Multi-channel electromagnetic spectrum intelligent prediction method and system

Abstract

The embodiments of the present application relate to the field of electromagnetic spectrum prediction technology and disclose a multi-channel electromagnetic spectrum intelligent prediction method and system, including: sorting the acquired multi-channel spectrum data by correlation, and forming a two-dimensional relationship neighboring time-frequency matrix based on the correlation sorting results; performing a multi-channel three-dimensional convolution operation on the two-dimensional relationship neighboring time-frequency matrix, deeply mining the spatial features while keeping the temporal features unchanged, and obtaining multiple time series sequences that have completed spatial feature mining; simultaneously inputting the multiple time series sequences that have completed spatial feature mining into an LSTM network, and using the LSTM network to deeply mine the temporal features; and compressing and fusing the output of the LSTM network through a fully connected layer to obtain the predicted occupancy probability of the next time slot of the multi-channel spectrum. This method effectively achieves accurate prediction of future spectrum occupancy status, can know the changes in the multi-channel spectrum situation in advance, and effectively improves the utilization rate of spectrum resources.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of electromagnetic spectrum prediction, and in particular to a multi-channel electromagnetic spectrum intelligent prediction method and system. Background Art

[0002] In today's information-driven and intelligent age, the electromagnetic spectrum, as the primary carrier of modern information transmission and military reconnaissance, is a crucial strategic resource. With the advancement of electronic technology, the electromagnetic spectrum has gradually become one of the primary combat spaces, operating across sea, land, air, and space. Electromagnetic countermeasures have become a crucial component that cannot be ignored. Electromagnetic frequency-based equipment and systems for various functions, such as communication, detection, reconnaissance, navigation, and cognition, are ubiquitous on the battlefield, and their number is growing exponentially. This has led to the increasing complexity of the electromagnetic spectrum, evolving into a multi-input, multi-output, multi-mode, and contested space shared by multiple parties. Electronic countermeasures and security management in complex electromagnetic environments are becoming increasingly crucial for national defense and safeguarding China's image.

[0003] With evolving mission requirements and environmental dynamics, traditional electronic reconnaissance alone can no longer meet the stringent demands of today's complex electromagnetic environments. Innovating electromagnetic spectrum attack and defense theories and technologies to deliver precise strikes against the enemy and gain an advantage in electromagnetic control and countermeasures in the silent battlefield of the electromagnetic spectrum is an urgent and crucial research topic. All electromagnetic spectrum activity relies on spectrum resources, and multi-channel electromagnetic spectrum prediction technology can help seize the initiative on the battlefield. On the one hand, predicting the frequency usage of enemy electromagnetic devices with swarm effects can reveal their behavioral relationships within the network and hidden information within it, identify their communication relationships, and even infer their hierarchical position within the communication network, ultimately locating key nodes within the network. On the other hand, predicting the frequency usage of these friendly electromagnetic devices with swarm effects can integrate spatial electromagnetic spectrum information, providing comprehensive information support for spectrum control, optimizing spectrum resources, and improving the reliability of spectrum control.

[0004] Numerous spectrum prediction methods have been proposed. Classic ones include Bayesian inference algorithms, exponential smoothing algorithms, autoregressive models, support vector machines, and multilayer perceptron models. Newer methods include recurrent networks, long short-term memory networks, deep residual networks, generative adversarial networks, and algorithms based on channel-space attention modules. Most of the aforementioned spectrum prediction methods focus on reasoning about spectral behavior in the temporal dimension, while joint prediction of time, space, and frequency is still in its early stages of research. In particular, regarding multi-channel joint prediction of the electromagnetic spectrum and how to achieve efficient, real-time, and accurate inference, many open theoretical and technical issues remain worthy of further exploration.

[0005] Through the above analysis, the current spectrum prediction method has the following defects.

[0006] First, the spectrum prediction technologies currently proposed only consider the characteristic information of one-dimensional time series, or process multiple one-dimensional time series simultaneously through multiple channels, without processing spatial features.

[0007] Second, the multi-channel spectrum prediction technology currently proposed is still in the early exploratory stage, the prediction accuracy cannot be guaranteed and the algorithm complexity is too high. Summary of the Invention

[0008] To solve the above technical problems, the embodiments of the present application provide a multi-channel electromagnetic spectrum intelligent prediction method and system. By fully mining and extracting spatial features, it achieves accurate prediction of future spectrum occupancy status, and obtains advance knowledge of multi-channel spectrum situation changes, thereby improving spectrum resource utilization.

[0009] To achieve the above-mentioned purpose, an embodiment of the present application proposes a multi-channel electromagnetic spectrum intelligent prediction method, comprising the following steps: sorting the acquired multi-channel spectrum data by correlation, forming a two-dimensional relationship neighboring time-frequency matrix based on the correlation sorting results, and performing zero padding before the first row and after the last row; performing a multi-channel three-dimensional convolution operation on the two-dimensional relationship neighboring time-frequency matrix, deeply mining the spatial features while keeping the time features unchanged, and obtaining multiple time series sequences that have completed spatial feature mining; simultaneously inputting the multiple time series sequences that have completed spatial feature mining into a preset LSTM network, and using the LSTM network to deeply mine the time features; compressing and fusing the output of the LSTM network through a fully connected layer to obtain the predicted occupancy probability of the next time slot of the multi-channel spectrum.

[0010] To achieve the above-mentioned purpose, an embodiment of the present application also proposes a multi-channel electromagnetic spectrum intelligent prediction system, which includes: a sorting module, which is used to sort the acquired multi-channel spectrum data by correlation, form a two-dimensional relationship neighboring time-frequency matrix based on the correlation sorting results, and perform zero padding before the first row and after the last row of the two-dimensional relationship neighboring time-frequency matrix; a spatial feature mining module, which is used to perform multi-channel three-dimensional convolution operations on the two-dimensional relationship neighboring time-frequency matrix, deeply mine spatial features while keeping the time features unchanged, and obtain multiple time series sequences that have completed spatial feature mining; a time feature mining module, which is used to simultaneously input multiple time series sequences that have completed spatial feature mining into a preset LSTM network, and use the LSTM network to deeply mine time features; a compression fusion prediction module, which is used to compress and fuse the output of the LSTM network through a fully connected layer to obtain the predicted occupancy probability of the next time slot of the multi-channel spectrum.

[0011] To achieve the above-mentioned objectives, an embodiment of the present application also proposes an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a multi-channel electromagnetic spectrum intelligent prediction method as described above.

[0012] To achieve the above objectives, an embodiment of the present application further proposes a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement a multi-channel electromagnetic spectrum intelligent prediction method as described above.

[0013] The embodiment of the present application proposes a multi-channel electromagnetic spectrum intelligent prediction method, which sorts the acquired multi-channel spectrum data by linear correlation analysis and nonlinear correlation analysis, forms a two-dimensional relationship neighboring time-frequency matrix based on the correlation sorting results, and then uses a multi-channel three-dimensional convolution operation to process the two-dimensional relationship neighboring time-frequency matrix, thereby realizing the mining of spatial features between channels in the spatial dimension without destroying the temporal sequence of the multi-channel spectrum data, and upgrading spectrum prediction from one-dimensional time series mining to time-space two-dimensional feature mining. By constructing an LSTM network and a fully connected layer, the time dimension features of the multi-channel spectrum data are deeply mined, and the final time-space two-dimensional features are compressed and extracted to obtain the state transition probability of the high-order hidden Markov model, that is, the spectrum occupancy probability of the multi-channel in the next time slot is obtained, thereby improving the accuracy of the multi-channel spectrum prediction. Finally, an accurate prediction of the future spectrum occupancy state is achieved, thereby knowing the changes in the multi-channel spectrum situation in advance, thereby effectively improving the utilization rate of spectrum resources. The method also has strong generalization capabilities.

[0014] Optionally, let the input time slot be T and the multi-channel spectrum data be k is the total number of channels, is the spectrum data of the i-th channel, and the correlation sorting of the acquired multi-channel spectrum data includes:

[0015] Calculate the Phi correlation coefficient and mutual information entropy between the spectrum data of each channel;

[0016] The mutual information entropy is normalized by standard, and then the Phi correlation coefficient and the mutual information entropy after standard normalization are linearly weighted to obtain the comprehensive correlation parameter;

[0017] Based on the comprehensive correlation parameter, the spectrum data of each channel is sorted by correlation;

[0018] The Phi correlation coefficient is calculated using the following formula:

[0019]

[0020] Among them, i≠j, a i and a j Respectively and The number of individuals with the same value, b i and b j Respectively and The number of individuals with different values ​​in , express and Phi correlation coefficient between them;

[0021] Mutual information entropy is calculated by the following formula:

[0022]

[0023] in, and Respectively and The marginal probability of express and The joint probability of express and The mutual information entropy between .

[0024] Optionally, the mutual information entropy is normalized by standard normalization, and the Phi correlation coefficient and the normalized mutual information entropy are linearly weighted to obtain a comprehensive correlation parameter, which is implemented by the following formula:

[0025]

[0026] Among them, Norm[·] represents standard normalization, R ij for and Comprehensive correlation parameters between

[0027] The method of forming a two-dimensional neighboring time-frequency matrix based on the correlation sorting results and performing zero padding before the first row and after the last row includes:

[0028] The spectrum data of the first channel is placed in the first row of the two-dimensional neighboring time-frequency matrix. The channel with the largest comprehensive correlation parameter with the spectrum data of the first channel is taken as the neighboring channel, and its spectrum data is placed in the second row. The spectrum data of the channel with the largest comprehensive correlation parameter with the spectrum data of the neighboring channel other than the first channel is placed in the third row, and so on. Finally, zero padding is performed before the first row and after the last row to form a two-dimensional neighboring time-frequency matrix.

[0029] Optionally, the multi-channel three-dimensional convolution operation is performed on the two-dimensional neighboring time-frequency matrix to deeply mine the spatial features while keeping the temporal features unchanged, including:

[0030] Under the premise of keeping the temporal features unchanged, a multi-channel three-dimensional convolution operation is performed on the two-dimensional neighboring time-frequency matrix to deeply mine the spatial features. A different convolution kernel is used for each channel, and each convolution kernel represents the extraction of a local feature.

[0031] A multi-channel three-dimensional convolution operation with a total number of channels M is performed on the two-dimensional neighboring time-frequency matrix to deeply mine spatial features. This is achieved through the following formula:

[0032]

[0033] Among them, [X T ] represents the two-dimensional relationship adjacent time-frequency matrix, conv m <·> represents the convolution kernel used for the mth channel, Represents the convolution result of the mth channel.

[0034] Optionally, after obtaining the convolution results of each channel, the convolution results of each channel are compressed from the region partition of each channel output using an average pooling operation to reduce the dimension, and finally obtain multiple time series that complete spatial feature mining;

[0035] The average pooling operation is expressed as:

[0036] X″ T =[x″1,x″2,…,x″ T ];

[0037]

[0038] Where, t=1,2,…,T, represents the convolution result of the mth channel in the tth time slot, x″ t represents the spatial feature deep mining result of the t-th time slot, X″ T It represents the final time series of multiple completed spatial feature mining.

[0039] Optionally, before the obtained multi-channel spectrum data is sorted by correlation, an LSTM network model is pre-built, and multiple time series sequences that have completed spatial feature mining are simultaneously input into the preset LSTM network, and the LSTM network is used to deeply mine the time features, which is achieved by the following formula:

[0040] Y=LSTM(X″ T )=[y 1 ,y 2 ,…,y i ,…,y k ];

[0041] Among them, LSTM(·) represents the LSTM network, y i represents the output of the LSTM network for the i-th channel, and Y represents the output of the LSTM network.

[0042] Optionally, compressing and fusing the output of the LSTM network through a fully connected layer to obtain a predicted occupancy probability of the next time slot of the multi-channel spectrum includes:

[0043] Establish a multi-channel high-order hidden Markov model and define x t,i ∈{0,1} is the spectrum state of channel i at time slot t, x t,i =1 means channel i is occupied at time slot t, x t,i =0 means that channel i is idle at time slot t;

[0044] Based on the multi-channel high-order hidden Markov model, the problem of estimating the spectrum state of k channels at the next moment is transformed into the problem of estimating the state transition probability of k channels at the next moment;

[0045] The output of the LSTM network is compressed and fused through a fully connected layer. The output of the LSTM network is processed into k dimensions through linear transformation and activation function, representing the spectrum occupancy probability of k channels at the next moment, that is, the state transition probability of the high-order hidden Markov model;

[0046] The prediction result is output for each channel based on the magnitude relationship of the spectrum occupancy probability. If the predicted spectrum occupancy probability of the i-th channel at the next moment is greater than 0.5, the spectrum state of the i-th channel at the next moment is considered to be occupied. Otherwise, it is considered to be idle. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the related technologies, the following is a brief introduction to the drawings required for use in the embodiments of the present application or the description of the related technologies. Obviously, the following drawings are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. The drawings described here are only used to explain the present application and are not used to limit the present application.

[0048] Figure 1 This is a flow chart of a multi-channel electromagnetic spectrum intelligent prediction method provided in one embodiment of the present application;

[0049] Figure 2 Schematic diagram of the training loss of the conv-LSTM network model based on correlation analysis provided in one embodiment of the present application;

[0050] Figure 3 This is a graph showing changes in the accuracy of occupancy probability prediction for eight channels, provided in one embodiment of the present application;

[0051] Figure 4 is a structural diagram of a multi-channel electromagnetic spectrum intelligent prediction system provided in another embodiment of the present application;

[0052] Figure 5 It is a structural diagram of an electronic device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below with reference to the accompanying drawings. In the various embodiments of the present application, many technical details are proposed to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in the present application can be implemented. The division of the following embodiments is only for the convenience of description and should not constitute any limitation on the specific implementation of the present application. The various embodiments can be combined with each other and referenced to each other under the premise of no contradiction.

[0054] An embodiment of the present application proposes a multi-channel electromagnetic spectrum intelligent prediction method, which is applied to an electronic device, wherein the electronic device can be a terminal or a server. In this embodiment and the following embodiments, the electronic device is described using a server as an example. The implementation details of the multi-channel electromagnetic spectrum intelligent prediction method proposed in this embodiment are specifically described below. The following content is only the implementation details provided for easy understanding and is not necessary for implementing this solution.

[0055] The specific process of the multi-channel electromagnetic spectrum intelligent prediction method proposed in this embodiment can be as follows: Figure 1 As shown, including:

[0056] S1, sort the obtained multi-channel spectrum data by correlation, form a two-dimensional neighboring time-frequency matrix based on the correlation sorting result, and perform zero padding before the first row and after the last row.

[0057] In the specific implementation, the server acquires multi-channel spectrum data in real time and sorts the obtained multi-channel spectrum data based on its relevance. The relevance sorting is based on both linear and nonlinear factors. After completing the relevance sorting, the server can construct a two-dimensional neighboring time-frequency matrix based on the relevance sorting results, and perform zero padding before the first row and after the last row. The zero padding operation effectively maintains dimensional consistency.

[0058] Assume that the input time slot is T and the multi-channel spectrum data is k is the total number of channels, is the spectrum data of the i-th channel. When the server sorts the acquired multi-channel spectrum data by correlation, it needs to calculate the Phi correlation coefficient and mutual information entropy between the spectrum data of each channel. The Phi correlation coefficient can reflect the linear relationship between the spectrum data of two channels, while the mutual information entropy can reflect the nonlinear relationship between the spectrum data of two channels. Subsequently, the server needs to apply standard normalization to the mutual information entropy, and then linearly weight the Phi correlation coefficient and the standard normalized mutual information entropy to obtain a comprehensive correlation parameter. Finally, based on the comprehensive correlation parameter, the spectrum data of each channel is sorted by correlation.

[0059] In one example, the Phi correlation coefficient is calculated using the following formula:

[0060]

[0061] Among them, i≠j, a i and a j Respectively and The number of individuals with the same value, b i and b j Respectively and The number of individuals with different values ​​in , express and Phi correlation coefficient between them.

[0062] In one example, mutual information entropy is calculated as follows:

[0063]

[0064] in, and Respectively and The marginal probability of express and The joint probability of express and The mutual information entropy between .

[0065] In one example, the server applies standard normalization to the mutual information entropy, and then linearly weights the Phi correlation coefficient and the normalized mutual information entropy to obtain a comprehensive correlation parameter, which is implemented by the following formula:

[0066]

[0067] Among them, Norm[·] represents standard normalization, R ij for and The comprehensive correlation parameters between .

[0068] In one example, when the server forms a two-dimensional relationship neighboring time-frequency matrix based on the correlation sorting results, it needs to place the spectrum data of the first channel in the first row of the two-dimensional relationship neighboring time-frequency matrix, and take the channel with the largest comprehensive correlation parameter with the spectrum data of the first channel as the adjacent channel (which can be recorded as channel d), and place the spectrum data of channel d in the second row, and then place the spectrum data of the channel with the largest comprehensive correlation parameter with the spectrum data of channel d (recorded as channel e) other than the first channel in the third row, and so on. After the spectrum data of all channels are placed in the two-dimensional relationship neighboring time-frequency matrix, zero padding is performed before the first row and after the last row of the two-dimensional relationship neighboring time-frequency matrix, thus forming a two-dimensional relationship neighboring time-frequency matrix.

[0069] In one example, the formula for the two-dimensional relational neighbor time-frequency matrix is ​​as follows:

[0070]

[0071] S2 performs a multi-channel three-dimensional convolution operation on the two-dimensional neighboring time-frequency matrix, deeply mining the spatial features while keeping the temporal features unchanged, and obtains multiple time series that complete the spatial feature mining.

[0072] In the specific implementation, after obtaining the two-dimensional relationship neighboring time-frequency matrix, the server needs to perform a multi-channel three-dimensional convolution operation on the two-dimensional relationship neighboring time-frequency matrix, deeply mining the spatial features while keeping the time features unchanged, and obtain multiple time series sequences that complete the spatial feature mining.

[0073] In one example, the server first needs to perform a multi-channel 3D convolution operation on the 2D neighboring time-frequency matrix, while maintaining the temporal characteristics unchanged, to deeply mine spatial features. Each channel uses a different convolution kernel, and each convolution kernel extracts a local feature. Performing a multi-channel 3D convolution operation on the 2D neighboring time-frequency matrix with a total number of channels M to deeply mine spatial features can be achieved using the following formula:

[0074]

[0075] Among them, [X T ] represents the two-dimensional relationship adjacent time-frequency matrix, conv m <·> represents the convolution kernel used for the mth channel, Represents the convolution result of the mth channel.

[0076] In an example, the specific formula of a convolution kernel is as follows:

[0077]

[0078] Among them, let l+1 be the current layer, Z l Represents the output of the previous layer, W l+1 Represents the convolution kernel sparse matrix of this convolution layer, Represents the convolution operator and b is the offset.

[0079] It should be noted that the convolution operation itself is linear, so the convolution output data needs to be activated by the ReLU activation function to improve the nonlinear representation ability. The ReLU function can be expressed as:

[0080]

[0081] In one example, after obtaining the convolution results of each channel, the server also needs to use the average pooling operation to compress the convolution results of each channel from the area division of each channel output to reduce the dimension, and finally obtain multiple time series that complete spatial feature mining.

[0082] In one example, the average pooling operation is formulated as:

[0083] X″ T =[x″1,x″2,…,x″ T ];

[0084]

[0085] Where, t=1,2,…,T, represents the convolution result of the mth channel in the tth time slot, x″ t represents the spatial feature deep mining result of the t-th time slot, X″ T It represents the final time series of multiple completed spatial feature mining.

[0086] S3, multiple time series sequences that have completed spatial feature mining are simultaneously input into the preset LSTM network, and the LSTM network is used to deeply mine the temporal features.

[0087] In the specific implementation, the server needs to build an LSTM network model in advance. After completing the mining and extraction of spatial features, multiple time series sequences that have completed spatial feature mining are simultaneously input into the preset LSTM network, and the LSTM network is used to deeply mine the time features.

[0088] In an example, the LSTM network is a chain result, and the specific operation of the LSTM neuron can be expressed by the formula:

[0089] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0090]

[0091] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0092]

[0093] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0094] h t =o t *tanh(c t );

[0095] Among them, i t is the input gate coefficient, is the new memory at the current moment, f t is the forget gate coefficient, ct-1 is the memory of the previous moment, c t For all memories of the current moment, o t is the output gate coefficient, h t-1 is the output of the previous neuron in the same layer, h t is the output of the neuron at the current moment, W i 、W c 、W f and W o They are the input gate weight matrix, the new memory weight matrix, the forget gate weight matrix and the output gate weight matrix, b i 、b c 、b f and b o They are the input gate offset, new memory offset, forget gate offset and output gate offset, respectively, x t is the output of the corresponding neuron in the previous hidden layer or input layer, [h t-1 ,x t ] means h t-1 and x t Connected splicing, * represents vector bitwise multiplication, tanh(·) represents the hyperbolic tangent function, and σ(·) represents the sigmoid function.

[0096] The hyperbolic tangent function and the sigmoid function can be expressed as:

[0097] tanh(x)=(e x -e -x ) / (e x +e -x );

[0098] σ(x)=1 / (1+e -x ).

[0099] In one example, multiple time series that have completed spatial feature mining are simultaneously input into a preset LSTM network, and the LSTM network is used to deeply mine the temporal features, which is achieved through the following formula:

[0100] Y=LSTM(X″ T )=[y 1 ,y 2 ,…,y i ,…,y k ];

[0101] Among them, LSTM(·) represents the LSTM network, y i represents the output of the LSTM network for the i-th channel, and Y represents the output of the LSTM network.

[0102] S4, compresses and fuses the output of the LSTM network through the fully connected layer to obtain the predicted occupancy probability of the next time slot of the multi-channel spectrum.

[0103] In the specific implementation, after completing the mining of spatial features and temporal features, the server can compress and fuse the output of the LSTM network through the fully connected layer to obtain the predicted occupancy probability of the next time slot of the multi-channel spectrum.

[0104] The server must first establish a multi-channel high-order hidden Markov model and define x t,i ∈{0,1} is the spectrum state of channel i at time slot t, x t,i =1 means channel i is occupied at time slot t, x t,i =0 indicates that channel i is in idle state at time slot t.

[0105] To estimate the spectrum state of k channels at the next moment based on a multi-channel high-order hidden Markov model, it is necessary to estimate the following state transition probabilities:

[0106]

[0107] Among them, f is the transition probability function for predicting different spectrum states of multiple channels, X k represents the spectrum state sequence of the kth channel in the past t time slots, P(x t+1,k =0)+P(x t+1,k =1)=1, which is the predicted probability of whether the spectrum of channel k is occupied at time slot t, and a binary decision is made on the spectrum occupancy based on this probability, as shown below:

[0108]

[0109] in, It indicates that the kth channel is predicted to be idle at the next moment (next time slot), It indicates that the kth channel is predicted to be occupied at the next moment.

[0110] This shows that, based on the multi-channel high-order hidden Markov model, the server can transform the problem of estimating the spectral state of k channels at the next moment into the problem of estimating the state transition probabilities of k channels at the next moment. The server then compresses and fuses the output of the LSTM network through a fully connected layer. This LSTM network output is processed into a k-dimensional form through a linear transformation and an activation function. This represents the spectrum occupancy probability of the k channels at the next moment, i.e., the state transition probability of the high-order hidden Markov model. A prediction result is then output for each channel based on the relative magnitude of the spectrum occupancy probabilities. If the predicted spectrum occupancy probability of the i-th channel at the next moment is greater than 0.5, the spectrum state of the i-th channel at the next moment is considered occupied; otherwise, it is considered idle.

[0111] The final result can be expressed as:

[0112] o T+1 =σ(Y·w+b)=[y 1 T+1 ,y a T+1 ,...,y h T+1 ];

[0113] P(x i T+1 =1) = conv_LSTM(X T )=y i T+1 ;

[0114] Among them, conv_LSTM(·) can represent the entire prediction model.

[0115] In this embodiment, the obtained multi-channel spectrum data is sorted by correlation through linear correlation analysis and nonlinear correlation analysis, and a two-dimensional relationship neighboring time-frequency matrix is ​​formed according to the sorting results. Then, a multi-channel three-dimensional convolution operation is used to process the two-dimensional relationship neighboring time-frequency matrix. This effectively realizes the mining of spatial features between channels in the spatial dimension without destroying the temporal sequence of the multi-channel spectrum data, and upgrades the spectrum prediction from one-dimensional time series mining to time-space two-dimensional feature mining. By constructing an LSTM network and a fully connected layer, the time dimension features of the multi-channel spectrum data are deeply mined, and the final time-space two-dimensional features are compressed and extracted to obtain the state transition probability of the high-order hidden Markov model, that is, the spectrum occupancy probability of the multi-channel in the next time slot is obtained, thereby improving the accuracy of the multi-channel spectrum prediction. Finally, an accurate prediction of the future spectrum occupancy state is achieved, so that the multi-channel spectrum situation changes are known in advance, thereby effectively improving the spectrum resource utilization rate.

[0116] The steps of the various methods above are divided only for clarity of description. They can be combined into one step or some steps can be decomposed into multiple steps during implementation. As long as they include the same logical relationship, they are all within the scope of protection of this application. Adding insignificant modifications or introducing insignificant designs to the algorithm or process without changing the core design of the algorithm and process are all within the scope of protection of this application.

[0117] In one embodiment, in order to evaluate the performance of a multi-channel electromagnetic spectrum intelligent prediction method proposed in this application, we conducted relevant simulation experiments. In the simulation experiment, the simulation platform selected Pytorch, an open source framework for deep learning based on the Python language, and used queuing theory to generate 40,000 groups of 8-channel spectrum data as a data set. 36,000 groups of samples were randomly selected as training sets to train the model, 2,000 groups of samples were selected as validation sets to supervise the training, and the last 2,000 groups of samples were used as test sets to test the training results. The channels are first sorted by correlation analysis combining the Phi correlation coefficient and mutual information entropy, and then a conv-LSTM network model (prediction model) is built.

[0118] Some parameters of the model are selected as follows: batch size batch_size is set to 100, the number of iterations epoch is set to 150, the number of output channels of the convolutional layer is set to 32, the input data shape is set to (8, 48), the convolution kernel size is set to 3, the pooling layer is set to use maximum pooling, the number of hidden layers of the LSTM network is set to 2, the hidden layer dimension is set to 48, and the output data dimension of the fully connected layer is set to (1, 8).

[0119] The binary logarithmic loss function is used as the objective function, and the Adam algorithm is used as the training optimizer. The parameters of the Adam optimizer are initialized as follows: the learning rate η is set to 0.005, the attenuation coefficient β1 of the first-order moment estimate is set to 0.9, and the attenuation coefficient β2 of the second-order moment estimate is set to 0.999.

[0120] The loss curves of the training set and test set of the model training of a multi-channel electromagnetic spectrum intelligent prediction method proposed in this application are as follows: Figure 2 As shown, it can be seen that the convergence process of the loss curves of the training set and the test set is smooth enough, and there is no excessive fluctuation in the whole process.

[0121] The prediction accuracy change diagram of 8 channels in a multi-channel electromagnetic spectrum intelligent prediction method proposed in this application is as follows: Figure 3As shown in the figure, it can be seen that the improvement process of the 8-channel spectrum state prediction accuracy is basically stable. Only a very small number of channels have a decrease in accuracy within a reasonable range. This is because the purpose of training is to improve the joint prediction accuracy. During the iteration process, the accuracy of individual channels decreases but the accuracy of other channels increases, thus ensuring that the overall accuracy remains improved.

[0122] It can be seen that the multi-channel electromagnetic spectrum intelligent prediction method proposed in this application is reliable.

[0123] Another embodiment of the present application proposes a multi-channel electromagnetic spectrum intelligent prediction system. The following is a detailed description of the implementation details of the multi-channel electromagnetic spectrum intelligent prediction system proposed in this embodiment. The following content is only for the convenience of understanding the implementation details and is not necessary for the implementation of this embodiment. Figure 4 This is a structural diagram of a multi-channel electromagnetic spectrum intelligent prediction system proposed in this embodiment. The system specifically includes: a sorting module M1, a spatial feature mining module M2, a temporal feature mining module M3 and a compression fusion prediction module M4.

[0124] The sorting module M1 is used to sort the acquired multi-channel spectrum data by correlation, form a two-dimensional relationship neighboring time-frequency matrix based on the correlation sorting result, and perform zero padding before the first row and after the last row of the two-dimensional relationship neighboring time-frequency matrix.

[0125] The spatial feature mining module M2 is used to perform multi-channel three-dimensional convolution operations on the two-dimensional relationship neighboring time-frequency matrix, deeply mining the spatial features while keeping the temporal features unchanged, and obtaining multiple time series sequences that complete the spatial feature mining.

[0126] The time feature mining module M3 is used to simultaneously input multiple time series that have completed spatial feature mining into the preset LSTM network, and use the LSTM network to deeply mine time features.

[0127] The compression fusion prediction module M4 is used to compress and fuse the output of the LSTM network through the fully connected layer to obtain the predicted occupancy probability of the next time slot of the multi-channel spectrum.

[0128] It is worth mentioning that all modules involved in this embodiment are logical modules. In actual applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, to highlight the innovation of this application, this embodiment does not include units that are not closely related to solving the technical problem proposed by this application. However, this does not mean that other units do not exist in this embodiment.

[0129] It is not difficult to find that this embodiment is a system embodiment corresponding to the above-mentioned method embodiments, and this embodiment can be implemented in conjunction with the above-mentioned method embodiments. The relevant technical details and technical effects mentioned in the above-mentioned method embodiments are still valid in this embodiment, and to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the above-mentioned method embodiments.

[0130] Another embodiment of the present application provides an electronic device, the specific structure of which is as follows: Figure 5 As shown, it includes: at least one processor C1; and a memory C2 communicatively connected to the at least one processor C1; wherein the memory C2 stores instructions that can be executed by the at least one processor C1, and the instructions are executed by the at least one processor C1 to enable the at least one processor C1 to execute a multi-channel electromagnetic spectrum intelligent prediction method as described in the above-mentioned method embodiments.

[0131] The memory and processor can be connected using a bus. The bus can include any number of interconnected buses and bridges, connecting various circuits within one or more processors and the memory. The bus can also connect various other circuits, such as peripherals, voltage regulators, and power management circuits. These are well known in the art and will not be described further herein. The bus interface is responsible for providing an interface between the bus and the transceiver. The transceiver can be a single component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium.

[0132] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.

[0133] Another embodiment of the present application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement a multi-channel electromagnetic spectrum intelligent prediction method as described in the above method embodiments.

[0134] That is, those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented by instructing related hardware through a program, wherein the program is stored in a storage medium and includes a number of instructions for causing a device (such as a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps in the above-described method embodiments. The storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0135] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and that in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A multi-channel electromagnetic spectrum intelligent prediction method, characterized in that: include: The obtained multi-channel spectrum data is sorted by correlation, and a two-dimensional neighboring time-frequency matrix is ​​formed based on the correlation sorting result, and zero padding is performed before the first row and after the last row; Perform multi-channel three-dimensional convolution operations on the two-dimensional neighboring time-frequency matrix, deeply mining the spatial features while keeping the temporal features unchanged, and obtain multiple time series that complete the spatial feature mining; Multiple time series that have completed spatial feature mining are simultaneously input into the preset LSTM network, and the LSTM network is used to deeply mine the temporal features; The output of the LSTM network is compressed and fused through the fully connected layer to obtain the predicted occupancy probability of the next time slot of the multi-channel spectrum.

2. A multi-channel electromagnetic spectrum intelligent prediction method according to claim 1, characterized in that: Assume that the input time slot is T and the multi-channel spectrum data is i=1,2,...,k, k is the total number of channels, is the spectrum data of the i-th channel, and the correlation sorting of the acquired multi-channel spectrum data includes: Calculate the Phi correlation coefficient and mutual information entropy between the spectrum data of each channel; The mutual information entropy is normalized by standard, and then the Phi correlation coefficient and the mutual information entropy after standard normalization are linearly weighted to obtain the comprehensive correlation parameter; Based on the comprehensive correlation parameter, the spectrum data of each channel is sorted by correlation; The Phi correlation coefficient is calculated using the following formula: Among them, i≠j, a i and a j Respectively and The number of individuals with the same value, b i and b j Respectively and The number of individuals with different values ​​in , express and Phi correlation coefficient between them; Mutual information entropy is calculated by the following formula: in, and Respectively and The marginal probability of express and The joint probability of express and The mutual information entropy between .

3. The multi-channel electromagnetic spectrum intelligent prediction method according to claim 2, characterized in that: The mutual information entropy is normalized by standard, and then the Phi correlation coefficient and the normalized mutual information entropy are linearly weighted to obtain the comprehensive correlation parameter, which is achieved by the following formula: Among them, Norm[·] represents standard normalization, R ij for and Comprehensive correlation parameters between The method of forming a two-dimensional neighboring time-frequency matrix based on the correlation sorting results and performing zero padding before the first row and after the last row includes: The spectrum data of the first channel is placed in the first row of the two-dimensional neighboring time-frequency matrix. The channel with the largest comprehensive correlation parameter with the spectrum data of the first channel is taken as the neighboring channel, and its spectrum data is placed in the second row. The spectrum data of the channel with the largest comprehensive correlation parameter with the spectrum data of the neighboring channel other than the first channel is placed in the third row, and so on. Finally, zero padding is performed before the first row and after the last row to form a two-dimensional neighboring time-frequency matrix.

4. The multi-channel electromagnetic spectrum intelligent prediction method according to claim 1, characterized in that: The multi-channel three-dimensional convolution operation is performed on the two-dimensional neighboring time-frequency matrix to deeply mine the spatial features while keeping the temporal features unchanged, including: Under the premise of keeping the temporal features unchanged, a multi-channel three-dimensional convolution operation is performed on the two-dimensional neighboring time-frequency matrix to deeply mine the spatial features. A different convolution kernel is used for each channel, and each convolution kernel represents the extraction of a local feature. A multi-channel three-dimensional convolution operation with a total number of channels M is performed on the two-dimensional neighboring time-frequency matrix to deeply mine spatial features. This is achieved through the following formula: Among them, [X T ] represents the two-dimensional relationship adjacent time-frequency matrix, conv m <·> represents the convolution kernel used for the mth channel, Represents the convolution result of the mth channel.

5. The multi-channel electromagnetic spectrum intelligent prediction method according to claim 4, characterized in that: After obtaining the convolution results of each channel, the average pooling operation is used to compress the convolution results of each channel from the area division of each channel output to reduce the dimension, and finally obtain multiple time series that complete spatial feature mining; The average pooling operation is expressed as: X″ T =[x″1,x″2,…,x″ T ]; Where, t=1,2,...,T, represents the convolution result of the mth channel in the tth time slot, x″ t represents the spatial feature deep mining result of the t-th time slot, X″ T It represents the final time series of multiple completed spatial feature mining.

6. A multi-channel electromagnetic spectrum intelligent prediction method according to claim 5, characterized in that: Before sorting the obtained multi-channel spectrum data by correlation, an LSTM network model is pre-built. Multiple time series that have completed spatial feature mining are simultaneously input into the preset LSTM network, and the LSTM network is used to deeply mine the time features. This is achieved through the following formula: Y=LSTM(X″ T )=[y 1 ,y 2 ,...,y i ,...,y k ]; Among them, LSTM(·) represents the LSTM network, y i represents the output of the LSTM network for the i-th channel, and Y represents the output of the LSTM network.

7. A multi-channel electromagnetic spectrum intelligent prediction method according to any one of claims 1 to 6, characterized in that: The method of compressing and fusing the output of the LSTM network through the fully connected layer to obtain the predicted occupancy probability of the next time slot of the multi-channel spectrum includes: Establish a multi-channel high-order hidden Markov model and define x t,i ∈{0, 1} as the spectrum state of channel i at time slot t, x t,i =1 means channel i is occupied at time slot t, x t,i =0 means that channel i is idle at time slot t; Based on the multi-channel high-order hidden Markov model, the problem of estimating the spectrum state of k channels at the next moment is transformed into the problem of estimating the state transition probability of k channels at the next moment; The output of the LSTM network is compressed and fused through a fully connected layer. The output of the LSTM network is processed into k dimensions through linear transformation and activation function, representing the spectrum occupancy probability of k channels at the next moment, that is, the state transition probability of the high-order hidden Markov model; The prediction result is output for each channel based on the magnitude relationship of the spectrum occupancy probability. If the predicted spectrum occupancy probability of the i-th channel at the next moment is greater than 0.5, the spectrum state of the i-th channel at the next moment is considered to be occupied. Otherwise, it is considered to be idle.

8. A multi-channel electromagnetic spectrum intelligent prediction system, characterized in that: include: A sorting module is used to sort the acquired multi-channel spectrum data by correlation, form a two-dimensional neighboring time-frequency matrix based on the correlation sorting results, and perform zero padding before the first row and after the last row of the two-dimensional neighboring time-frequency matrix; The spatial feature mining module is used to perform multi-channel three-dimensional convolution operations on the two-dimensional neighboring time-frequency matrix, deeply mining the spatial features while keeping the temporal features unchanged, and obtaining multiple time series that complete the spatial feature mining; The temporal feature mining module is used to simultaneously input multiple time series that have completed spatial feature mining into a preset LSTM network, and use the LSTM network to deeply mine temporal features; The compression fusion prediction module is used to compress and fuse the output of the LSTM network through the fully connected layer to obtain the predicted occupancy probability of the next time slot of the multi-channel spectrum.

9. An electronic device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; In which, the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a multi-channel electromagnetic spectrum intelligent prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it can implement a multi-channel electromagnetic spectrum intelligent prediction method as described in any one of claims 1 to 7.

Images