CN112616160B - Intelligent short-wave frequency cross-frequency-band real-time prediction method and system - Google Patents
Intelligent short-wave frequency cross-frequency-band real-time prediction method and system Download PDFInfo
- Publication number
- CN112616160B CN112616160B CN202011449836.1A CN202011449836A CN112616160B CN 112616160 B CN112616160 B CN 112616160B CN 202011449836 A CN202011449836 A CN 202011449836A CN 112616160 B CN112616160 B CN 112616160B
- Authority
- CN
- China
- Prior art keywords
- frequency
- short
- wave
- data
- communication
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/10—Scheduling measurement reports ; Arrangements for measurement reports
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/08—Testing, supervising or monitoring using real traffic
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an intelligent short-wave frequency cross-frequency-band real-time prediction method and system. The method comprises the following steps: training a prediction model based on a neural network by using historical sampling data, wherein the prediction model is used for predicting the communication quality of other frequencies according to frequency characteristic information of a certain frequency; determining the range of the next available frequency after the short-wave communication is interrupted according to the correlation among different frequencies; acquiring frequency characteristic information at the short-wave communication interruption moment; inputting frequency characteristic information of short-wave communication interruption time into the prediction model, and acquiring communication quality of each frequency in the range; and selecting the frequency with the best communication quality in the range as the next accessible frequency selected after the short-wave communication is interrupted. According to the method, other frequency points with high link establishment reliability near the communication interruption frequency point are used as the prediction target, so that the training complexity of the neural network facing a large amount of nonlinear data can be reduced, and the prediction target is more combined with the operation requirement of the actually-working equipment.
Description
Technical Field
The invention belongs to the technical field of wireless communication, and particularly relates to an intelligent short-wave frequency cross-frequency-band real-time prediction method and system.
Background
In actual short-wave communication, a communication link is often suddenly interrupted due to the unstable influence of a short-wave channel, the selection of the next frequency with good communication effect depends on experience to a great extent, the instantaneity and the reliability cannot be guaranteed, and the processing method is limited in the face of sudden short-wave communication interruption.
The frequency prediction method based on the neural network can predict the short wave frequency in real time by means of the nonlinear approximation capability of the neural network. However, in actual use, the existing method often needs to acquire full-frequency-band communication data, cannot perform cross-frequency-band prediction based on used frequency-band data, and is not strong in practicability. Meanwhile, under the influence of changeful short wave channels, the communication data of the same frequency band obtained by tests also changes violently in a short time, the data obtained by single sampling is not representative, and the reliability of training data is not high.
Disclosure of Invention
Aiming at least one defect or improvement requirement in the prior art, the invention provides an intelligent short-wave frequency cross-frequency-band real-time prediction method and system, which take other frequency points with high link establishment reliability near a communication interruption frequency point as prediction targets, can reduce the training complexity of a neural network facing a large amount of nonlinear data, and the prediction targets are more combined with the equipment operation requirements of actual work.
In order to achieve the above object, according to a first aspect of the present invention, there is provided an intelligent short-wave frequency cross-band real-time prediction method, including the steps of:
acquiring historical sampling data of short wave communication, and training a prediction model based on a neural network by using the historical sampling data, wherein the prediction model is used for predicting the communication quality of other frequencies according to frequency characteristic information of a certain frequency;
determining the range of the next available frequency after the short-wave communication is interrupted according to the correlation among different frequencies;
acquiring frequency characteristic information of short-wave communication interruption time;
inputting frequency characteristic information of short-wave communication interruption time into the prediction model, and acquiring communication quality of each frequency in the range;
and selecting the frequency with the best communication quality in the range as the next accessible frequency selected after the short-wave communication is interrupted.
Preferably, the determining the range of the next available frequency after the short-wave communication interruption according to the correlation between different frequencies is specifically:
calculating a correlation coefficient r between different frequencies by adopting a Pearson correlation coefficient method, wherein the calculation formula is as follows:
wherein, Xi,YiThe sample data is represented by a sample data,
denotes the sample mean, n denotes the number of samples, σx,σYRepresents the sample standard deviation;
and determining the range of the next available frequency after the short-wave communication is interrupted according to the correlation coefficient r.
Preferably, the frequency characteristic information at the time of interruption of the short-wave communication includes: the time of the short-wave communication interruption moment, the chain breakage frequency and the communication signal-to-noise ratio of different moments before interruption.
Preferably, before the prediction model is trained by using the historical sample data, the historical sample data is preprocessed, and the preprocessing includes the steps of:
performing data smoothing on the historical sampling data, wherein the data smoothing is to fill data with missing signal-to-noise ratio by adopting a signal-to-noise ratio mean value;
performing nonlinear unit conversion on the data subjected to the data smoothing processing, wherein the nonlinear unit conversion refers to the conversion of signal-to-noise ratio data with the unit of dB into the ratio of signal power to noise power;
and performing data enhancement on the data converted by the nonlinear unit to expand the training sample.
Preferably, the prediction model comprises a classification network and a regression network;
the classification network is used for predicting whether communication links under other frequencies can be communicated according to the frequency characteristic information of a certain frequency;
the regression network is used to predict a signal-to-noise ratio for a frequency over which the communication link may be passed.
Preferably, the classification network includes 1 input layer, 4 full-connection layers, and 1 output layer, the activation function of the classification network employs a sigmoid function, and the objective function of the classification network employs a cross entropy function:
wherein H (X) represents information entropy, P (x)i) Representing the probability predicted by the category.
Preferably, the regression network comprises 1 input layer, 5 fully-connected layers and one output layer, the activation function of the regression network adopts a ReLU function, and the adaptive learning rate of the regression network is 0.01-0.0005.
Preferably, the objective function of the regression network is:
wherein, ypreFor the network to predict value, yiIs a function of the actual value of the measured value,n is the number of samples.
Preferably, the range is a frequency within 2MHZ of the frequency at the time of interruption.
According to a second aspect of the present invention, there is provided an intelligent short-wave frequency cross-band real-time prediction system, comprising:
the device comprises a pre-training module, a neural network-based prediction model and a communication quality prediction module, wherein the pre-training module is used for acquiring historical sampling data of short-wave communication and training the prediction model based on the neural network by using the historical sampling data, and the prediction model is used for predicting the communication quality of other frequencies according to frequency characteristic information of a certain frequency;
the correlation calculation module is used for determining the range of the next available frequency after the short-wave communication is interrupted according to the correlation among different frequencies;
the interruption frequency characteristic information acquisition module is used for acquiring frequency characteristic information at the short-wave communication interruption moment;
the prediction module is used for inputting frequency characteristic information of short-wave communication interruption time into the prediction model and acquiring the communication quality of each frequency in the range;
and the output module is used for selecting the frequency with the best communication quality in the range as the next accessible frequency selected after the short-wave communication is interrupted.
In general, compared with the prior art, the invention has the following beneficial effects: on the basis of fitting the short wave frequency change rule by using the neural network, the traditional neural network learning target is changed by combining the actual frequency selection operation process, namely, the frequency selection of the full-frequency band in the whole period is not taken as the prediction target, but the frequency points with high reliability of link establishment nearby the communication interruption frequency point are taken as the prediction target according to the frequency selection pain point in the actual short wave communication process. When short-wave communication is interrupted, the best frequency point in the frequency points is directly selected as the next accessible frequency point, cross-frequency-band prediction is achieved, training complexity of a neural network facing a large amount of nonlinear data is reduced through the method, and the prediction target is more combined with the operation requirement of actual working equipment.
Drawings
FIG. 1 is a diagram of an application scenario provided by an embodiment of the present invention;
FIG. 2 is a schematic diagram of a method provided by an embodiment of the invention;
FIG. 3 is a diagram of data correlation verification results provided by an embodiment of the present invention;
FIG. 4 is a comparison graph of waveforms after nonlinear transformation of data provided by an embodiment of the present invention;
FIG. 5 is a block diagram of a prediction network provided by an embodiment of the present invention;
FIG. 6 is a schematic diagram of a neural network structure of a prediction module according to an embodiment of the present invention;
FIG. 7 is a graph illustrating training error trends of a prediction network according to an embodiment of the present invention;
fig. 8 is a comparison chart of partial predicted result waveforms according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Fig. 1 is an application scenario diagram of an intelligent short-wave frequency cross-frequency-band real-time prediction method provided by an embodiment of the present invention. For a commonly used neural network prediction method, in order to predict the communication quality of a certain frequency point, regression prediction needs to be performed according to historical data of the frequency point, as shown in fig. 1 (a); in the actual frequency optimization process, the communication quality of the full frequency band needs to be predicted, and the frequency point with the best communication quality is selected to finally obtain the predicted communication frequency. Using this method requires obtaining communication quality data of the full band in advance, as shown in fig. 1 (b); actual communication can not intentionally carry out data acquisition on communication quality of other frequency bands under the condition that the existing frequency can be used. The prediction lacks full band data. The main thought of the cross-frequency-band prediction method provided by the invention is as follows: and changing a neural network prediction mode, predicting the communication effect of other frequency bands by using the accessible frequency band historical data, and realizing cross-frequency band prediction. As shown in FIG. 1 (c); in the actual communication process, the communication effect of other frequency points near the interrupt frequency point can be used as a prediction target by mining the correlation between adjacent frequency points, so that the frequency band prediction range is reduced, the input information amount is increased, and the prediction precision is effectively improved. As shown in fig. 1 (d).
Fig. 2 is a schematic diagram illustrating a principle of an intelligent short-wave frequency cross-band real-time prediction method according to an embodiment of the present invention, where the method includes steps S1 to S5.
And S1, acquiring historical sampling data of the short wave communication, and training a prediction model based on a neural network by using the historical sampling data, wherein the prediction model is used for predicting the communication quality of other frequencies according to the frequency characteristic information of a certain frequency.
Only historical sampling data is collected during pre-training, and full-frequency-band communication quality data does not need to be collected during prediction at the subsequent short-wave communication interruption moment. However, the prediction method in the prior art generally does not need to collect the communication quality data of the full frequency band when performing prediction.
Frequency characteristic information of a plurality of sampling points is extracted from historical sampling data. The frequency characteristic information preferably includes communication time, communication frequency, and communication quality. The communication quality preferably uses signal-to-noise ratio data.
And performing regression training on the communication quality of other frequency points near the short wave frequency of each sample by the prediction function of the neural network. Specifically, frequency characteristic data is input into a neural network; obtaining a trained prediction network G through back propagation and gradual iteration; the prediction network G can fit the frequency-signal-to-noise ratio waveform relationship at different times.
And S2, determining the range of the next available frequency after the short-wave communication is interrupted according to the correlation among different frequencies, and recording the range as df.
Through the correlation calculation of each frequency point of the historical data and the frequency points of the full frequency band, other frequency points with strong correlation with the frequency characteristics at the communication interruption frequency points are searched, a frequency range is provided for predicting the next available frequency in the step S4, and the next available frequency does not need to be selected in the full frequency band.
And S3, acquiring frequency characteristic information of the short-wave communication interruption moment.
Only the frequency characteristic information of the short-wave communication interruption moment needs to be acquired as the input of the classification model. The frequency characteristic information of the short-wave communication interruption moment is the same as the type of the pre-training data, and preferably comprises the time of the short-wave communication interruption moment, the chain breakage frequency and the communication signal-to-noise ratio of different moments before interruption. Time when a preferred communication interruption occurs; the link-breaking frequency at the moment of communication interruption; the communication signal-to-noise ratio one hour before the communication interruption occurs; the communication signal-to-noise ratio two hours before the communication interruption occurs; the signal-to-noise ratio of the communication the day before the communication interruption occurred.
The frequency characteristic information at the time of short-wave communication interruption can change along with the ionosphere change at the time and place of communication interruption.
And S4, inputting the frequency characteristic information of the short-wave communication interruption time into the prediction model, and acquiring the communication quality of each frequency in the df range near the interruption frequency.
And inputting the frequency characteristic information of the short-wave communication interruption moment into the prediction model, so that the communication quality of other frequencies in the df range near the interruption frequency can be obtained.
And S5, selecting the frequency with the best communication quality in the df range near the interrupt frequency as the next accessible frequency selected after the short-wave communication is interrupted.
A preferred implementation of step S2 is described in detail below.
Fig. 3 is a data correlation thermodynamic diagram of an intelligent short-wave frequency cross-band real-time prediction method according to an embodiment of the present invention, where a frequency selection method according to an embodiment of the present invention is to determine a selection range of accessible frequencies near a communication break, and we find that next accessible frequencies are generally distributed in a certain rule near the communication break and have a certain correlation therebetween. The associated data has the possibility of being predictable, and the relevance verification is carried out on each frequency point of the existing data at the full-band frequency point to obtain the selection range of the accessible frequency points. The method comprises the following steps:
using the pearson correlation coefficient as a method of quantifying the correlation, the pearson correlation coefficient between two variables is defined as the quotient of the covariance and the standard deviation between the variables:
the pearson correlation coefficient for sample data can be derived from the equation:
wherein, Xi,YiThe sample data is represented by a sample data,
denotes the sample mean, n denotes the number of samples, σX,σYSample standard deviations are indicated.
Assuming that the correlation thermodynamic diagram is calculated by using the historical sampling data of one month, as a result, as shown in fig. 3, it can be seen that the range in which a certain frequency point in the data has a strong correlation with its nearby frequency point (the thermodynamic diagram is brighter) is about 20 sampling points, that is, the range of 2MHz near the frequency point. Further, it is inferred that the signal-to-noise ratio characteristic of a certain frequency point is highly correlated with data of nearby 2 MHz. Thus, it is preferable to use a frequency range of 2MHz around the interrupt frequency as an input for predicting the signal-to-noise ratio characteristic of the frequency bin. If a large amount of historical data is used as support, calculation can be performed all at once, and the range suitable for most conditions is obtained. If the data volume is limited, full-frequency-band full-time data (such as once a month) can be collected discontinuously to deal with the change of the ionosphere along with time, so that the prediction is more accurate.
A preferred implementation of step S1 is described in detail below.
Before training a prediction model by using historical sampling data, preprocessing the historical sampling data, wherein the preprocessing comprises the following steps:
and carrying out data smoothing processing on the historical sampling data. Under the influence of random fluctuation of an ionosphere channel, the communication signal-to-noise ratio under the same frequency can be changed drastically in a short time. Meanwhile, due to the limitation of a sampling method, the signal-to-noise ratio on a certain frequency is only acquired once, and the data reliability at a non-effective value is seriously influenced. And filling the suspected missing part of the signal-to-noise ratio in the data by using a signal-to-noise ratio mean value filling method.
And carrying out nonlinear unit conversion on the data subjected to the data smoothing processing. Fig. 4 is a waveform comparison diagram after nonlinear transformation is performed on data preprocessing of the intelligent short-wave frequency cross-frequency-band real-time prediction method provided by the embodiment of the invention. Since the signal-to-noise ratio of the data is in dB, the signal-to-noise ratio data where the signal is not passed is theoretically- ∞ dB. Waveform data with large gradient difference is not suitable for being predicted by using a neural network, so that signal-to-noise ratio data is converted into a proportional unit of Ps/Pn, namely a ratio of signal power to noise power. Meanwhile, the conversion formula exponential form is nonlinear conversion, so that the converted signal waveform is more advantageous in describing places with high signal-to-noise ratio, and the method is not conspirant with the prediction target of people.
And performing data enhancement (Mix-up) on the data after nonlinear unit conversion. Because the acquisition of the original data has sampling intervals and the frequency point characteristics in the intervals are not well described, the data is enhanced by using a Mix-up method. The linear modeling can improve the generalization performance of the model.
The calculation formula of the data enhancement is as follows:
wherein, δ is a fusion ratio, and the value range is 0 to 1. x is the number ofi，yiThe sample data before the non-enhancement is represented,
representing the enhanced sample data.
Preferably, 10% of data in the processed data are randomly selected as a test set sample on the principle that the test number is not in one day as much as possible, and the rest of data are used as training set samples after the Mix-up data enhancement.
Fig. 5 is a schematic diagram of a prediction model network structure provided in an embodiment of the present invention, where the structure includes:
a classification module: the method is used for predicting whether the communication link is accessible at other frequencies according to the frequency characteristic information of a certain frequency. That is, the input data is firstly predicted to be two types of accessible link and inaccessible link, and the two types of accessible link and inaccessible link are respectively predicted to improve the network efficiency. And for the type of output which is not communicated, the output is directly 0, which represents that the received signal-to-noise ratio is minus infinity, and the frequency point is not communicated.
A regression module: for predicting a signal-to-noise ratio of frequencies over which a communication link may be communicated. The data of a type which can be passed is directly input into a regression network which is trained, and a predicted waveform is obtained.
And synthesizing the output of the classification module and the regression module to obtain the signal-to-noise ratio waveform of other frequency points near the input frequency on a full frequency band (3-30 MHz).
And then the optimal communication signal-to-noise ratio frequency point near the input can be obtained through calculation and used as the output of the network.
Fig. 6 is a schematic diagram of a neural network structure of a prediction model according to an embodiment of the present invention, where the output of the kth neuron is:
y(k)＝[g(W1·x)+b1][g(W2·x)+b2]...[g(WL·x)+bL]·βk+b2|k|
wherein g (-) represents an excitation function, Wi∈Rn，bi∈RnAnd i is 1, …, and L represents the input weight of the ith neuron in the hidden layer. Beta is ak∈RLAnd k is 1, and m represents the input weight of the kth neuron in the output layer.
The structure includes:
classifying the network: the method comprises an input layer, four full-connection layers (64 nodes), an output layer, an activation function selection 'sigmoid' function, and a target function which is a cross entropy function:
regression network: the system comprises an input layer, five fully-connected layers (64 nodes), an output layer, an activation function selection 'ReLU' function and an adaptive learning rate of 0.01-0.0005.
The metric learning mechanism comprises: according to different emphasis points of frequency prediction, a metric learning mechanism is introduced. Because the invention calculates the optimal frequency of the maximum signal-to-noise ratio of different frequency points, the measurement of high signal-to-noise ratio is improved, so that the prediction network can pay more attention to the fitting of the frequency points with high signal-to-noise ratio. There are numerous models for metric learning, and embodiments of the present invention focus on using metric learning in the loss function of the neural network. Generally, the loss function used by the regression network for sequence prediction is the Mean Square Error (MSE) function:
the embodiment of the invention adds higher y on the basis of MSEiThe loss function that alters the regression network is:
wherein, ypreFor network prediction value, yiN is the number of samples for the actual value.
I.e. an amplification is given to the signal error with the signal-to-noise ratio larger than 1(snr larger than 0db), so that the signal error occupies a larger position in the backward propagation. Signals with a signal-to-noise ratio of less than 1(snr less than 0db) are given an attenuation that makes them less dominant in back propagation. This makes the network more sensitive to signal errors with larger signal-to-noise ratios.
In addition, adding Dropout or regularization techniques in each hidden layer of the built neural network prevents the overfitting problem. Dropout discards neurons with probability, changing the structure of the network itself and making the generalization of the model more efficient. Regularization limits the size of the parameters by adding a sum term to the original objective function. Using L2 regularization that is more applicable and prevents overfitting, the corresponding penalty term is the L2 norm:
from the above equation, it can be seen that the L2 regularization achieves regularization by adding the square sum of all the feature coefficients to the original objective function.
Fig. 7 is a training error trend chart of the prediction model according to the embodiment of the present invention.
The sample set data is historical signal-to-noise ratio data of the Wuhan-Beijing communication link in 2019 and 10 months, and time, frequency, signal-to-noise ratio at the current moment, signal-to-noise ratio before one hour, signal-to-noise ratio before two hours and signal-to-noise ratio before one day are selected as input characteristics. Processing the data into a plurality of groups of full-frequency-band short-wave communication data in hours, randomly selecting 10% of the data as a test sample according to the principle that the data is not in one day as much as possible in the test data, and enhancing the rest data by using Mix-up data to serve as a training sample.
The error function curve graph of the training shows that the descending trend of the error of the test set is close to the descending trend of the error of the training set and finally tends to convergence, and the prediction network can fit the nonlinear change rule of the short wave frequency selection problem and has good generalization capability.
Fig. 8 is a prediction result waveform comparison diagram of the intelligent short-wave frequency cross-band real-time prediction method provided by the embodiment of the invention.
The regression network results predicted results with errors ranging from 8% to 12% for the test set, with an average error RMSE of 0.25. Since the raw data is scaled to scale units and the error is scaled to dB, the RMSE range is 0.96dB to 2.23dB for data that we are more concerned with signal-to-noise ratios greater than 0. And due to the nonlinearity of the conversion formula, the higher the signal-to-noise ratio is (the more concerned is), the lower the prediction error is, and the lower the signal-to-noise ratio is (the less concerned is), the higher the prediction error is.
In summary, the method for predicting the accessible frequency near the short-wave communication interruption provided by the embodiment of the invention fully considers the frequency spectrum environment of the high-speed change of the short wave and the operation condition of actual frequency selection, and can select the next accessible frequency with a better effect under the condition of short-wave communication interruption.
The embodiment of the invention provides an intelligent short-wave frequency cross-frequency-band real-time prediction system, which comprises:
the pre-training module is used for acquiring historical sampling data of short-wave communication, training a prediction model based on a neural network by using the historical sampling data, and predicting the communication quality of other frequencies according to frequency characteristic information of a certain frequency;
the correlation calculation module is used for determining the range of the next available frequency after the short-wave communication is interrupted according to the correlation among different frequencies;
the interruption frequency characteristic information acquisition module is used for acquiring frequency characteristic information at the short-wave communication interruption moment;
the prediction module is used for inputting the frequency characteristic information of the short-wave communication interruption moment into the prediction model and acquiring the communication quality of each frequency in the range;
and the output module is used for selecting the frequency with the best communication quality in the range as the next accessible frequency selected after the short-wave communication is interrupted.
The implementation principle and technical effect of the system are the same as those of the method, and are not described herein again.
It must be noted that in any of the above embodiments, the methods are not necessarily executed in order of sequence number, and as long as it cannot be assumed from the execution logic that they are necessarily executed in a certain order, it means that they can be executed in any other possible order.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. An intelligent short-wave frequency cross-frequency-band real-time prediction method is characterized by comprising the following steps:
acquiring historical sampling data of short-wave communication, and training a prediction model based on a neural network by using the historical sampling data, wherein the prediction model is used for predicting the communication quality of other frequencies according to frequency characteristic information of a certain frequency;
determining the range of the next available frequency after the short-wave communication is interrupted according to the correlation among different frequencies;
acquiring frequency characteristic information at the short-wave communication interruption moment;
inputting frequency characteristic information of short-wave communication interruption time into the prediction model, and acquiring communication quality of each frequency in the range;
and selecting the frequency with the best communication quality in the range as the next accessible frequency selected after the short-wave communication is interrupted.
2. The intelligent short-wave frequency cross-band real-time prediction method of claim 1, wherein the determination of the next available frequency range after the short-wave communication interruption according to the correlation among different frequencies is specifically:
calculating a correlation coefficient r between different frequencies by adopting a Pearson correlation coefficient method, wherein the calculation formula is as follows:
wherein, Xi,YiThe sample data is represented by a sample data,
denotes the sample mean, n denotes the number of samples, σX,σYRepresents the sample standard deviation;
and determining the range of the next available frequency after the short-wave communication is interrupted according to the correlation coefficient r.
3. The intelligent short-wave frequency cross-frequency-band real-time prediction method of claim 1, wherein the frequency characteristic information at the short-wave communication interruption time comprises: the time of the short-wave communication interruption moment, the chain breakage frequency and the communication signal-to-noise ratio of different moments before interruption.
4. The intelligent short-wave frequency cross-band real-time prediction method of claim 1, wherein the historical sample data is preprocessed before the prediction model is trained by the historical sample data, and the preprocessing comprises the following steps:
performing data smoothing on the historical sampling data, wherein the data smoothing is to fill data with missing signal-to-noise ratio by adopting a signal-to-noise ratio mean value;
performing nonlinear unit conversion on the data subjected to the data smoothing processing, wherein the nonlinear unit conversion refers to the conversion of signal-to-noise ratio data with the unit of dB into the ratio of signal power to noise power;
and performing data enhancement on the data converted by the nonlinear unit to expand the training sample.
5. The intelligent short-wave frequency cross-band real-time prediction method of claim 1, wherein the prediction model comprises a classification network and a regression network;
the classification network is used for predicting whether communication links under other frequencies can be communicated according to the frequency characteristic information of a certain frequency;
the regression network is used to predict a signal-to-noise ratio for a frequency over which the communication link may be passed.
6. The intelligent short-wave frequency cross-band real-time prediction method of claim 5, wherein the classification network comprises 1 input layer, 4 full-connection layers and 1 output layer, an activation function of the classification network adopts a sigmoid function, and an objective function of the classification network adopts a cross-entropy function:
wherein H (X) represents information entropy, P (x)i) Represents a category xiThe resulting probability is predicted.
7. The intelligent short-wave frequency cross-band real-time prediction method of claim 5, wherein the regression network comprises 1 input layer, 5 full-link layers and one output layer, the activation function of the regression network adopts a ReLU function, and the adaptive learning rate of the regression network is 0.01 to 0.0005.
8. The intelligent short-wave frequency cross-band real-time prediction method of claim 5, wherein an objective function of the regression network is as follows:
wherein, ypreFor the network to predict value, yiN is the number of samples for the actual value.
9. The method of claim 5, wherein the range is a frequency within 2MHz of a frequency of the interruption time.
10. An intelligent short wave frequency cross-frequency-band real-time prediction system is characterized by comprising:
the device comprises a pre-training module, a neural network-based prediction model and a communication quality prediction module, wherein the pre-training module is used for acquiring historical sampling data of short-wave communication and training the prediction model based on the neural network by using the historical sampling data, and the prediction model is used for predicting the communication quality of other frequencies according to frequency characteristic information of a certain frequency;
the correlation calculation module is used for determining the range of the next available frequency after the short-wave communication is interrupted according to the correlation among different frequencies;
the interruption frequency characteristic information acquisition module is used for acquiring frequency characteristic information at the short-wave communication interruption moment;
the prediction module is used for inputting frequency characteristic information of short-wave communication interruption time into the prediction model and acquiring the communication quality of each frequency in the range;
and the output module is used for selecting the frequency with the best communication quality in the range as the next accessible frequency selected after the short-wave communication is interrupted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011449836.1A CN112616160B (en) | 2020-12-12 | 2020-12-12 | Intelligent short-wave frequency cross-frequency-band real-time prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011449836.1A CN112616160B (en) | 2020-12-12 | 2020-12-12 | Intelligent short-wave frequency cross-frequency-band real-time prediction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112616160A CN112616160A (en) | 2021-04-06 |
| CN112616160B true CN112616160B (en) | 2022-06-21 |
Family
ID=75232982
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011449836.1A Active CN112616160B (en) | 2020-12-12 | 2020-12-12 | Intelligent short-wave frequency cross-frequency-band real-time prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112616160B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114301557B (en) * | 2021-12-16 | 2023-12-29 | 中国人民解放军国防科技大学 | Short wave frequency selection method and system based on combination of predictive data and historical data |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106878943A (en) * | 2017-01-11 | 2017-06-20 | 中国人民解放军国防信息学院 | Short wave network Situation Awareness realization method and system |
| CN107180260A (en) * | 2017-06-02 | 2017-09-19 | 西安电子科技大学 | Short wave communication frequency selecting method based on Evolutionary Neural Network |
| CN110730046A (en) * | 2019-10-18 | 2020-01-24 | 中国人民解放军陆军工程大学 | Cross-frequency-band spectrum prediction method based on deep migration learning |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10425294B2 (en) * | 2014-01-06 | 2019-09-24 | Cisco Technology, Inc. | Distributed and learning machine-based approach to gathering localized network dynamics |
| CN109067427B (en) * | 2018-08-16 | 2019-11-22 | 北京科技大学 | A kind of frequency hop sequences prediction technique based on Optimization-type wavelet neural network |
-
2020
- 2020-12-12 CN CN202011449836.1A patent/CN112616160B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106878943A (en) * | 2017-01-11 | 2017-06-20 | 中国人民解放军国防信息学院 | Short wave network Situation Awareness realization method and system |
| CN107180260A (en) * | 2017-06-02 | 2017-09-19 | 西安电子科技大学 | Short wave communication frequency selecting method based on Evolutionary Neural Network |
| CN110730046A (en) * | 2019-10-18 | 2020-01-24 | 中国人民解放军陆军工程大学 | Cross-frequency-band spectrum prediction method based on deep migration learning |
Non-Patent Citations (3)
| Title |
|---|
| EMD-Based Multi-Model Prediction for Network Traffic in Software-Defined Networks;Longfei dai et.al;《2014 IEEE 11th International Conference on Mobile Ad Hoc and Sensor Systems》;20150209;全文 * |
| 基于BP神经网络的短波通信信道预测方法;张想等;《工业控制计算机》;20171025(第10期);全文 * |
| 短波通信频率自优化技术分析研究;徐池 等;《通信技术》;20200410;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112616160A (en) | 2021-04-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886498B (en) | EMD-GRU short-term power load prediction method based on feature selection | |
| CN110705692B (en) | Nonlinear dynamic industrial process product prediction method of space-time attention network | |
| CN110502806B (en) | Wireless spectrum occupancy rate prediction method based on LSTM network | |
| CN110443417A (en) | Multiple-model integration load forecasting method based on wavelet transformation | |
| CN111144552B (en) | Multi-index grain quality prediction method and device | |
| Li et al. | Domain adaptation remaining useful life prediction method based on AdaBN-DCNN | |
| CN106656357B (en) | Power frequency communication channel state evaluation system and method | |
| CN115098999A (en) | Multi-mode fusion fuel cell system performance attenuation prediction method | |
| CN112616160B (en) | Intelligent short-wave frequency cross-frequency-band real-time prediction method and system | |
| CN113126038A (en) | High-frequency ground wave radar working frequency optimization method, system, storage medium and application | |
| CN114584230B (en) | Predictive channel modeling method based on countermeasure network and long-term and short-term memory network | |
| CN115796351A (en) | Rainfall shorthand prediction method and device based on variational modal decomposition and microwave attenuation | |
| CN114239990A (en) | Time series data prediction method based on time series decomposition and LSTM | |
| CN109450573A (en) | A kind of frequency spectrum sensing method based on deep neural network | |
| CN116128690B (en) | Carbon emission cost value calculation method, device, equipment and medium | |
| CN116561569A (en) | Industrial power load identification method based on EO feature selection and AdaBoost algorithm | |
| CN111160419B (en) | Deep learning-based electronic transformer data classification prediction method and device | |
| CN113051809A (en) | Virtual health factor construction method based on improved restricted Boltzmann machine | |
| CN112307918A (en) | Diagnosis method for transformer direct-current magnetic biasing based on fuzzy neural network | |
| Fu et al. | Speech quality objective assessment using neural network | |
| CN111523258B (en) | Microseism effective signal first arrival pickup method and system based on MS-Net network | |
| CN117972536B (en) | Pulse classification method and system | |
| CN114330924B (en) | Complex product change strength prediction method based on generating type countermeasure network | |
| CN114548459B (en) | Ticket data regulation and control method and system and computer readable storage medium | |
| CN109521176B (en) | Virtual water quality monitoring method based on improved deep extreme learning machine |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |