Short-wave communication highest available frequency prediction method based on same-ground multi-source detection data fusion

Abstract

The invention discloses a method for predicting the highest available frequency of short-wave communication based on same-ground multi-source detection data fusion, which comprises the following steps of: A, preprocessing data and acquiring the data; b, correlation analysis: performing correlation analysis on the obtained data; c, calculating various information entropies; step D, the predictability of the sequence is solved, the predictability of the sequence is solved by using different entropies, and the effectiveness of the quantification method is more intuitively observed through a predictability boundary; and step E, completing the MUF prediction task. According to the short wave available frequency prediction method based on multi-source data fusion provided by the invention, the correlation among the multi-source data of each station is evaluated so as to determine the correlation change among the multi-source data and the differential time-varying property of the correlation change, and the prediction requirement of the multi-source data fusion is proposed. According to the prediction method provided by the invention, the model is complete, the physical significance is clear, and the accuracy of short-wave highest available frequency spatio-temporal reasoning is effectively improved.

Description

Technical Field

[0001] The invention belongs to the technical field of shortwave highest available frequency prediction, and in particular relates to a method for predicting the highest available frequency of shortwave communication by fusing co-local multi-source detection data. Background Art

[0002] Shortwave (HF) communication can achieve long-distance transmission without relying on fixed facilities through the ionosphere, and has been widely used in commercial and military fields. According to the recommendations of the International Telecommunication Union (ITU), the maximum usable frequency (MUF) is the key factor in determining the appropriate operating frequency of shortwave communication. Accurate prediction of MUF can speed up the frequency selection process, enhance transmission capacity and improve communication quality. However, due to the time-varying characteristics of the ionosphere, predicting the available propagation frequency is still a key factor restricting the improvement of shortwave communication performance. Existing work mainly focuses on the single variable prediction of foF2 or MUF. However, a single detection element cannot fully reflect the changes in the ionosphere, and the noise and error of a single detection element will affect the prediction, which cannot meet the shortwave communication needs in unfamiliar areas. At the same time, the ionosphere is jointly affected by the physical quantities of multiple existing corresponding observation elements, which characterize the change process of the ionosphere from different angles. Therefore, using the fusion of multiple ionospheric data to predict the MUF of shortwave propagation can help improve the prediction accuracy and alleviate the problems of resource waste, poor channel quality, and long link establishment time caused by blind frequency selection.

[0003] Currently, a variety of ionospheric models have been developed to describe and predict the characteristics of the ionosphere, such as the International Reference Ionosphere (IRI) model, which provides a standardized method to estimate ionospheric parameters at different altitudes and conditions. However, these models are usually based on average statistical parameters and may not accurately reflect the short-term dynamic changes of the ionosphere to meet the frequency selection requirements of shortwave communications. In order to improve the prediction accuracy, researchers began to explore the possibility of multi-source data fusion, combining GNSS observations, ground plummets and other means to explore the relationship between multi-source data to enhance the temporal and spatial resolution and accuracy of the model.

[0004] Data fusion algorithms play a key role in integrating data from different sources. These algorithms are able to process and fuse various observation data, such as GNSS, plummet and remote sensing satellite data, to produce a more accurate ionospheric model.

[0005] For example, Kalman filter and its variants, such as extended Kalman filter, spherical harmonic model and ionosphere-thermosphere coupled physical model. In ionospheric modeling, these algorithms can evolve the law of ionospheric state change and effectively integrate multi-source observation data, thereby improving the accuracy and reliability of ionospheric parameter prediction such as MUF.

[0006] Although the existing models and algorithms have made significant progress in ionospheric research, they still face the challenges of nonlinear and multi-scale characteristics of the ionosphere and fine-grained accurate prediction. Deep learning provides an effective solution to this challenge with its remarkable nonlinear fitting ability. However, there are still the following limitations in applying deep learning to ionospheric multi-source data fusion. On the one hand, the difficulty in characterizing the relationship between different data sources requires the robustness of the data fusion algorithm. On the other hand, the existing deep learning models are still insufficiently interpretable when applied to the ionosphere. Therefore, it is very important to deeply mine the ionospheric multi-source data and use interpretable methods to fuse multi-source data based on deep learning models to improve the prediction performance. Therefore, how to analyze and utilize the elastic correlation between multi-source data in the ionospheric multi-source multi-variable fusion prediction modeling, and then realize the interpretable spatiotemporal reasoning and fusion prediction of multi-source ionospheric parameters based on deep learning, has become a key issue to improve the prediction performance of the shortwave highest available frequency under the sparse coverage of multi-source heterogeneous detection equipment. Summary of the invention

[0007] The present invention aims to solve the problem raised in the background technology and proposes a method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location.

[0008] In order to achieve the purpose of the present invention, the present invention discloses a method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location, comprising the following steps:

[0009] Step A: Data preprocessing; obtaining multi-source data;

[0010] Step B, correlation analysis: performing correlation analysis on the highest available frequency MUF, total electron content TEC, geomagnetic activity index Kp and geomagnetic storm ring current index Dst of the acquired data respectively;

[0011] Step C, calculating multiple information entropies; calculating the information entropies of multiple combinations of MUF, TEC, Kp and Dst data sequences;

[0012] Step D: Obtain the predictability of the sequence; the predictability of the sequence is obtained using different entropies, and the effectiveness of the quantification method is more intuitively observed through the predictability boundary;

[0013] Step E, completing the MUF prediction task; considering the correlation between multi-source data to complete the MUF prediction task.

[0014] Furthermore, step A specifically includes the following sub-steps:

[0015] Step A1, data acquisition: obtain the propagation factor M(3000)F2, vertical detection critical frequency foF2, total electron content TEC (Total Electron Content), geomagnetic activity index Kp (Kp index) and geomagnetic storm ring current index Dst (Disturbance Storm Time) of the ionosphere F2 layer at 3000 km at different locations

[0016] Step A2, calculate F2(3000)MUF, the specific formula is F2(3000)MUF=M(3000)F2×foF2.

[0017] Furthermore, step B specifically includes the following decomposition steps:

[0018] Step B1, correlation analysis between MUF and TEC data;

[0019] Step B2, performing correlation analysis between MUF and Dst index;

[0020] Step B3, performing correlation analysis between MUF and Kp index;

[0021] Step B4: Correlation analysis based on monthly median offset.

[0022] Furthermore, the correlation analysis uses the Pearson coefficient as a metric for correlation analysis between variables;

[0023] The Pearson correlation coefficient between two variables X and Y is defined as the covariance of the two variables divided by the product of their standard deviations:

[0024]

[0025] Among them, ρ is the global correlation coefficient; when actually measuring the correlation between variables, the covariance and standard deviation of the estimated sample are used to obtain the sample correlation coefficient r:

[0026]

[0027] Where, X i , Y i Represent the i-th sample of two different variables, Represent the mean of the two samples respectively.

[0028] Furthermore, step B1 specifically includes the following decomposition steps:

[0029] Step B11, calculating the overall correlation coefficient and 24-hour local correlation coefficient of TEC and MUF at different locations throughout the year;

[0030] Step B12, performing time series decomposition of TEC and MUF data based on an additive model, and performing correlation analysis on the decomposed series components respectively;

[0031] Step B4 specifically includes the following decomposition steps:

[0032] Step B41, observing and analyzing the correlation between the monthly median offsets of MUF and TEC and Kp and Dst;

[0033] Step B42: observe and analyze the correlation between the monthly median offset of MUF and the monthly median offset of TEC in different regions and the Kp index.

[0034] Furthermore, step C specifically includes the following sub-steps:

[0035] Step C1, using the sample entropy SE (Sample Entropy) method to calculate and analyze the complexity of the univariate sequence;

[0036] Step C2, using the multivariate sample entropy (MSE) method to calculate and analyze the complexity of the multivariate sequence;

[0037] Step C3: Use the multivariate constrained sample entropy McSE (Multivariate Constraint Sample Entropy) method to reflect the intrinsic characteristics of multivariate time series, measure the upper limit of the predictability of multivariate time series, and improve the performance of the prediction algorithm, providing guidance for the selection and design of the prediction model; given N variable time series and the correlation graph G. McSE uses the personalized page ranking PPR (Personalized Page Rank) method to find the variables that have the greatest impact on a certain variable i and remove the variables without influence to reduce the computational consumption and improve the performance.

[0038] Furthermore, step C2 specifically includes the following decomposition steps:

[0039] Step C21: for p measurement functions h k (y i ) The observed time series of p variables Multivariate embedding reconstruction based on a composite delay vector:

[0040]

[0041] in, is the embedding vector, τ=[τ 1 ,τ 2 ,…,τ p] is the time delay vector, the composite delay vector in Based on step C21, (Nn) composite delay vectors are formed where i = 1, 2, ..., Mn and n = max{M} × max{τ};

[0042] Step C22: transform any two vectors X m (i) and X m The distance between (j) is defined as:

[0043] d[X m (i),X m (j)]=max l＝1,…,m {|x(i+l-1)-x(j+l-1)|}

[0044] Step C23: For a given composite delay vector X m (i) and threshold r, calculate d[X m (i),X m (j)]≤r,j≠i i , and then calculate the frequency of occurrence:

[0045]

[0046] And define:

[0047]

[0048] Step C24, expand the dimension of the multivariate delay vector from m to (m+1); this is achieved in p different ways, and a total of There are p×(Nn) vectors X in m+1 (i), where X m+1 (i) indicates that the embedding dimension of a specific variable k increases from m k Increase to (m k +1) any embedding vector;

[0049] Step C25: For a given X m+1 (i), calculate d[X m+1 (i),X m+1 vector Q of [j]≤r i , where j≠i, and then calculate the frequency of occurrence:

[0050]

[0051] And define:

[0052]

[0053] Step C26, for a given similarity tolerance level r, the estimated multivariate sample entropy is:

[0054]

[0055] Furthermore, step C3 specifically includes the following decomposition steps:

[0056] Step C31: For the i-th time series Find the maximum connected graph Let G' contain variable i; on graph G', calculate PPR p i ;

[0057] PPR Vector is defined as follows:

[0058] p=α·M·p+(1-α)·s

[0059] Among them, α is a damping factor, that is, the probability of random walk following the Markov process is α, and the probability of jumping to the restart node is 1-α, and α is usually set to 0.85; the vector Indicates the restarted node; for example, s = (0, 1, 1) indicates that the restarted nodes are the second and third nodes; M is the transition probability matrix generated according to the graph G;

[0060]

[0061] Where nbr(i) is the set of adjacent nodes of node i in graph G; M ij It is the ratio of the edge weight from node i to node j to the sum of the edge weights from node i to all adjacent nodes;

[0062] Step C32: (N-1) PPR values ​​p where 1≤j≠i≤N i In (j), select the first k largest PPR values ​​and use S i Indicates related variables, |S i |=k; Generate a new multivariate time series So that every variable j∈X' i All belong to S i gather;

[0063] Step C33: According to X' i , before calculating McSE, for d[X m+1 (i),X m+1 (j)]≤r Redefine the condition:

[0064] |X m (i)[l]-X m (j)[l]|≤r+r l l=1,2,…,m

[0065]

[0066] By introducing X m The sub-item r of the lth element in (i) l , α represents X m (i) The variable to which [l] belongs, enlarges the bounding box W by setting a larger size for the dimension corresponding to the more important variable in variable i i ;

[0067] W i =[X m (i)[1]-(r+r 1 ),X m (i)[1]+(r+r 1 )]×[X m (i)[2]-(r+r 2 ),X m (i)[2]+(r+r 2 )]…×[X m (i)[m]-(r+r m ),X m (i)[m]+(r+r m )]

[0068] Similarly, calculate Φ in multivariate sample entropy i,m (r) and Φ i,m+N (r); The multivariate constrained sample entropy of each variable i is defined as:

[0069]

[0070] Finally, the overall multivariate constrained sample entropy is:

[0071]

[0072] Among them, the adjacency matrix required to measure the transfer probability matrix is ​​generated using the Pearson coefficient and the transfer entropy TE (Transfer Entropy);

[0073] For transfer entropy, take X as source and Y as target, As a background process with the hope as condition; the transfer entropy is defined using time-local variables:

[0074]

[0075] Taking the average over time, we further get:

[0076]

[0077] Further, in step D, different upper bounds of predictability Π are calculated from different entropies max ; Based on Fano inequality, the upper bound of predictability Π max The following equation can be solved:

[0078] H=-[Πlog 2 Π+(1-Π)log 2 (1-Π)]+(1-Π)log 2 (M-1)

[0079] Where M is the number of state symbols and H is the entropy of the selection.

[0080] Furthermore, in step E, according to the results of step C and step D, a Transformer model is selected to introduce correlation between data into the original multi-source data to complete the task of predicting MUF.

[0081] Compared with the prior art, the significant progress of the present invention is that: the present invention provides a shortwave broadcast frequency scheduling method based on multi-source data processing, which is scientifically designed and mines the correlation and statistical characteristics between the real-world MUF and the multi-source ionospheric data TEC, Kp, and Dst. By using the Pearson coefficient to evaluate the correlation between the multi-source data of each station, the correlation change between the multi-source data with latitude and the differentiated time-varying nature of the correlation change are determined, and the prediction requirements for multi-source data fusion are proposed. The multi-source data fusion prediction experiment was carried out on the most advanced deep learning prediction model at present, and the effectiveness of the proposed capability requirements was verified. The complexity of the multi-source data fusion sequence of different combinations was calculated. Through multivariate sample entropy analysis, the entropy sequences of different combinations were observed, and it was found that the time-varying nature of the correlation change in the short term will lead to an increase in the complexity of the sequence, thereby affecting the accuracy of nonlinear fitting. A prediction scheme based on cross-variable time-varying constraints is proposed, with a complete model and clear physical meaning, which effectively improves the accuracy of shortwave highest available frequency spatiotemporal reasoning.

[0082] In order to more clearly illustrate the functional characteristics and structural parameters of the present invention, further description is given below in conjunction with the accompanying drawings and specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0084] Figure 1 A flowchart of a method for predicting the highest available frequency of shortwave communication by fusing co-located multi-source detection data provided by the present invention;

[0085] Figure 2A schematic diagram of the fusion prediction of multi-source heterogeneous detection data of the ionosphere for the integration of space and earth provided by the present invention;

[0086] Figure 3 Schematic diagram of the ionospheric multi-source data fusion prediction solution provided by the present invention. DETAILED DESCRIPTION

[0087] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0088] See also Figures 1 to 3 ,like Figure 2 As shown in , consider a shortwave communication link scenario that includes multiple space-to-ground ionospheric detection methods such as ionospheric vertical detection stations, satellite detection, and the Earth's magnetic field, as shown in Figure 3 As shown in the figure, a multi-source data fusion prediction scheme is given under the condition of multi-source detection in the same place. Figure 1 As shown, the present invention provides a method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location, comprising the following steps:

[0089] Step A, data preprocessing: obtaining multi-source data;

[0090] Step B, correlation analysis: perform correlation analysis on the acquired data: MUF, TEC, Kp and Dst respectively;

[0091] Step C, calculating multiple information entropies: calculating the information entropies of multiple combinations of MUF, TEC, Kp and Dst data sequences;

[0092] Step D, calculate the predictability of the sequence: calculate the predictability of the sequence using different entropies, and observe the effectiveness of the quantification method more intuitively through the predictability boundary;

[0093] Step E, completing the MUF prediction task: considering the correlation between multi-source data to complete the MUF prediction task. In the present invention, in specific implementation, step A specifically includes the following sub-steps:

[0094] Step A1, data acquisition: obtain the propagation factor M(3000)F2, vertical detection critical frequency foF2, total electron content (TEC), geomagnetic activity index (Kp index) and geomagnetic storm ring current index (Dst) of the ionosphere F2 layer at 3000 km at different locations

[0095] Step A2, calculate F2(3000)MUF, the specific formula is as follows:

[0096] F2(3000)MUF=M(3000)F2×foF2, formula (1);

[0097] Step B: In the correlation analysis of this step, the Pearson coefficient is mainly used as a measurement indicator for the correlation analysis between variables.

[0098] The Pearson correlation coefficient between two variables X and Y is defined as the covariance of the two variables divided by the product of their standard deviations:

[0099]

[0100] Among them, the global correlation coefficient is often represented by the Greek lowercase letter ρ. When actually measuring the correlation between variables, the covariance and standard deviation of the estimated sample are mainly used to obtain the sample correlation coefficient, which is often represented by r:

[0101]

[0102] Where, X i , Y i Represent the i-th sample of two different variables, Represent the mean of the two samples respectively.

[0103] In the present invention, in specific implementation, step B specifically includes the following sub-steps:

[0104] Step B1, correlation analysis between MUF and TEC data;

[0105] In the present invention, step B1 specifically includes the following decomposition steps:

[0106] Step B11, calculate the overall correlation coefficient and 24-hour local correlation coefficient of TEC and MUF at different locations throughout the year

[0107] Step B12, performing time series decomposition of the TEC and MUF data based on the additive model, and performing correlation analysis on the decomposed series components respectively.

[0108] Step B2, performing correlation analysis between MUF and Dst index;

[0109] Step B3, performing correlation analysis between MUF and Kp index;

[0110] Step B4, correlation analysis based on monthly median offset:

[0111] In the present invention, step B4 specifically includes the following decomposition steps:

[0112] Step B41, observe and analyze the correlation between the monthly median offsets of MUF and TEC and Kp and Dst.

[0113] Step B42, observe and analyze the correlation between the monthly median offset of MUF and the monthly median offset of TEC in different regions and the Kp index.

[0114] In the present invention, in specific implementation, step C specifically includes the following sub-steps:

[0115] Step C1, using the sample entropy (SE) method to calculate and analyze the complexity of the univariate sequence;

[0116] In the present invention, step C1 specifically includes the following decomposition steps:

[0117] Step C11, combine the sequences into m-dimensional vectors in order, as shown below:

[0118]

[0119] Step C12, define the distance between any vectors X(i) and x(j) as:

[0120]

[0121] And calculate the distance between any vector and any other vector:

[0122] d[X(i),X(j)].

[0123] Step C13, according to the similarity tolerance r (r>0), calculate the number of d[X(i),X(j)]<r for each vector i, and make a ratio with the total number of vectors Nm, recorded as It can be expressed as:

[0124]

[0125] Step C14, increase the dimension by 1, that is, m+1, and repeat the previous steps to obtain B m+1 (r). Therefore, the sample entropy can be obtained as follows:

[0126] SampEn(m,r,N)=-ln[B m+1 (r) / B m (r)]

[0127] The present invention selects r=STD as the similarity tolerance, wherein STD is the standard deviation of the original sequence.

[0128] Step C2, using the Multivariate Sample Entropy (MSE) method to calculate and analyze the complexity of the multivariate sequence;

[0129] In the present invention, step C2 specifically includes the following decomposition steps:

[0130] Step C21, for p measurement functions hk(y i ) The observed time series of p variables Multivariate embedding reconstruction based on a composite delay vector:

[0131]

[0132] in, is the embedding vector, τ=[τ 1 ,τ 2 ,…,τ p ] is the time delay vector, the composite delay vector in Based on the above theory, (Nn) composite delay vectors are formed Where i = 1, 2,…, Nn and n = max{M}×max{τ}.

[0133] Step C22, any two vectors X m (i) and X m The distance between (j) is defined as:

[0134] d[X m (i),X m (j)]=max l＝1,…,m {|x(i+l-1)-x(j+l-1)|}.

[0135] Step C23, for a given composite delay vector X m (i) and threshold r, calculate d[X m (i),X m (j)]≤r,j≠i i , and then calculate the frequency of occurrence:

[0136]

[0137] And define:

[0138]

[0139] Step C24, expand the dimension of the multivariate delay vector from m to (m+1). This can be achieved in p different ways, and a total of There are p×(Nn) vectors X inm+1 (i), where X m+1 (i) indicates that the embedding dimension of a specific variable k increases from m k Increase to (m k +1).

[0140] Step C25, for a given X m+1 (i), calculate d[X m+1 (i),X m+1 vector Q of [j]≤r i , where j≠i, and then calculate the frequency of occurrence:

[0141]

[0142] And define:

[0143]

[0144] Step C26, for a given similarity tolerance level r, the estimated multivariate sample entropy is:

[0145]

[0146] Step C3 uses the Multivariate Constraint Sample Entropy (McSE) method to reflect the intrinsic characteristics of multivariate time series, measure the upper limit of the predictability of multivariate time series, and improve the performance of the prediction algorithm, providing guidance for the selection and design of the prediction model. and the correlation graph G. McSE uses the Personalized Page Rank (PPR) method to find out the variables that are important to a certain variable i and remove unimportant variables to reduce computational consumption and improve performance.

[0147] In the present invention, step C3 specifically includes the following decomposition steps:

[0148] Step C31, for the i-th time series Find the maximum connected graph Let G' contain variable i. On graph G', PPR p can be calculated i .

[0149] PPR Vector The definition of

[293] :

[0150] p=α·M·p+(1-α)·s

[0151] Among them, α is a damping factor, that is, the probability of random walk following the Markov process is α, and the probability of jumping to the restart node is 1-α, which is usually set to 0.85. The vector Indicates the restarted node. For example, s = (0, 1, 1) indicates that the restarted nodes are the second and third nodes. M is the transition probability matrix generated according to the graph G.

[0152]

[0153] Where nbr(i) is the set of adjacent nodes of node i in graph G. ij It is the ratio of the edge weight from node i to node j to the sum of the edge weights from node i to all adjacent nodes.

[0154] Step C32, in the (N-1) PPR values ​​p where 1≤j≠i≤N i In (j), select the first k largest PPR values ​​and use S i Indicates related variables, |S i |=k. Generate a new multivariate time series So that every variable j∈X' i All belong to S i gather.

[0155] Step C33, according to X' i , before calculating McSE, for d[X m+1 (i),X m+1 (j)]≤r Redefine the condition:

[0156] |X m (i)[l]-X m (j)[l]|≤r+r l l=1,2,…,m

[0157]

[0158] By introducing X m The sub-item r of the lth element in (i) l , α represents X m (i) The variable to which [l] belongs, enlarges the bounding box W by setting a larger size for the dimension corresponding to the more important variable in variable i i .

[0159] W i =[X m (i)[1]-(r+r 1 ),X m (i)[1]+(r+r 1 )]×[X m (i)[2]-(r+r2 ),X m (i)[2]+(r+r 2 )]…×[X m (i)[m]-(r+r m ),X m (i)[m]+(r+r m )]

[0160] Similarly, we can calculate the Φ in the multivariate sample entropy i,m (r) and Φ i,m+N (r). Multivariate constraints for each variable i

[0161] The sample entropy is defined as:

[0162]

[0163] Finally, the overall multivariate constrained sample entropy is:

[0164]

[0165] Among them, the adjacency matrix required to measure the transfer probability matrix is ​​generated using the Pearson coefficient and transfer entropy (TE).

[0166] For transfer entropy, take X as source and Y as target, As a background process, we want to condition the transfer entropy. The transfer entropy is defined in terms of time-local variables:

[0167]

[0168] By averaging over time, we can further obtain:

[0169]

[0170] Step D, different upper bounds of predictability Π can be calculated from different entropies max Based on Fano's inequality, the upper bound of predictability is Π max The following equation can be solved:

[0171] H=-[Πlog 2 Π+(1-Π)log 2 (1-Π)]+(1-Π)log 2 (M-1)

[0172] Where M is the number of state symbols and H is the entropy of the selection.

[0173] Step E: Based on the results of Step C and Step D, select the Transformer model to introduce data correlation into the original multi-source data to complete the task of predicting MUF.

[0174] In summary, compared with the prior art, the shortwave broadcast frequency scheduling method based on multi-source data processing provided by the present invention is scientifically designed and mines the correlation and statistical characteristics between the real-world MUF and the multi-source ionospheric data TEC, Kp, and Dst. By using the Pearson coefficient to evaluate the correlation between the multi-source data of each station, the correlation change between the multi-source data with latitude and the differentiated time-varying nature of the correlation change are determined, and the prediction requirements for multi-source data fusion are proposed. The multi-source data fusion prediction experiment was carried out on the most advanced deep learning prediction model to verify the effectiveness of the proposed capability requirements. The complexity of the multi-source data fusion sequence of different combinations was calculated. Through multivariate sample entropy analysis, the entropy sequences of different combinations were observed, and it was found that the time-varying nature of the correlation change in the short term will lead to an increase in the complexity of the sequence, thereby affecting the accuracy of nonlinear fitting. A prediction scheme based on cross-variable time-varying constraints is proposed, with a complete model and clear physical meaning, which effectively improves the accuracy of shortwave highest available frequency spatiotemporal reasoning.

[0175] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0176] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location, characterized in that: The following steps are involved: Step A: Data preprocessing; obtaining multi-source data; Step B, correlation analysis: performing correlation analysis on the highest available frequency MUF, total electron content TEC, geomagnetic activity index Kp and geomagnetic storm ring current index Dst of the acquired data respectively; Step C, calculating multiple information entropies; calculating the information entropies of multiple combinations of MUF, TEC, Kp and Dst data sequences; Step D: Obtain the predictability of the sequence; the predictability of the sequence is obtained using different entropies, and the effectiveness of the quantification method is more intuitively observed through the predictability boundary; Step E, completing the MUF prediction task; considering the correlation between multi-source data to complete the MUF prediction task.

2. According to claim 1, a method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location, characterized in that: Step A specifically includes the following decomposition steps: Step A1, data acquisition: obtaining the propagation factor M(3000)F2, the vertical detection critical frequency foF2, the total electron content TEC, the geomagnetic activity index Kp and the geomagnetic storm ring current index Dst of the ionosphere F2 layer at 3000 kilometers at different locations; Step A2, calculate F2(3000)MUF, the specific formula is F2(3000)MUF=M(3000)F2×foF2.

3. The method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location according to claim 1 is characterized in that: Step B specifically includes the following decomposition steps: Step B1, correlation analysis between MUF and TEC data; Step B2, performing correlation analysis between MUF and Dst index; Step B3, performing correlation analysis between MUF and Kp index; Step B4: Correlation analysis based on monthly median offset.

4. The method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location according to claim 3 is characterized in that: The correlation analysis uses the Pearson coefficient as a metric for correlation analysis between variables; The Pearson correlation coefficient between two variables X and Y is defined as the covariance of the two variables divided by the product of their standard deviations: Among them, ρ is the global correlation coefficient; when actually measuring the correlation between variables, the covariance and standard deviation of the estimated sample are used to obtain the sample correlation coefficient r: In the formula, X i , Y i Represent the i-th sample of two different variables, Represent the mean of the two samples respectively.

5. The method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location according to claim 3 is characterized in that: Step B1 specifically includes the following decomposition steps: Step B11, calculating the overall correlation coefficient and 24-hour local correlation coefficient of TEC and MUF at different locations throughout the year; Step B12, performing time series decomposition of TEC and MUF data based on an additive model, and performing correlation analysis on the decomposed series components respectively; Step B4 specifically includes the following decomposition steps: Step B41, observing and analyzing the correlation between the monthly median offsets of MUF and TEC and Kp and Dst; Step B42: observe and analyze the correlation between the monthly median offset of MUF and the monthly median offset of TEC in different regions and the Kp index.

6. The method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location according to claim 1 is characterized in that: Step C specifically includes the following decomposition steps: Step C1, using the sample entropy SE method to calculate and analyze the complexity of the univariate sequence; Step C2, using the multivariate sample entropy MSE method to calculate and analyze the complexity of the multivariate sequence; Step C3: Use the multivariate constrained sample entropy McSE method to reflect the intrinsic characteristics of multivariate time series, measure the upper limit of the predictability of multivariate time series, and improve the performance of the prediction algorithm, providing guidance for the selection and design of the prediction model; Given a time series of N variables and the correlation graph G. McSE uses the personalized page ranking PPR method to find the variables that have the greatest impact on a certain variable i and remove the variables without impact.

7. The method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location according to claim 6 is characterized in that: Step C2 specifically includes the following decomposition steps: Step C21: for p measurement functions h k (y i ) The observed time series of p variables Multivariate embedding reconstruction based on a composite delay vector: in, is the embedding vector, τ=[τ1,τ2,…,τ p ] is the time delay vector, the composite delay vector in Based on step C21, (Nn) composite delay vectors are formed where i = 1, 2, ..., Nn and n = max{M} × max{τ}; Step C22: transform any two vectors X m (i) and X m The distance between (j) is defined as: d[X m (i),X m (j)]=max l＝1,…,m {|x(i+l-1)-x(j+l-1)|} Step C23: For a given composite delay vector X m (i) and threshold r, calculate d[X m (i),X m (j)]≤r,j≠i i , and then calculate the frequency of occurrence: And define: Step C24, expand the dimension of the multivariate delay vector from m to (m+1); this is achieved in p different ways, and a total of There are p×(Nn) vectors X in m+1 (i), where X m+1 (i) indicates that the embedding dimension of a specific variable k increases from m k Increase to (m k +1) any embedding vector; Step C25: For a given X m+1 (i), calculate d[X m+1 (i),X m+1 vector Q of [j]≤r i , where j≠i, and then calculate the frequency of occurrence: And define: Step C26, for a given similarity tolerance level r, the estimated multivariate sample entropy is:

8. The method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location according to claim 6 is characterized in that: Step C3 specifically includes the following decomposition steps: Step C31: For the i-th time series Find the maximum connected graph Let G' contain variable i; on graph G', calculate PPR p i ; PPR Vector is defined as follows: p=α·M·p+(1-α)·s Among them, α is a damping factor, that is, the probability of random walk following the Markov process is α, and the probability of jumping to the restart node is 1-α; the vector Indicates the node where the restart occurs; M is the transition probability matrix generated based on the graph G; Where nbr(i) is the set of adjacent nodes of node i in graph G; M ij It is the ratio of the edge weight from node i to node j to the sum of the edge weights from node i to all adjacent nodes; Step C32: (N-1) PPR values ​​p where 1≤j≠i≤N i In (j), select the first k largest PPR values ​​and use S i Indicates related variables, |S i |=k; Generate a new multivariate time series So that every variable j∈X' i All belong to S i gather; Step C33: According to X' i , before calculating McSE, for d[X m+1 (i),X m+1 (j)]≤r Redefine the condition: |X m (i)[l]-X m (j)[l]|≤r+r l l=1,2,…,m By introducing X m The sub-item r of the lth element in (i) l , α represents X m (i) The variable to which [l] belongs, enlarges the bounding box W by setting a larger size for the dimension corresponding to the more important variable in variable i i ; W i =[X m (i)[1]-(r+r1),X m (i)[1]+(r+r1)]×[X m (i)[2]-(r+r2),X m (i)[2]+(r+r2)]…×[X m (i)[m]-(r+r m ),X m (i)[m]+(r+r m )] Similarly, calculate Φ in multivariate sample entropy i,m (r) and Φ i,m+N (r); The multivariate constrained sample entropy of each variable i is defined as: Finally, the overall multivariate constrained sample entropy is: Among them, the adjacency matrix required to measure the transition probability matrix is ​​generated using the Pearson coefficient and the transfer entropy TE; For transfer entropy, take X as source and Y as target, As a background process with the hope as condition; the transfer entropy is defined using time-local variables: Taking the average over time, we further get:

9. The method for predicting the highest available frequency of shortwave communication by fusing multi-source detection data at the same location according to claim 1 is characterized in that: Step D, calculate different upper bounds of predictability Π from different entropies max ; Based on Fano inequality, the upper bound of predictability Π max The following equation can be solved: H=-[Πlog2Π+(1-Π)log2(1-Π)]+(1-Π)log2(M-1) Where M is the number of state symbols and H is the entropy of the selection.

10. The method for predicting the highest available frequency of shortwave communication by fusing co-located multi-source detection data according to claim 1, characterized in that: In step E, based on the results of step C and step D, a Transformer model is selected to introduce data correlation into the original multi-source data to complete the task of predicting MUF.

Images