot is: X t =[x(t),x(t+1)…x(t+l-1)] Where X is a transmission pressure data stream with a type characteristic containing T data, l is the length of the fixed time slot service flow, and x(t) is the data packet successfully sent in the tth time slot.

3. The cross-domain end-to-end data transmission pressure recognition method according to claim 2 is characterized in that: The calculation formula of the original scale coefficient is: u 0,k =∫x(t)φ 0,k (t)dt f 0,k (t)=ε(tk)-ε(tk-1) In the formula, u 0,k is the original scale factor, φ 0,k (t) is the scaling function, and ε(tk) is the step function shifted by k units.

4. The cross-domain end-to-end data transmission pressure recognition method according to claim 1 is characterized in that: The transmission pressure characteristic value is calculated as follows: Where P t is the transmission pressure evaluation value, S The rated transmission delay of the service flow in a fixed time slot determined by the service communication quality requirement, S t is the average delay of all data packets transmitted in a fixed time slot; The average delay calculation formula for all data packets transmitted within a fixed time slot is as follows: Where s(t) is the actual transmission delay of the successfully sent data packet at time t.

5. The cross-domain end-to-end data transmission pressure recognition method according to claim 4 is characterized in that: The S5 includes: S51, concatenating the fixed time slot service flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value, as shown in the following formula: Y t =[X t ,U t ,P t ] Where Y t For X t The transmission pressure characteristic signal sequence; S52: Determine a modeling data set based on the transmission pressure characteristic signal sequence: Y=[Y1,Y2,...,Y T-l ]; S53, dividing the modeling data set into a training set and a test set, as shown in the following formula: In the formula, σ is the division ratio, E is the training set, and C is the test set.

6. The cross-domain end-to-end data transmission pressure recognition method according to claim 5 is characterized in that: The S6 includes: S61, inputting the training set into the LSTM neural network model to perform feature extraction to obtain temporal abstract features; S62, based on the time series abstract features, with the goal of minimizing the loss function, using the root mean square error formula to perform error calculation to obtain the trained LSTM neural network model; S63: Input the test set into the trained LSTM neural network model to obtain a prediction result.

7. A cross-domain end-to-end data transmission pressure recognition device, characterized in that: include: A service flow segmentation module is used to segment the historical service flow and determine the fixed time slot service flow based on the given service data transmission requirements; A wavelet transform module, used for performing a one-dimensional discrete wavelet transform according to the fixed time slot service flow to obtain an original scale coefficient; A coefficient decomposition module is used to perform coefficient decomposition according to the original scale coefficients by using a multi-fractal wavelet model and an optimized decomposition series to obtain an optimized decomposition coefficient sequence; A characteristic value determination module, used to determine a transmission pressure characteristic value according to an average transmission delay and a rated transmission delay of each data packet in a time slot corresponding to the fixed time slot service flow; A signal division module, used for splicing the fixed time slot service flow, the optimized decomposition coefficient sequence and the transmission pressure characteristic value to obtain a transmission pressure characteristic signal sequence and then dividing it according to proportion to obtain a training set and a test set of a neural network; A model training module, which trains the LSTM neural network model using the training set, and then transfers the test set to the trained LSTM neural network model to obtain a prediction result; The coefficient decomposition module comprises: The wavelet decomposition submodule is used to select Haar wavelet, perform multi-fractal wavelet model decomposition on the original scale coefficients, and obtain an optimized coefficient set. The process is as follows: In the formula, u j,2k is the 2kth decomposition coefficient in the j-level decomposition; A j-1,k The random variable decomposed at level m-1 approximately follows Gaussian distribution, H is the Hurst parameter of the data stream of historical information, p is the self-similar variable of the random variable, and var is the variance; The series decomposition submodule is used to determine the optimized decomposition coefficient sequence according to the multi-fractal wavelet model and the optimal decomposition series, as shown in the following formula: Where M0 is the optimal decomposition level.

8. The cross-domain end-to-end data transmission pressure recognition device according to claim 7, characterized in that: The calculation formula of the original scale coefficient is: u 0,k =∫x(t)φ 0,k (t)dt f 0,k (t)=ε(tk)-ε(tk-1) In the formula, u 0,k is the original scale factor, φ 0,k (t) is the scaling function, and ε(tk) is the step function shifted by k units.

Images