Power line communication time delay prediction method and device and electronic equipment

Abstract

The invention discloses a power line communication time delay prediction method and device and electronic equipment, and belongs to the technical field of power line communication control, and the power line communication time delay prediction method comprises the steps: decomposing a preprocessed initial time delay sequence into a trend component, a season component and a primary residual component at a time; decomposing the primary residual component into K sub-modes and a secondary residual component; only one time delay sequence needs to be collected, required priori knowledge is little, channel state information does not need to be obtained, and the calculation complexity is low; inputting the K + 3 component sequences into K + 3 trained bidirectional long-short-term memory networks in a one-to-one correspondence manner, and superposing the obtained K + 3 prediction results to obtain a final communication time delay prediction result; and the defect of low prediction precision caused by neglecting the residual error in the conventional method is overcome. Compared with a traditional linear prediction regression method, the method can also predict the time delay data in a longer time duration in a certain precision range under the condition that the historical time delay data is less.

Description

Technical Field

[0001] The present invention belongs to the technical field of power line communication control, and more specifically, relates to a power line communication delay prediction method, device and electronic equipment. Background Art

[0002] Power Line Communication (PLC) technology refers to a communication method that uses power lines to transmit data and media signals. This technology loads high frequency information onto current and then transmits it through wires. The adapter that receives the information then separates the high frequency from the current and transmits it to a computer or phone to achieve information transmission.

[0003] Power line communication is widely used in the field of Internet of Things. It is susceptible to electromagnetic interference and environmental influences, and there are a lot of sudden large delays, which challenges the real-time and reliability of communication. By predicting communication delays, allowing the system to respond and control in advance and taking compensation methods, the performance and service quality of the communication system can be improved.

[0004] At present, there are few studies on the time series prediction of power line communication delay, and they mainly focus on relatively stable delay series, and there is almost no prediction related to more difficult sudden large delays. Even if there are a few power line communication delay prediction schemes, the time series prediction methods of power line communication delay are often of low accuracy. Summary of the invention

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a power line communication delay prediction method, device and electronic device, which aims to solve the technical problem that the time series prediction method of the existing power line communication delay is often of low accuracy.

[0006] To achieve the above object, according to one aspect of the present invention, a method for predicting power line communication delay is provided, comprising:

[0007] S1: Identify and replace the noise points in the current delay sequence of the power line to obtain the initial delay sequence;

[0008] S2: Decomposing the initial time delay sequence into a trend component T(t), a seasonal component S(t) and a first-order residual component E(t);

[0009] S3: Decompose the first residual component E(t) to obtain K sub-modes {IMF1(t), IMF2(t),…,IMF K (t)} and the quadratic residual component e(t);

[0010] S4: Generate the current execution sample sequences corresponding to the trend component T(t), the seasonal component S(t), the K sub-modes and the quadratic residual component e(t);

[0011] S5: Inputting each type of currently executed sample sequence into the corresponding bidirectional long short-term memory network for prediction, and obtaining prediction results of each type of currently executed sample sequence;

[0012] S6: superimposing the prediction results of various types of currently executed sample sequences to obtain a prediction result of a final communication delay.

[0013] In one embodiment, the S1 includes:

[0014] S11: Detecting noise points of the current time delay sequence;

[0015] S12: Use a moving window average value to replace the noise point to obtain the initial time delay sequence.

[0016] In one embodiment, the S11 includes:

[0017] 1) Selecting an unvisited point P from the current delay sequence;

[0018] 2) Calculate the number of points of all neighbors of point P within the preset range ε. If the number of neighbor points exceeds the threshold, point P is considered a core point, otherwise it is a non-core point;

[0019] 3) If point P is a core point, add point P and all points in its neighborhood to a cluster; expand and check the neighborhood of each point in the cluster; if the number of points in the neighborhood is greater than the preset value and the point has not been processed, add it to the cluster;

[0020] 4) If point P is a non-core point, it is processed into a boundary point and a noise point; if point P does not have enough neighbors and is not in the neighborhood of any core point, it is marked as a noise point, otherwise it is classified as a boundary point;

[0021] 5) Repeat iterations 1)-4) until all points in the current delay sequence have been visited.

[0022] In one embodiment, the initial time delay sequence X(t) is the sum of the trend component T(t), the seasonal component S(t) and the first residual component E(t);

[0023] The first residual component E(t) is K sub-modes {IMF1(t),IMF2(t),…,IMF K (t)} and the sum of the quadratic residual component e(t).

[0024] In one embodiment, the S4 includes:

[0025] The trend component T(t), seasonal component S(t), K sub-modes and quadratic residual component e(t) are respectively used as sequences to be processed; for each sequence to be processed {x1, x2, …, x l}Use the formula Represents the current execution sample sequence generated;

[0026] Among them, {x1,,x2,…,x m} and {x m+1 ,,x m+2 ,…,x m+n} represent the label and features of the first sample respectively; {x2,x3,…,x m+1} and {x m+2 ,x m+3 ,…,x m+n+1} respectively represent the label and features of the second sample, {x i ,x i+1 ,…,x m+i-1} and {x m+i ,x m+i+1 ,…,x m+n+i-1} respectively represent the label and feature of the i-th sample; {x l-n-m+1 ,x l-n-m+2 ,…,x l-n} and {x l-n+1 ,x l-n+2 ,…,x l} respectively represent the label and feature of the (ln-m+1)th sample, i∈[1,ln-m+1]; l is the length of the sequence to be processed, m is the number of delay values ​​in the observation window, and n is the number of delay values ​​in the prediction window.

[0027] In one embodiment, S3 includes: using a variational mode decomposition algorithm to decompose the first-order residual E(t) into K sub-modes {IMF1(t), IMF2(t), ..., IMF K (t)} and quadratic residual e(t); wherein the parameter sub-mode number K and penalty factor α of the variational mode decomposition algorithm are determined by the grey wolf optimization algorithm.

[0028] In one embodiment, the fitness function of the gray wolf optimization algorithm takes the local envelope entropy as: in, is the probability distribution of the ith sub-mode, is the envelope sequence of the ith submode, u i (t) is the ith submode, H[u i (t)] is the i (t) Perform Hilbert transform.

[0029] According to another aspect of the present invention, there is provided a power line communication delay prediction device, comprising:

[0030] A preprocessing module, used for identifying and replacing noise points in the current delay sequence of the power line to obtain an initial delay sequence;

[0031] A primary decomposition module, used for decomposing the initial time delay sequence into a trend component T(t), a seasonal component S(t) and a primary residual component E(t);

[0032] A secondary decomposition module is used to decompose the first residual component E(t) to obtain K sub-modes {IMF1(t), IMF2(t),…,IMF K (t)} and the quadratic residual component e(t);

[0033] A sample generation module, used to generate the current execution sample sequence corresponding to each of the trend component T(t), the seasonal component S(t), the K sub-modes and the quadratic residual component e(t);

[0034] A network prediction module is used to input each type of currently executed sample sequence into a corresponding bidirectional long short-term memory network for prediction, and obtain prediction results of each type of currently executed sample sequence;

[0035] The post-processing module is used to superimpose the prediction results of various types of currently executed sample sequences to obtain the prediction results of the final communication delay.

[0036] According to another aspect of the present invention, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0037] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0038] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0039] (1) This scheme provides a power line communication delay prediction method, which decomposes the preprocessed initial delay sequence into a trend component T(t), a seasonal component S(t) and a primary residual component E(t); then decomposes the primary residual component E(t) into K sub-modes {IMF1(t), IMF2(t), …, IMF K(t)} and the quadratic residual component e(t); less prior knowledge is required, only a delay sequence of the past needs to be collected, no channel state information is required, and the computational complexity is low; then the above K+3 component sequences are input into K+3 trained bidirectional long short-term memory networks one by one to obtain K+3 prediction results; then the K+3 prediction results are superimposed to obtain the final prediction result of the communication delay; it overcomes the defect of the previous method that the residual is ignored and the prediction accuracy is low. Compared with the traditional linear prediction regression method, this method can work with less historical delay data and can predict the delay data for a longer time within a certain accuracy range.

[0040] (2) In this solution, the noise points are replaced by the moving window average value to obtain the initial time delay sequence, which can remove redundancy and retain the delay characteristics of the original sequence to a large extent.

[0041] (3) In this scheme, the DBSCAN algorithm is used to identify noise points. There are many different states in power line communication, such as idle, occupied, and congested. Therefore, the frequency distribution of delay will have multiple density centers. The DBSCAN algorithm can perfectly cluster different delay states and identify abnormal data.

[0042] (4) In this scheme, the initial time-delay sequence X(t) is the sum of the trend component T(t), the seasonal component S(t) and the first-order residual component E(t); the first-order residual component E(t) is the K sub-modes {IMF1(t), IMF2(t), …, IMF K (t)} and the sum of the quadratic residual component e(t); the above decomposition process retains the original delay characteristics to the greatest extent, which can ensure the accuracy of the subsequent prediction process.

[0043] (5) In this scheme, the starting point is initialized at a time delay value, and m data are taken as the label of a sample; then n data are taken as the feature of the sample; this method takes into account the observation window size m and the prediction window size n, and can traverse the entire sequence to be processed, which can ensure the accuracy of the subsequent prediction process.

[0044] (6) The number of sub-modes K and the penalty factor α of the variational mode decomposition algorithm in this scheme are determined by the gray wolf optimization algorithm. The local envelope entropy of the decomposed subsequence is selected as the fitness function, and the envelope entropy of each modal IMF is calculated. The optimization goal is to minimize it. At this time, α and K corresponding to the mode are the optimal VMD decomposition parameter combination; this method can achieve high-precision secondary decomposition of the residual component. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flowchart of a method for predicting power line communication delay provided in Embodiment 1 of the present invention;

[0046] Figure 2 A schematic diagram of a power line communication delay prediction method provided in Embodiment 1 of the present invention;

[0047] Figure 3 A histogram comparing the errors of the prediction method provided in Example 1 of the present invention and the method based on primary decomposition and LSTM network;

[0048] Figure 4 A power line delay data curve diagram and a frequency distribution diagram provided in Example 1 of the present invention;

[0049] Figure 5 This is a diagram showing the effect of data outlier processing provided in Example 1 of the present invention;

[0050] Figure 6 Iterative convergence curve diagram of the Grey Wolf Optimization Algorithm provided in Example 1 of the present invention;

[0051] Figure 7 A comparison chart of the predicted curve and the actual curve provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0053] Example 1

[0054] like Figure 1 and Figure 2 As shown, this embodiment provides a method for predicting power line communication delay, including: S1-S6. Specifically, S1: identifying and replacing noise points in the current delay sequence of the power line to obtain an initial delay sequence; S2: decomposing the initial delay sequence into a trend component T(t), a seasonal component S(t) and a primary residual component E(t); S3: decomposing the primary residual component E(t) to obtain K sub-modes {IMF1(t), IMF2(t),…,IMF K(t)} and the quadratic residual component e(t); S4: Generate the current execution sample sequences corresponding to the trend component T(t), the seasonal component S(t), the K sub-modes and the quadratic residual component e(t); S5: Input each type of current execution sample sequence into the corresponding bidirectional long short-term memory network for prediction, and obtain the prediction results of each type of current execution sample sequence; S6: Superimpose the prediction results of each type of current execution sample sequence to obtain the final communication delay prediction result, that is, the prediction results of each subsequence {p1(t), p2(t), …, p k+3 (t)} is synthesized to obtain the prediction result of the final communication delay Right now

[0055] Among them, the initial time delay sequence X(t) is the sum of the trend component T(t), the seasonal component S(t) and the first residual component E(t), that is, X(t) = T(t) + S(t) + E(t). The first residual component E(t) is K sub-modes {IMF1(t), IMF2(t), …, IMF K (t)} and the sum of the quadratic residual component e(t), that is, E(t) = IMF1(t) + IMF2(t) + ... + IMF K (t)+e(t). After two decompositions, the obtained subsequences have the following relationship with the initial sequence: X(t)=T(t)+S(t)+IMF1+IMF2+…+IMF K +e(t).

[0056] The bidirectional long short-term memory network used in this scheme is trained. Regarding the training process, the loss function can be set to mean square error MSE, the optimizer can be set to Adam, and the batch_size can be set to 16. K+3 bidirectional long short-term memory networks are established, and the training set D of each subsequence is used. tr Training on.

[0057] Figure 3 The error bar comparison chart of the embodiment of the present invention and the method based on one-step decomposition and LSTM network is shown. The prediction errors of the present invention and the standard LSTM model, the prediction model based on STL decomposition and LSTM, and the prediction model based on VMD decomposition and LSTM are compared. The results show that the delay prediction mean square error, mean absolute error, and mean percentage error of the design method of the present invention are all the smallest, and the goodness of fit is the largest. It can be seen that the estimation accuracy of the design method of the present invention is significantly better than that of the traditional one-step decomposition method.

[0058] In one embodiment, S1 includes: S11: detecting noise points of the current delay sequence; S12: replacing the noise points with a moving window average value to obtain an initial delay sequence; for example, Figure 4 The left part is the power line delay data curve used in this example, and the right part is the delay value frequency distribution diagram. The delay sequence curve has many spikes, and the delay value frequency distribution is long-tailed, indicating that there are some discrete large delays. The frequency distribution of the data has three peaks, namely the density center. The original sequence length is 2000, that is, X(t) = {x1, x2, …, x 2000}, use DBSCAN density clustering algorithm to detect outliers X in the original data i In order to prevent the clustering from being too tight or too loose, the parameter radius ε was set to 20 and the minimum number of neighbors MinPts was set to 30 after manual debugging. Finally, the data was clustered into 3 categories, which is consistent with the number of density centers in the frequency distribution diagram. Then, the moving window average Replace the outlier, where N is the window length centered on the outlier, and in this example, it is 10. Preferably, in other examples, the most suitable window length can be selected according to actual conditions.

[0059] Figure 5 This is the effect diagram of data outlier processing. The parameter radius ε of DBSCAN density clustering is set to 20, and the minimum number of neighbors MinPts is set to 30. The sample points are clustered into 3 categories, which is consistent with Figure 4 The abnormal points in the delay series curve are marked with red dots, and the smoothing window replacement method is used to deal with the abnormal values.

[0060] In one embodiment, the DBSCAN density clustering algorithm is used to detect outliers in the original data, and the outliers are replaced by the moving window average. There are many different states of power line communication, such as idle, occupied, and congested. The frequency distribution of delay will have multiple density centers. The DBSCAN algorithm can perfectly cluster different delay states and identify abnormal data.

[0061] Specifically, the iterative steps of the DBSCAN density clustering algorithm corresponding to S11 are:

[0062] 1) Select an unvisited point: Select an unvisited point P from the dataset.

[0063] 2) Calculate the number of points in the ε neighborhood to determine whether P is a core point: Calculate all neighbors of point P within the ε range. If the number of points in the neighborhood is greater than or equal to MinPts, then P is a core point.

[0064] 3) If P is a core point: If P is a core point, add P and all points in its neighborhood to a cluster. Expand each point in the cluster and check its neighborhood. If the number of points in the neighborhood is greater than MinPts and the point has not been processed, add it to the cluster and further check its neighborhood.

[0065] 4) If it is not a core point, it is processed as a boundary point and a noise point: If a point does not have enough neighbors (that is, less than MinPts points are in the ε neighborhood) and is not in the neighborhood of any core point, the point is marked as a noise point. Otherwise, the point will be classified as a boundary point, belonging to a cluster of core points, but it is not a core point itself.

[0066] 5) Repeat the above iterative steps until all points have been visited: Check all points in the data set until all points are assigned to clusters or marked as noise points.

[0067] In one embodiment, S4 includes: taking the trend component T(t), the seasonal component S(t), the K sub-modes and the quadratic residual component e(t) as sequences to be processed respectively; for each sequence to be processed, using represents the current execution sample sequence generated; where {x1,,x2,…,x m} and {x m+1 ,,x m+2 ,…,x m+n} represent the label and features of the first sample respectively; {x2,x3,…,x m+1} and {x m+2 ,x m+3 ,…,x m+n+1} respectively represent the label and features of the second sample, {x i ,x i+1 ,…,x m+i-1} and {x m+i ,x m+i+1 ,…,x m+n+i-1} respectively represent the label and feature of the i-th sample; {x l-n-m+1 ,x l-n-m+2 ,…,x l-n} and {x l-n+1 ,x l-n+2 ,…,x l} respectively represent the label and feature of the (ln-m+1)th sample, i∈[1,lnm-+1]; l is the length of the sequence to be processed, m is the number of delay values ​​in the observation window, and n is the number of delay values ​​in the prediction window.

[0068] Among them, after decomposition, {T(t), S(t), IMF1(t), IMF2(t),…, IMF KEach sequence in (t),e(t)} will perform the sample generation process. The observation window size is m, and the prediction window size is n. The data in the observation window is used as the sample feature, and the data in the prediction window is used as the sample label. The window position is continuously moved to construct the sample data set D. Assume that the original sequence X(t) = {x1, x2, …, x l}. When constructing a sample, the starting point is initialized at the position of the first delay value, and m data are taken as the label of the first sample, that is, {x1,,x2,…,x m}; Continue to take n data as the features of the first sample, that is, {x m+1 ,,x m+2 ,…,x m+n}. Move the starting point backward one position to x2, take m data as the features of the second sample, and then take n data as the label of the second sample. Move the starting point backward one position, and repeat the above process to complete the sample construction. The final constructed sample data set is:

[0069]

[0070] In one embodiment, S3 includes: using a variational mode decomposition algorithm to decompose the first-order residual E(t) into K sub-modes {IMF1(t), IMF2(t), …, IMF K (t)} and quadratic residual e(t); where the number of sub-modes K and penalty factor α of the variational mode decomposition algorithm are determined by the gray wolf optimization algorithm. Furthermore, the fitness function of the gray wolf optimization algorithm takes the local envelope entropy as: in, is the probability distribution of the ith sub-mode, is the envelope sequence of the ith submode, u i (t) is the ith submode, H[u i (t)] is the i (t) Perform Hilbert transform.

[0071] Figure 6 It is the iterative convergence curve of the gray wolf optimization algorithm. In this example, the local envelope entropy of the decomposed subsequence is selected as the fitness function, and the envelope entropy of each modal IMF is calculated. The optimization goal is to minimize it. At this time, α and K corresponding to the mode are the best VMD decomposition parameter combination. K is limited to the interval [3,20], and α is limited to the interval [1000,3000]. During the iteration process, the fitness function is constantly decreasing and finally converges to K = 9, α = 520.524.

[0072] Figure 7The comparison chart of the predicted curve and the real curve of this embodiment shows that the predicted curve and the real curve basically coincide with each other, and can accurately predict sudden large delays, with only slight jitter errors. Five measurement indicators are used to evaluate the reliability of the model from different angles. The mean square error MSE of the prediction results is 54.00ms, the root mean square error RMSE is 7.35ms, the mean absolute error MAE is 5.75ms, the average relative error MAPE is 0.0767, and the goodness of fit R2 is 0.986. For a large number of delay values ​​above 250ms in this example, the first four indicators all show that this model has high accuracy, and the goodness of fit R2 is 0.986, which is very close to 1, further showing that this model has a very high degree of fit to the original sequence.

[0073] Regarding the second decomposition, variational mode decomposition can be used to decompose the first residual E(t) into K sub-modes {IMF1(t),IMF2(t),…,IMF K (t)} and the quadratic residual e(t), the sum of the decomposed subsequences is equal to the original sequence, that is, E(t) = IMF1(t) + IMF2(t) + ... + IMF K After two decompositions, the subsequences finally obtained have the following relationship with the initial sequence: X(t) = T(t) + S(t) + IMF1 + IMF2 + ... + IMF K +e(t), and the set of each subsequence is recorded as U = {u1(t),u2(t),…,u K+3 (t)}. The number of sub-modal parameters K and the penalty factor α of VMD decomposition are optimized by the gray wolf optimization algorithm, and the fitness function takes the local envelope entropy E i First, for the subsequence u i (t) Perform Hilbert transform demodulation to obtain the envelope signal sequence Then calculate the probability distribution of the signal Finally, calculate the envelope entropy of the signal

[0074] Example 2

[0075] The present embodiment provides a power line communication delay prediction device, including: a preprocessing module, a primary decomposition module, a secondary decomposition module, a sample generation module, a network prediction module and a post-processing module. The preprocessing module is used to identify and replace the noise points in the current delay sequence of the power line to obtain an initial delay sequence; the primary decomposition module is used to decompose the initial delay sequence into a trend component T(t), a seasonal component S(t) and a primary residual component E(t); the secondary decomposition module is used to decompose the primary residual component E(t) to obtain K sub-modes {IMF1(t), IMF2(t),…,IMF K(t)} and the quadratic residual component e(t); a sample generation module is used to generate the current execution sample sequences corresponding to the trend component T(t), the seasonal component S(t), the K sub-modes and the quadratic residual component e(t); a network prediction module is used to input each type of current execution sample sequence into the corresponding bidirectional long short-term memory network for prediction, and obtain the prediction results of each type of current execution sample sequence; a post-processing module is used to superimpose the prediction results of each type of current execution sample sequence to obtain the final communication delay prediction result.

[0076] Example 3

[0077] This embodiment provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0078] Example 4

[0079] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0080] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting power line communication delay, characterized in that: include: S1: Identify and replace the noise points in the current delay sequence of the power line to obtain the initial delay sequence; S2: Decomposing the initial time delay sequence into a trend component T(t), a seasonal component S(t) and a first-order residual component E(t); S3: Decompose the first residual component E(t) to obtain K sub-modes {IMF1(t), IMF2(t),…,IMF K (t)} and the quadratic residual component e(t); S4: Generate the current execution sample sequences corresponding to the trend component T(t), the seasonal component S(t), the K sub-modes and the quadratic residual component e(t); S5: Inputting each type of currently executed sample sequence into the corresponding bidirectional long short-term memory network for prediction, and obtaining prediction results of each type of currently executed sample sequence; S6: superimposing the prediction results of various types of currently executed sample sequences to obtain a prediction result of a final communication delay.

2. The power line communication delay prediction method according to claim 1, characterized in that: The S1 includes: S11: Detecting noise points of the current time delay sequence; S12: Use a moving window average value to replace the noise point to obtain the initial time delay sequence.

3. The power line communication delay prediction method according to claim 2, characterized in that: The S11 includes: 1) Selecting an unvisited point P from the current delay sequence; 2) Calculate the number of points of all neighbors of point P within the preset range ε. If the number of neighbor points exceeds the threshold, point P is considered a core point, otherwise it is a non-core point; 3) If point P is a core point, add point P and all points in its neighborhood to a cluster; expand and check the neighborhood of each point in the cluster; if the number of points in the neighborhood is greater than the preset value and the point has not been processed, add it to the cluster; 4) If point P is a non-core point, it is processed into a boundary point and a noise point; if point P does not have enough neighbors and is not in the neighborhood of any core point, it is marked as a noise point, otherwise it is classified as a boundary point; 5) Repeat iterations 1)-4) until all points in the current delay sequence have been visited.

4. The power line communication delay prediction method according to claim 1, characterized in that: The initial time delay sequence X(t) is the sum of the trend component T(t), the seasonal component S(t) and the first residual component E(t); The first residual component E(t) is K sub-modes {IMF1(t), IMF2(t), ..., IMF K (t)} and the sum of the quadratic residual component e(t).

5. The power line communication delay prediction method according to claim 4, characterized in that: The S4 includes: The trend component T(t), seasonal component S(t), K sub-modes and quadratic residual component e(t) are respectively used as sequences to be processed; for each sequence to be processed {x1, x2, ..., x l }Use the formula Represents the current execution sample sequence generated; Among them, {x1,,x2,…,x m } and {x m+1 ,,x m+2 , …, x m+n } represent the label and features of the first sample respectively; {x2, x3, …, x m+1 } and {x m+2 , x m+3 , …, x m+n+1 } respectively represent the label and features of the second sample, {x i , x i+1 , …, x m+i-1 } and {x m+i , x m+i+1 , …, x m+n+i-1 } respectively represent the label and feature of the i-th sample; {x l-n-m+1 , x l-n-m+2 , …, x l-n } and {x l-n+1 , x l-n+2 , …, x l } respectively represent the label and feature of the (1-n-m+1)th sample, i∈[1,ln-m+1]; l is the length of the sequence to be processed, m is the number of delay values ​​in the observation window, and n is the number of delay values ​​in the prediction window.

6. The power line communication delay prediction method according to claim 1, characterized in that: The S3 includes: The variational mode decomposition algorithm is used to decompose the first-order residual E(t) into K sub-modes {IMF1(t), IMF2(t), …, IMF K (t)} and quadratic residual e(t); wherein the parameter sub-mode number K and penalty factor α of the variational mode decomposition algorithm are determined by the grey wolf optimization algorithm.

7. The method for predicting power line communication delay according to claim 6, characterized in that: The fitness function of the gray wolf optimization algorithm takes the local envelope entropy as: in, is the probability distribution of the ith sub-mode, is the envelope sequence of the ith submode, u i (t) is the ith submode, H[u i (t)] is the i (t) Perform Hilbert transform.

8. A power line communication delay prediction device, characterized in that: include: A preprocessing module, used for identifying and replacing noise points in the current delay sequence of the power line to obtain an initial delay sequence; A primary decomposition module, used for decomposing the initial time delay sequence into a trend component T(t), a seasonal component S(t) and a primary residual component E(t); A secondary decomposition module is used to decompose the first residual component E(t) to obtain K sub-modes {IMF1(t), IMF2(t), ..., IMF K (t)} and the quadratic residual component e(t); A sample generation module, used to generate the current execution sample sequence corresponding to each of the trend component T(t), the seasonal component S(t), the K sub-modes and the quadratic residual component e(t); A network prediction module is used to input each type of currently executed sample sequence into a corresponding bidirectional long short-term memory network for prediction, and obtain prediction results of each type of currently executed sample sequence; The post-processing module is used to superimpose the prediction results of various types of currently executed sample sequences to obtain the prediction results of the final communication delay.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images