Wind speed prediction method along high-speed railway lines based on secondary decomposition and reconstruction

Abstract

The present invention belongs to the technical field of railway disaster monitoring, and specifically relates to a method for predicting wind speed along a high-speed railway based on secondary decomposition and reconstruction, including: S1: analyzing the original wind speed data along the high-speed railway to construct a structural time series; S2: using CEEMDAN to perform a primary decomposition of the wind speed time series to obtain a series of signal components; S3: calculating the SE value of the signal, quantifying the complexity of the signal components, dividing the IMF components into three groups according to the SE, and reconstructing the signal groups to generate HIMF, MIMF and LIMF respectively; S4: using VMD to further decompose the HIMF to obtain a series of relatively stable subsequences; S5: predicting the signal components respectively, adding the obtained results to obtain the final predicted wind speed. The present invention avoids modal aliasing and noise introduction caused by excessive decomposition, improves the effectiveness of signal decomposition, and improves the accuracy of strong wind prediction along the high-speed railway.

Description

Technical Field

[0001] The invention belongs to the technical field of railway disaster monitoring, and in particular relates to a method for predicting wind speed along a high-speed railway based on secondary decomposition and reconstruction. Background Art

[0002] As an important mode of modern transportation, high-speed railways have fast running speeds, strong carrying capacity, and little impact on the environment. Instantaneous strong winds can cause foreign objects to invade the railway network, damage railway infrastructure, and cause train delays or even suspension of operations; strong side winds can affect the stability of moving trains, increasing the risk of train derailment or overturning. Therefore, quickly and accurately predicting wind speeds along the line is crucial to ensuring the safe operation of the railway system.

[0003] Given the significant randomness and volatility of wind, it is difficult to accurately predict the future wind speed change trend by directly analyzing the original wind speed data. Therefore, the existing wind speed prediction work usually combines signal decomposition technology to decompose the complex original wind speed data into multiple relatively stable sub-signals, which are then input into the prediction model respectively, and the prediction results of the obtained sub-signals are summed to obtain the final predicted wind speed. Common signal decomposition techniques include fully adaptive noise ensemble empirical mode decomposition (CEEMDAN), empirical mode decomposition (EMD) and integrated empirical mode decomposition (EEMD). CEEMDAN is widely used because it can effectively solve the modal aliasing problem of EMD and the noise residual problem of EEMD. However, if only a single decomposition is performed on the wind speed data, there will still be residual noise in some sub-signals (especially high-frequency sub-signals), which will affect the effect of the prediction model, and over-decomposition will lead to modal aliasing and noise introduction. In addition, if the deep learning network is directly used to predict all the decomposed sub-signals, the computational overhead is large and the time-consuming is long, which is not suitable for railway strong wind prediction scenarios with high real-time requirements. Summary of the invention

[0004] The purpose of the present invention is to provide a method for predicting wind speed along high-speed railways based on secondary decomposition and reconstruction, so as to solve the technical problems in the prior art that only a single decomposition is performed on wind speed data, noise still remains in the high-frequency sub-signal, and excessive decomposition leads to modal aliasing and noise introduction, thus affecting the effect of the prediction model.

[0005] The method for predicting wind speed along the high-speed railway based on secondary decomposition and reconstruction comprises the following steps:

[0006] S1: Analyze the original wind speed data along the high-speed railway and construct a tectonic time series;

[0007] S2: Use CEEMDAN to perform the initial decomposition of the wind speed time series and obtain a series of signal components;

[0008] S3: Calculate the SE value of the signal to quantify the complexity of the signal components. According to the SE, the IMF components of a series of signals are divided into three groups. The signal groups are reconstructed to generate HIMF, MIMF and LIMF respectively.

[0009] S4: VMD is used to further decompose HIMF and obtain a series of relatively stable subsequences;

[0010] S5: predict the signal components separately, add the obtained results, and obtain the final predicted wind speed, wherein the signal components include MIMF and LIMF, and the subsequence obtained by secondary decomposition of HIMF.

[0011] Preferably, the step S4 specifically includes the following sub-steps:

[0012] Step S4.1 Initialize parameters ,in Refers to IMF components, Refers to The center frequency of the IMF component, is the Lagrange coefficient, , Indicates the number of sub-signals obtained by VMD decomposition;

[0013] Step S4.2 Update parameters and , the calculation formula is as follows:

[0014]

[0015]

[0016] in, , and They represent the main signal to be decomposed in the frequency domain and the IMF components and Lagrangian coefficients in the frequency domain, and Iteration The second time IMF components and corresponding center frequencies, , , and Respectively for , , and The result after performing Fourier transform; is the quadratic penalty coefficient, which is used to weaken the interference of Gaussian noise. Indicates the number of iterations;

[0017] Step S4.3 Update parameters :

[0018]

[0019] in, Refers to the noise tolerance factor;

[0020] Step S4.4 determines whether to terminate the iteration based on the iteration termination condition;

[0021] The iteration termination condition is as follows:

[0022]

[0023] Refers to iteration The second time i IMF components, ;

[0024] Determine whether the iteration termination condition is met. If so, terminate the iteration; otherwise, return to step 4.2 until the iteration termination condition is met.

[0025] Step S4.5 Determine the key parameters and number of components in VMD decomposition based on the signal center frequency I The value of .

[0026] Preferably, the step S4.5 further includes the following sub-steps:

[0027] Step S4.5.1 Initialize the number of IMFs obtained after VMD decomposition ;

[0028] Step S4.5.2 obtains and stores the center frequency of each IMF component, represents the set of center frequencies of all IMF components;

[0029] Step S4.5.3 Calculate the maximum and minimum center frequencies of the IMF component signal. The corresponding formula is as follows:

[0030]

[0031]

[0032] in are respectively obtained by decomposition The maximum and minimum center frequencies of the IMF components;

[0033] Step S4.5.4: Determine whether the iteration termination condition is met. If so, output ,otherwise , return to step 4.5.2 and continue iterating until the termination condition is met. The iteration termination condition is as follows

[0034]

[0035] Determine the number of components at the end of the iteration I ;

[0036] Step S4.5.5 Output VMD decomposition I IMF components, respectively VIMF 1, VIMF 2,…, VIMF I .

[0037] Preferably, the step S2 specifically includes the following sub-steps:

[0038] Step S2.1 Add Gaussian white noise to the original wind speed sequence multiple times to generate a noise-enhanced wind speed sequence : ,in, refers to the original wind speed series, is the corresponding noise standard deviation, represents Gaussian white noise, is the total number of samples;

[0039] Step S2.2: Enhance the wind speed sequence with noise Performing Empirical Mode Decomposition Operation times, calculate the mean of the results, and the corresponding calculation formula is:

[0040]

[0041] in, represents the first intrinsic mode component obtained after CEEMDAN decomposition, Refers to the number of executions of EMD. is the first modal component obtained after EMD decomposition;

[0042] Step S2.3 Calculate the value obtained in step 2.2 Corresponding residual , the formula is:

[0043]

[0044] in, represents the first intrinsic mode component obtained after CEEMDAN decomposition, for The corresponding residual of Refers to the original wind speed series;

[0045] Step S2.4 Calculate the second intrinsic mode component obtained by CEEMDAN decomposition , and its corresponding calculation formula is:

[0046]

[0047] in, Refers to the number of executions of the EMD operation. is the corresponding noise standard deviation;

[0048] Step S2.5 Repeat steps 2.3 and 2.4 to obtain the intrinsic modal components and corresponding residuals obtained by CEEMDAN decomposition until the residual components meet becomes a monotonic function; the calculation formula of the intrinsic mode components and the corresponding residuals is as follows:

[0049]

[0050]

[0051] in, and Respectively The intrinsic mode components and the corresponding residuals, represents the first The intrinsic mode components, is the corresponding noise standard deviation.

[0052] Preferably, the step S3 specifically includes the following sub-steps:

[0053] Step S3.1 Convert the given continuous signal into a set of Dimensional embedding vector:

[0054]

[0055] in, Refers to The beginning Continuous wind speed values, i.e. dimensional embedding vector, represents the signal element in the embedding vector;

[0056] Step S3.2 calculates the distance between the two embedded vectors. The corresponding calculation method is as follows:

[0057]

[0058] in, express and The distance between two embedding vectors, and Indicates the corresponding signal element and the signal elements;

[0059] Step S3.3 For each , count all the conditions that meet The number of satisfy , then:

[0060]

[0061] in, It means that two embedding vectors match The probability of a point, Refers to similarity tolerance, Indicates that all conditions are met the number of

[0062] Step S3.4 Calculation The corresponding formula is as follows:

[0063]

[0064] in, Then The corresponding average value;

[0065] Step S3.5 Increase the dimension to Repeat steps 3.1-3.4 to count the number of The number of satisfy , and calculate the corresponding average value ;

[0066] Step S3.6 calculates the sample entropy value of the given continuous signal:

[0067]

[0068] in, express The sample entropy corresponding to the dimensional vector;

[0069] Step S3.7 sets a threshold interval according to the calculated sample entropy value, and divides the sub-signals into three groups of high complexity, medium complexity and low complexity according to the sample entropy values, thus completing signal reconstruction.

[0070] Preferably, step S5 uses a TCGRU network to predict the MIMF and LIMF signals respectively. The TCGRU network is composed of TCN and GRU, and can simultaneously extract short-term fluctuations and long-term dependencies of wind speed data, and also uses the TCGRU network to predict the subsequence obtained by secondary decomposition of HIMF.

[0071] Afterwards, the prediction results of each sub-signal are obtained and added together to output the final prediction value.

[0072] The calculation formula is: ,

[0073] in, Refers to the first The prediction result corresponding to the sub-signal is and They refer to the predicted values ​​corresponding to the reconstructed MIMF and LIMF signals respectively.

[0074] The present invention has the following advantages: the present invention uses CEEMDAN to perform preliminary decomposition of the original wind speed signal, and uses SE to quantify the complexity of the decomposed signal to achieve signal grouping reconstruction, effectively reduce the computational overhead, and use VMD to perform secondary decomposition of HIMF to generate a series of relatively stable subsequences, reducing the difficulty of prediction. Construct TCGRU to predict all signals obtained by VMD decomposition and the remaining reconstructed signals respectively, merge the prediction results of each component, and output the final predicted wind speed. Experiments were carried out on a real high-speed rail wind speed data set, and the experimental results verified the effectiveness of the proposed model in the field of strong wind prediction. The present invention iteratively calculates the maximum and minimum center frequencies of the sub-signals obtained in the VMD secondary decomposition, and determines the key parameter IMF component number I in the VMD in the secondary decomposition stage, which determines the number of sub-signals obtained by decomposition. In this way, the frequency difference of the signal components is strictly controlled to avoid modal aliasing and noise introduction due to excessive decomposition, thereby improving the effectiveness of signal decomposition. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 It is a flow chart of the method for predicting wind speed along a high-speed railway line based on secondary decomposition and reconstruction of the present invention.

[0076] Figure 2 Schematic diagram of the prediction network TCGRU structure in the present invention. DETAILED DESCRIPTION

[0077] The specific implementation modes of the present invention are further explained in detail below by describing the embodiments with reference to the accompanying drawings, so as to help those skilled in the art to have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.

[0078] like Figure 1 , Figure 2 As shown, the present invention provides a method for predicting wind speed along a high-speed railway based on secondary decomposition and reconstruction, comprising the following steps:

[0079] S1: Analyze the original wind speed data along the high-speed railway and construct a tectonic time series.

[0080] In this step, the monitoring data of wind speed sensors deployed along the railway, namely wind speed data, are collected, and a time series is constructed using the collected wind speed data.

[0081] S2: The wind speed time series is initially decomposed using the fully adaptive ensemble empirical mode decomposition with noise (CEEMDAN) to obtain a series of signal components.

[0082] This step specifically includes the following sub-steps:

[0083] Step S2.1 Add Gaussian white noise to the original wind speed sequence multiple times to generate a noise-enhanced wind speed sequence : ,in, refers to the original wind speed series, is the corresponding noise standard deviation, represents Gaussian white noise, is the total number of samples.

[0084] Step S2.2: Enhance the wind speed sequence with noise Performs the Empirical Mode Decomposition (EMD) operation times, calculate the mean of the results, and the corresponding calculation formula is:

[0085]

[0086] in, represents the first intrinsic mode component obtained after CEEMDAN decomposition, Refers to the number of executions of EMD. is the first modal component obtained after EMD decomposition.

[0087] Step S2.3 Calculate the value obtained in step 2.2 Corresponding residual , the formula is:

[0088]

[0089] in, represents the first intrinsic mode component obtained after CEEMDAN decomposition, for The corresponding residual of Refers to the original wind speed series.

[0090] Step S2.4 Calculate the second intrinsic mode component obtained by CEEMDAN decomposition , and its corresponding calculation formula is:

[0091]

[0092] in, Refers to the number of executions of the EMD operation. is the corresponding noise standard deviation.

[0093] Step S2.5 Repeat steps 2.3 and 2.4 to obtain the intrinsic modal components and corresponding residuals obtained by CEEMDAN decomposition until the residual components meet Become a monotonic function. The calculation formula of the intrinsic mode components and the corresponding residuals is as follows:

[0094]

[0095]

[0096] in, and Respectively The intrinsic mode components and the corresponding residuals, represents the first The intrinsic mode components, is the corresponding noise standard deviation.

[0097] S3: Calculate the sample entropy (SE) value of the signal to quantify the complexity of the signal components. According to the SE, a series of signal component intrinsic modal components (CIMF) are divided into three groups. The signal groups are reconstructed to generate high complexity signals (HIMF), medium complexity signals (MIMF) and low complexity signals (LIMF).

[0098] This step specifically includes the following sub-steps:

[0099] Step S3.1 Convert the given continuous signal into a set of Dimensional embedding vector:

[0100]

[0101] in, Refers to The beginning Continuous wind speed values, i.e. dimensional embedding vector, Represents the signal element in the embedding vector.

[0102] Step S3.2 calculates the distance between the two embedded vectors. The corresponding calculation method is as follows:

[0103]

[0104] in, express and The distance between two embedding vectors, and Indicates the corresponding signal element and the signal element.

[0105] Step S3.3 For each , count all the conditions that meet The number of satisfy , then:

[0106]

[0107] in, It means that two embedding vectors match The probability of a point, Refers to similarity tolerance, Indicates that all conditions are met The number of

[0108] Step S3.4 Calculation The corresponding formula is as follows:

[0109]

[0110] in, Then The corresponding average value.

[0111] Step S3.5 Increase the dimension to Repeat steps 3.1-3.4 to count the number of The number of satisfy , and calculate the corresponding average value .

[0112] Step S3.6 Calculate the SE value of the given continuous signal:

[0113]

[0114] in, express The sample entropy corresponding to the dimensional vector.

[0115] Step S3.7 sets the threshold interval according to the calculated SE value, and divides the sub-signals into three groups of high complexity (HIMF), medium complexity (MIMF) and low complexity (LIMF) according to the SE values ​​to complete the signal reconstruction.

[0116] S4: Variational mode decomposition (VMD) is used to further decompose the high-complexity signal to obtain a series of relatively stable subsequences.

[0117] This step specifically includes the following sub-steps:

[0118] Step S4.1 Initialize parameters ,in Refers to IMF components, Refers to The center frequency of the IMF component, is the Lagrange coefficient, , Indicates the number of sub-signals obtained by VMD decomposition.

[0119] Step S4.2 Update parameters and , the calculation formula is as follows:

[0120]

[0121]

[0122] in, , and They represent the main signal to be decomposed in the frequency domain and the IMF components and Lagrangian coefficients in the frequency domain, and Iteration The second time IMF components and corresponding center frequencies, , , and Respectively for , , and The result after performing Fourier transform. is the quadratic penalty coefficient, which is used to weaken the interference of Gaussian noise. Indicates the number of iterations.

[0123] Step S4.3 Update parameters :

[0124]

[0125] in, Refers to the noise tolerance factor.

[0126] Step S4.4 determines whether to terminate the iteration based on the iteration termination condition.

[0127] The iteration termination condition is as follows:

[0128]

[0129] Refers to iteration The second time i IMF components, .

[0130] Determine whether the iteration termination condition is met. If so, terminate the iteration; otherwise, return to step 4.2 until the iteration termination condition is met.

[0131] Step S4.5 Determine the key parameters and number of components in VMD decomposition based on the signal center frequency I The value of .

[0132] Step S4.5 also includes the following sub-steps:

[0133] Step S4.5.1 Initialize the number of IMFs obtained after VMD decomposition , .

[0134] Step S4.5.2 obtains and stores the center frequency of each IMF component, Represents the set of center frequencies of all IMF components.

[0135] Step S4.5.3 Calculate the maximum and minimum center frequencies of the IMF component signal. The corresponding formula is as follows:

[0136]

[0137]

[0138] in are respectively obtained by decomposition The maximum and minimum center frequencies of the IMF components.

[0139] Step S4.5.4: Determine whether the iteration termination condition is met. If so, output ,otherwise , return to step 4.5.2 and continue iterating until the termination condition is met. The iteration termination condition is as follows

[0140]

[0141] Determine the number of components at the end of the iteration I .

[0142] Step S4.5.5 Output VMD decomposition I IMF components, respectively VIMF 1, VIMF 2,…, VIMF I .

[0143] S5: predict the signal components separately, add the obtained results, and obtain the final predicted wind speed, wherein the signal components include medium complexity signals, low complexity signals, and subsequences obtained by secondary decomposition of high complexity signals.

[0144] This step uses the Temporal Convolutional Gated Recurrent Unit (TCGRU) network to predict the MIMF and LIMF signals respectively. The TCGRU network consists of a Temporal Convolutional Network (TCN) and a Gated Recurrent Unit (GRU), which can simultaneously extract short-term fluctuations and long-term dependencies of wind speed data. The TCGRU network is also used to predict the subsequences obtained by secondary decomposition of high-complexity signals.

[0145] Afterwards, the prediction results of each sub-signal are obtained and added together to output the final prediction value.

[0146] The calculation formula is: ,

[0147] in, Refers to the first The prediction result corresponding to the sub-signal is and They refer to the predicted values ​​corresponding to the reconstructed MIMF and LIMF signals respectively.

[0148] In order to evaluate the effectiveness of the present invention, experiments were performed to verify the effectiveness of the present invention, and the specific contents are as follows.

[0149] Table 1 shows the experimental results of different signal decomposition methods for a data set consisting of wind speed data along the Beijing-Baotou Railway. The results show that compared with the three existing decomposition processing methods of EMD, CEEMDAN and VMD, the method provided by the present invention performs best in multiple evaluation indicators. Specifically, the present method is optimal in both the mean square error (MSE) and the root mean square error (RMSE) indicators, with the results being 0.1214 and 0.3484, respectively. This shows that the present method has significant advantages in capturing the absolute error of wind speed data, which is crucial for the accuracy of strong wind predictions along high-speed railways. Although the present method does not perform as well as EMD and CEEMDAN in the relative error indicator of mean absolute percentage error (MAPE), in the task of strong wind prediction along high-speed railways, more attention is paid to absolute error because it is directly related to the accuracy of the decision-making process. In addition, in the determination coefficient R that can be used to evaluate the model's ability to capture wind speed fluctuations, the absolute error is directly related to the accuracy of the decision-making process. 2 In terms of indicators, the present invention also has excellent performance. These results verify that the present invention is more effective than the prior art in processing complex signals.

[0150] Table 1: Index comparison table between the present invention and existing different signal decomposition methods.

[0151]

[0152] The present invention is described above by way of example in conjunction with the accompanying drawings. It is obvious that the specific implementation of the present invention is not limited to the above-mentioned method. As long as various non-substantial improvements are made using the inventive concept and technical solution of the present invention, or the inventive concept and technical solution are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A method for predicting wind speed along high-speed railway lines based on secondary decomposition and reconstruction, characterized by: The following steps are involved: S1: Analyze the original wind speed data along the high-speed railway and construct a tectonic time series; S2: Use CEEMDAN to perform the initial decomposition of the wind speed time series and obtain a series of signal components; S3: Calculate the SE value of the signal to quantify the complexity of the signal components. According to the SE, divide the signal components of a series of signals into three groups, and reconstruct the signal groups to generate HIMF, MIMF and LIMF respectively. S4: VMD is used to further decompose HIMF and obtain a series of relatively stable subsequences; S5: predict the signal components separately, add the obtained results, and obtain the final predicted wind speed, where the signal components include MIMF and LIMF, and the subsequence obtained by secondary decomposition of HIMF; The step S4 specifically includes the following sub-steps: Step S4.1 Initialize parameters ,in Refers to IMF components, Refers to The center frequency of the IMF component, is the Lagrange coefficient, , Indicates the number of sub-signals obtained by VMD decomposition; Step S4.2 Update parameters and , the calculation formula is as follows: in, , and They represent the main signal to be decomposed in the frequency domain and the IMF components and Lagrangian coefficients in the frequency domain, and Iteration The second time IMF components and corresponding center frequencies, , , and Respectively for , , and The result after performing Fourier transform; is the quadratic penalty coefficient, used to weaken the interference of Gaussian noise; Indicates the number of iterations; Step S4.3 Update parameters : in, Refers to the noise tolerance factor; Step S4.4 determines whether to terminate the iteration based on the iteration termination condition; The iteration termination condition is as follows: Refers to iteration The second time i IMF components, ; Determine whether the iteration termination condition is met. If so, terminate the iteration; otherwise, return to step 4.2 until the iteration termination condition is met. Step S4.5 Determine the key parameters and number of components in VMD decomposition based on the signal center frequency I The value of The step S4.5 also includes the following sub-steps: Step S4.5.1 Initialize the number of IMFs obtained after VMD decomposition ; Step S4.5.2 obtains and stores the center frequency of each IMF component, Represents the set of center frequencies of all IMF components; Step S4.5.3 Calculate the maximum and minimum center frequencies of the IMF component signal. The corresponding formula is as follows: in are respectively obtained by decomposition The maximum and minimum center frequencies of the IMF components; Step S4.5.4: Determine whether the iteration termination condition is met. If so, output ,otherwise , return to step 4.5.2 and continue iterating until the termination condition is met. The iteration termination condition is as follows Determine the number of components at the end of the iteration I ; Step S4.5.5 Output VMD decomposition I IMF components, respectively VIMF 1, VIMF 2,…, VIMF I .

2. The method for predicting wind speed along a high-speed railway line based on secondary decomposition and reconstruction according to claim 1 is characterized in that: The step S2 specifically includes the following sub-steps: Step S2.1 Add Gaussian white noise to the original wind speed sequence multiple times to generate a noise-enhanced wind speed sequence : ,in, refers to the original wind speed series, is the corresponding noise standard deviation, represents Gaussian white noise, is the total number of samples; Step S2.2: Enhance the wind speed sequence with noise Performing Empirical Mode Decomposition Operation times, calculate the mean of the results, and the corresponding calculation formula is: in, represents the first intrinsic mode component obtained after CEEMDAN decomposition, Refers to the number of executions of EMD. is the first modal component obtained after EMD decomposition; Step S2.3 Calculate the value obtained in step 2.2 Corresponding residual , the formula is: in, represents the first intrinsic mode component obtained after CEEMDAN decomposition, for The corresponding residual of Refers to the original wind speed series; Step S2.4 Calculate the second intrinsic mode component obtained by CEEMDAN decomposition , and its corresponding calculation formula is: in, Refers to the number of executions of the EMD operation. is the corresponding noise standard deviation; Step S2.5 Repeat steps 2.3 and 2.4 to obtain the intrinsic modal components and corresponding residuals obtained by CEEMDAN decomposition until the residual components meet becomes a monotonic function; the calculation formula of the intrinsic mode components and the corresponding residuals is as follows: in, and Respectively The intrinsic mode components and the corresponding residuals, Represents the first The intrinsic mode components, is the corresponding noise standard deviation.

3. The method for predicting wind speed along a high-speed railway based on secondary decomposition and reconstruction according to claim 1 is characterized in that: The step S3 specifically includes the following sub-steps: Step S3.1 Convert the given continuous signal into a set of Dimensional embedding vector: in, Refers to The beginning Continuous wind speed values, i.e. dimensional embedding vector, represents the signal element in the embedding vector; Step S3.2 calculates the distance between the two embedded vectors. The corresponding calculation method is as follows: in, express and The distance between two embedding vectors, and Indicates the corresponding signal element and the signal elements; Step S3.3 For each , count all the conditions that meet The number of satisfy , then: in, It means that two embedding vectors match The probability of a point, Refers to similarity tolerance, Indicates that all conditions are met the number of Step S3.4 Calculation The corresponding formula is as follows: in, Then The corresponding average value; Step S3.5 Increase the dimension to Repeat steps 3.1-3.4 to count the number of The number of satisfy , and calculate the corresponding average value ; Step S3.6 calculates the sample entropy value of the given continuous signal: in, express The sample entropy corresponding to the dimension vector; Step S3.7 sets a threshold interval according to the calculated sample entropy value, and divides the sub-signals into three groups of high complexity, medium complexity and low complexity according to the sample entropy values, thus completing signal reconstruction.

4. The method for predicting wind speed along a high-speed railway line based on secondary decomposition and reconstruction according to claim 1 is characterized in that: The step S5 uses a TCGRU network to predict the MIMF and LIMF signals respectively; the TCGRU network is composed of TCN and GRU, and can simultaneously extract the short-term fluctuations and long-term dependencies of the wind speed data, and also uses the TCGRU network to predict the subsequence obtained by the secondary decomposition of the HIMF; After that, obtain the prediction results of each sub-signal, add them up and output the final prediction value. The calculation formula is: , in, Refers to the first The prediction result corresponding to the sub-signal is and They refer to the predicted values ​​corresponding to the reconstructed MIMF and LIMF signals respectively.

Images