A method of channel-separated spectral prediction

Abstract

The application discloses a channel-separated spectrum prediction method, which belongs to the technical field of spectrum prediction.The application solves the problems of low prediction efficiency, low prediction accuracy and large storage resource occupation existing in the prior spectrum prediction method.The application fuses the channel dimension and the batch dimension, realizes the separation of the channel, and places the focus of the spectrum prediction model on the extraction of the time correlation, so that the influence of the channel interference and the large difference in the data distribution of the channels on the model feature extraction process is avoided.After the model outputs the prediction result, the batch dimension is split, the results of the channels are restored to the original dimension, and the future information of all the channels is predicted simultaneously.The method of the application combines the advantages of the channel independent prediction method and the channel joint prediction method, can improve the accuracy and efficiency of the spectrum data prediction, improves the generalization ability of the model through the sharing of the multi-channel model, and reduces the resource occupation.The method of the application can be applied to the technical field of spectrum prediction.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of spectrum prediction, and particularly relates to a spectrum prediction method for channel separation. BACKGROUND

[0002] In recent years, with the rapid development of mobile Internet, Internet of Things (IoT) and Internet of Vehicles (IoV), the demand for limited spectrum resources is increasing. This surge in demand urgently requires more intelligent and perfect spectrum utilization strategies. To achieve this goal, the electromagnetic spectrum state needs to be obtained in advance. There are mainly two methods to obtain this information: spectrum sensing and spectrum prediction. Spectrum sensing passively detects the current radio spectrum state using various signal processing methods. In contrast, spectrum prediction actively analyzes historical spectrum usage patterns to predict future usage. Accurate spectrum prediction (low prediction loss) can help new users identify idle or underutilized spectrum, enabling dynamic spectrum access and improving overall spectrum utilization. However, due to the development of wireless communication technology and the deployment of numerous heterogeneous wireless communication systems, spectrum prediction still faces many challenges brought by complex electromagnetic environments.

[0003] Spectrum prediction is to predict the future spectrum usage of a channel using time correlation and inter-channel correlation. Currently, the development of neural networks has enabled single-channel prediction to achieve high prediction accuracy. It is well known that multi-channel prediction, which simultaneously predicts the future spectrum state of multiple channels, still has many problems. Multi-channel prediction methods mainly include channel-independent prediction and channel-joint prediction. Channel-joint prediction method, i.e., training a model for multiple channels and using the model to predict the future results of all channels, has high prediction efficiency. The similarity between channels can provide inter-channel correlation when channel data is missing or insufficient, which can assist in predicting low-quality channels. When there is sufficient data between channels, the temporal correlation dominates the prediction process. At this time, the similarity between channels is not enough to provide additional information, and instead, it can lead to a decrease in prediction accuracy due to inter-channel interference and large differences in channel distribution. For this, we should use the channel-independent method. The channel-independent method trains a model for each channel to make predictions, reducing inter-channel interference and large differences in channel distribution that do not affect the feature extraction process of the model. However, in this method, each model focuses more on its own channel, inevitably leading to a decrease in its generalization ability. Sequentially predicting each channel will also bring a large time delay and reduce prediction efficiency. In addition, the resource occupation of this method is closely related to the number of channels, and a large number of channels will occupy a large amount of storage resources. SUMMARY

[0004] The application aims to solve the problems of low prediction efficiency, low prediction accuracy and large storage resource occupation of existing spectrum prediction methods, and proposes a channel-separated spectrum prediction method.

[0005] The technical scheme adopted by the application to solve the above technical problems is a channel-separated spectrum prediction method, which specifically comprises the following steps:

[0006] Step one, define the spectrum data of P time slots that need to be predicted in the future, and construct a corresponding spectrum prediction model according to the value of P;

[0007] And define the historical time slot number H corresponding to the historical spectrum data sequence input by the spectrum prediction model, O represents the set of all values of H;

[0008] Step two, the historical spectrum data sequence input by the spectrum prediction model is denoted as Wherein, M represents the number of channels that need to be predicted for spectrum data in each batch, and B represents the number of batches;

[0009] And the spectrum data of each channel in the historical spectrum data sequence X is normalized respectively, and the channel dimension of the normalized input data is fused into the batch dimension to obtain the processed historical spectrum data sequence

[0010] Step three, use the processed historical spectrum data sequence X' to train the spectrum prediction model, and obtain the trained spectrum prediction model corresponding to the current H value;

[0011] Step four, for each value of H in the set O, steps two and three are performed, that is, the trained spectrum prediction model corresponding to each H value is obtained respectively;

[0012] Then select the spectrum prediction model with the smallest mean square error loss from the trained spectrum prediction models corresponding to each H value;

[0013] Step five, process the historical spectrum data sequence of the channel to be predicted, and input the processed data into the selected spectrum prediction model to obtain the spectrum data prediction result of the channel to be predicted;

[0014] Then split the batch dimension of the spectrum data prediction result and restore it to the original channel dimension to obtain the final prediction result.

[0015] Further, the corresponding spectrum prediction model is constructed according to the value of P, specifically:

[0016] When P = 1, the constructed spectrum prediction model is a Seq2Seq model based on LSTM or an improved Autoformer model.

[0017] When P > 1, the constructed spectrum prediction model is an improved Autoformer model.

[0018] Further, the improved Autoformer model comprises an encoder and a decoder, and the encoder comprises N en encoding units, and the decoder comprises N de decoding units.

[0019] In the improved Autoformer model, the input of the improved Autoformer model is subjected to a linear mapping layer, the output of the linear mapping layer is taken as the input of the first encoding unit, the seasonal part output by the first encoding unit is taken as the input of the second encoding unit, the seasonal part output by the second encoding unit is taken as the input of the third encoding unit, and so on, the seasonal part output by the N en -1 encoding unit is taken as the input of the N en th encoding unit, and the seasonal part output by the N en th encoding unit is obtained.

[0020] The input of the improved Autoformer model is decomposed into the cumulative structure of the trend cycle component and the cumulative structure of the seasonal component, the cumulative structure of the trend cycle component, the cumulative structure of the seasonal component, and the seasonal part output by the N en th encoding unit are taken as the input of the first decoding unit, the trend cycle part, the seasonal part output by the first decoding unit, and the seasonal part output by the N en th encoding unit are taken as the input of the second decoding unit, and so on, the trend cycle part, the seasonal part output by the N de -1 decoding unit, and the seasonal part output by the N en th encoding unit are taken as the input of the N de th decoding unit, and the spectrum data prediction result is obtained according to the output of the N de th decoding unit.

[0021] Further, the working process of the encoder is as follows:

[0022] First, the dimension of the processed historical spectrum data sequence X' is expanded, X' is expanded to d dimensions, and the dimension expansion result X en is as follows:

[0023]

[0024] wherein, Embedding(·) represents the operation of the linear mapping layer;

[0025] The output X en of the linear mapping layer is taken as the input of the first encoding unit, and the input of the first encoding unit is taken as the input of the first correlation extraction module, and the output X (1) of the first correlation extraction module is:

[0026] X (1) = Auto-Correlation1(X en )

[0027] wherein, Auto-Correlation1(·) represents the operation of the first correlation extraction module;

[0028] The output of the first correlation extraction module is added to X en , and the addition result X (2) is:

[0029] X (2) = Auto-Correlation1(X en )+X en

[0030] The addition result X (2) is taken as the input of the first sequence decomposition module, and the trend cycle part X t(3) and the seasonal part X s(3) output by the first sequence decomposition module are:

[0031] The seasonal part X s(3) is taken as the input of the first forward propagation module, and the output X (4) of the first forward propagation module is:

[0032] X (4) = FFN1(X s(3) )

[0033] wherein, FFN1(·) represents the operation of the first forward propagation module;

[0034] The output of the first forward propagation module is added to the seasonal part X s(3) , and the addition result X (5) is:

[0035] X (5) = X (4) +X s(3)

[0036] The addition result X (5) is taken as the input of the second sequence decomposition module, and the trend cycle part output by the second sequence decomposition module is denoted as Xt(6) , the seasonal part of the output of the second sequence decomposition module is recorded as X s(6) , and X s(6) The seasonal component as the output of the first coding unit;

[0037] Then the seasonal part X output by the first encoding unit is s(6) As the input of the second coding unit; the seasonal part output by the second coding unit is used as the input of the third coding unit;

[0038] And so on, the Nth en -1 The seasonal part of the output of the coding unit is used as the Nth en The input of the encoding unit is obtained en Seasonal component of the encoding unit output

[0039] Furthermore, the cumulative structure of the trend cycle component and the cumulative structure of the seasonal component are obtained by:

[0040]

[0041] Among them, the processed historical spectrum data sequence X′ is passed through the sequence decomposition module to obtain the historical trend cycle sequence and historical seasonal sequence. It is the last H / 2 row of the historical trend cycle sequence. represents the last H / 2 rows of the historical seasonal series, mean(X′ H ) P represents a vector of P rows and 1 column consisting of the mean values ​​of each row in the historical spectrum data sequence X′, O P represents a 0 vector with P rows and 1 columns, represents the cumulative structure of the trend cycle component, Represents the cumulative structure of the seasonal component.

[0042] Furthermore, the working process of the decoder is:

[0043] Will and Nth en Seasonal component of the encoding unit output As the input of the first decoding unit, in the first decoding unit, As the input of the second correlation extraction module, the output of the second correlation extraction module is combined with Add, and use the addition result as the input of the third sequence decomposition module, and output X through the third sequence decomposition module s(7) and X t(7) :

[0044]

[0045] wherein Auto-Correlation2(·) represents the operation of the second correlation extraction module, SeriesDecomp3(·) represents the operation of the third sequence decomposition module;

[0046] X s(7) and As the input of the third correlation extraction module, the output of the third correlation extraction module is obtained by adding X s(7) and X s(8) as the input of the fourth sequence decomposition module, and the output of the fourth sequence decomposition module is obtained t(8) :

[0047]

[0048] wherein Auto-Correlation3(·) represents the operation of the third correlation extraction module, SeriesDecomp4(·) represents the operation of the fourth sequence decomposition module;

[0049] X s(8) as the input of the second forward propagation module, and the output of the second forward propagation module is obtained by adding s(8) X s(9) as the input of the fifth sequence decomposition module, and the output of the fifth sequence decomposition module is obtained t(9) :

[0050] X s(9) , t(9) = SeriesDecomp5(FFN2(X s(8) ) + X s(8) )

[0051] wherein FFN2(·) represents the operation of the second forward propagation module; SeriesDecomp5(·) represents the operation of the fifth sequence decomposition module;

[0052] X t(7) , X t(8) and X t(9) are added to obtain the addition result of the trend periodic component

[0053]

[0054] X s(9) as the seasonal part of the output of the first decoding unit, and as the trend periodic part of the output of the first decoding unit, and X s(9) and​ as the input of the second decoding unit;

[0055] Similarly, the Nth de seasonal part and the trend periodic part output by the decoding unit; the Nth de seasonal part output by the decoding unit is denoted as the trend periodic part output by the decoding unit is denoted as de the trend periodic part output by the decoding unit is denoted as the trend periodic part output by the decoding unit is denoted as de the trend periodic part output by the decoding unit is denoted as and the seasonal part are merged:

[0056]

[0057] wherein Y de represents the merging result;

[0058] the merging result Y de is mapped to obtain a spectrum data prediction result Y':

[0059] Y'=W o Y de +b o

[0060] wherein W o and b o are model parameters.

[0061] Further, the working process of the first correlation extraction module is as follows:

[0062] Step 1, linearly mapping X en to obtain sequences q, k and v;

[0063] Step 2, initializing τ=1;

[0064] Step 3, performing right circular shift on the sequence v according to the time delay τ to obtain a right circular shift result Roll(v, τ);

[0065] then calculating the autocorrelation function of X en

[0066]

[0067] wherein q t represents the tth element in the sequence q, and k t-τ represents the t-τth element in the sequence k;

[0068] Step 4, judging whether τ=H is satisfied; ​

[0069] If satisfied, proceed to step 5;

[0070] If not, set τ = τ + 1 and return to step 3;

[0071] Step 5: According to Select the k most correlated delays:

[0072]

[0073] Among them, τ1, τ2, ..., τ k represents the selected k delays;

[0074] For the selected k delays corresponding to Processing:

[0075]

[0076] in, Represent k delays τ1, τ2, ..., τ k The corresponding weights, softmax(·) is the normalized exponential function;

[0077] Then the output of the first correlation extraction module is Auto-Correlation(q,k,v):

[0078]

[0079] Furthermore, the first correlation extraction module performs calculations based on FFT and IFFT, and the specific calculation method is:

[0080] Step 1: Use sequence q as the input of the first linear layer, sequence k as the input of the second linear layer, and sequence v as the input of the third linear layer;

[0081] Step 2: Pass the output of the first linear layer through the first FFT module, pass the output of the second linear layer through the second FFT module, multiply the output of the first FFT module by the output of the second FFT module, and use the multiplication result as the input of the first IFFT module. The autocorrelation function value corresponding to each delay is output through the first IFFT module.

[0082] Then select the delays corresponding to the first k largest autocorrelation function values, and record the selected k delays as τ1, τ2, ..., τ k ;

[0083] Step 3: Delay the output of the third linear layer by τ1, τ2, ..., τ k The delay processing is to obtain delays τ1, τ2, ..., τk Corresponding delay processing results;

[0084] Step 4: multiply the delay processing result corresponding to delay τ1 by the autocorrelation function value corresponding to delay τ1, multiply the delay processing result corresponding to delay τ2 by the autocorrelation function value corresponding to delay τ2, ..., multiply the delay τ k Corresponding delay processing results and delay τ k The corresponding autocorrelation function values ​​are multiplied to obtain the delays τ1, τ2, ..., τ k The corresponding multiplication results are then accumulated to obtain the accumulated result;

[0085] The accumulated result is passed through the fourth linear layer, and the output of the fourth linear layer is used as the output of the first correlation extraction module.

[0086] Furthermore, the working process of the first sequence decomposition module is:

[0087] X t(3) =AvgPool(Padding(X (2) ))

[0088] X s(3) =X (2) -X t(3)

[0089] Among them, Padding(·) represents the data filling operation, AvgPool(·) represents the moving average operation, Represents the trend cycle part of the first sequence decomposition module output, Represents the seasonal part of the output of the first series decomposition module.

[0090] Furthermore, the working process of the first forward propagation module is:

[0091] FFN1(X s(3) )=w2relu(w1X s(3) +b1)+b2

[0092] Where FFN1(·) represents the operation of the first forward propagation module, w1, w2, b1 and b2 are model parameters, and relu(·) is the activation function.

[0093] The beneficial effects of the present invention are:

[0094] The application fuses the channel dimension and the batch dimension, realizes the separation of the channel, and the spectrum prediction model focuses on the extraction of time correlation, avoids the influence of channel interference and large differences in data distribution of each channel on the model feature extraction process. After the model output prediction result, the batch dimension is split, the results of each channel are restored to the original dimension, and the future information of all channels is predicted at the same time. The method of the application combines the advantages of channel independent and channel joint prediction method, can use a single model to predict the future results of multiple channels at the same time, thereby improving the accuracy and efficiency of spectrum data prediction. The multi-channel model sharing also makes the model no longer pay too much attention to each channel, improves the generalization ability of the model, and reduces the resource occupation. BRIEF DESCRIPTION OF DRAWINGS

[0095] Figure 1 is a two-dimensional power spectral density graph of a data set;

[0096] Figure 2 is a three-dimensional power spectral density graph of a data set;

[0097] Figure 3 is a structure diagram of a spectrum prediction model;

[0098] Figure 4 is a spectrum occupation state prediction diagram;

[0099] Figure 5 is a structure diagram of an improved Autoformer model;

[0100] Figure 6 is a schematic diagram of the calculation process of the correlation extraction module;

[0101] Figure 7 is an average MSE comparison diagram under different prediction methods;

[0102] Figure 8 is an average MAE comparison diagram under different prediction methods. DETAILED DESCRIPTION

[0103] Embodiment one: combination Figure 3 This embodiment describes a spectrum prediction method with channel separation. The method specifically includes the following steps:

[0104] Step one, define the spectrum data that needs to be predicted in the future P time slots (i.e. the power spectral density value that needs to be predicted in the future P time slots), and construct the corresponding spectrum prediction model according to the value of P;

[0105] And define the historical time slot number H corresponding to the historical spectrum data sequence input of the spectrum prediction model, O represents the set composed of all values of H (here, some values can be set according to experience to form the set, and H time slots should be selected continuously);

[0106] Step two, when the value of H is any value in set O, the historical spectrum data sequence input into the spectrum prediction model is denoted as wherein, represents a real number, M represents the number of channels that need to be predicted for each batch, and B represents the number of batches;

[0107] and the spectrum data of each channel in the historical spectrum data sequence X is normalized respectively, and then the channel dimension of the normalized input data is fused into the batch dimension according to the channel separation method, to obtain the processed historical spectrum data sequence

[0108] Step three, the processed historical spectrum data sequence X' is used to train the spectrum prediction model, the training label is the actual spectrum data of the prediction time slot, and the trained spectrum prediction model corresponding to the current H value is obtained;

[0109] Step four, for each value of H in set O, steps two and three are performed, that is, the trained spectrum prediction model corresponding to each H value is obtained respectively;

[0110] Then, the spectrum prediction model with the smallest mean square error loss is selected from the trained spectrum prediction models corresponding to each H value;

[0111] Step five, after processing the historical spectrum data sequence of the channel to be predicted (the processing method is the same as that of the historical spectrum data sequence during training), the processed data is input into the selected spectrum prediction model, and the spectrum data prediction result of the channel to be predicted is obtained;

[0112] Then, the batch dimension of the spectrum data prediction result is split and restored to the original channel dimension to obtain the final prediction result.

[0113] In the traditional method, the spectral data of multiple channels is included in each batch of model input data, so that the model training process is affected by the mutual interference between channels and the large difference in channel data distribution. The method of the present application first separates the channels and fuses them into the batch dimension, so that each batch only contains data of one channel, and a common prediction model is trained using multiple channels, thereby reducing the mutual interference between channels and the large difference in channel data distribution, and improving the prediction accuracy. Then, the improved Autoformer model is used for prediction. In the encoder part, the trend period component is gradually separated out, and the correlation extraction module is used to extract the complex seasonal fluctuation component. In the decoder, the autocorrelation and cross autocorrelation extraction module is used to gradually establish the seasonal component of the to-be-predicted part, and the trend period component is gradually added to the final prediction result. Finally, the batch dimension of the decoder prediction result is split and restored to the original channel. In the training stage, the historical time slot length and the prediction time slot length are changed to divide the spectral data of the training set, and the model of the corresponding prediction scene is trained. The trained model is applied to the actual multi-channel multi-step joint prediction, and the known spectral power spectral density is used to predict the spectral data in the unknown future time slot, and on this basis, the next step of spectrum sensing, spectrum decision, spectrum sharing and spectrum switching operation is performed.

[0114] Specific implementation method two: different from the specific implementation method one, the spectral prediction model is constructed according to the value of P, specifically:

[0115] When P = 1, the constructed spectral prediction model is an LSTM-based Seq2Seq model or an improved Autoformer model;

[0116] When P > 1, the constructed spectral prediction model is an improved Autoformer model.

[0117] The other steps and parameters are the same as those of the specific implementation method one.

[0118] In the actual training process, the value of P can be determined according to the actual prediction requirements, and then the spectral prediction model is constructed according to the value of P, and the value of H is changed to obtain the trained spectral prediction model under each H value, and then the best spectral prediction model and the H value corresponding to the best spectral prediction model are determined according to the training results. In the actual prediction process, for each channel, first, the spectral data of the continuous H time slots ending at the current time slot is obtained, then the channel dimension is fused into the batch dimension to obtain the input data, and then the data of each channel in the input data is normalized, and the normalized result is used as the input of the best spectral prediction model. The batch dimension of the output result of the best spectral prediction model is split and restored to the original channel dimension to obtain the final prediction result.

[0119] Specific implementation three: combination Figure 5 This embodiment is described. The difference between this embodiment and specific implementation one or two is that the improved Autoformer model comprises an encoder and a decoder, and the encoder comprises N en encoding units, and the decoder comprises N de decoding units.

[0120] In the improved Autoformer model, the input of the improved Autoformer model is subjected to a linear mapping layer, the output of the linear mapping layer is taken as the input of the first encoding unit, the seasonal part output by the first encoding unit is taken as the input of the second encoding unit, the seasonal part output by the second encoding unit is taken as the input of the third encoding unit, and so on, the seasonal part output by the N en -1 encoding unit is taken as the input of the N en th encoding unit, and the seasonal part output by the N en th encoding unit is obtained.

[0121] The input of the improved Autoformer model is decomposed into the cumulative structure of the trend cycle component and the cumulative structure of the seasonal component, the cumulative structure of the trend cycle component, the cumulative structure of the seasonal component and the seasonal part output by the N en th encoding unit obtained by decomposition are taken as the input of the first decoding unit, the trend cycle part, the seasonal part output by the first decoding unit and the seasonal part output by the N en th encoding unit are taken as the input of the second decoding unit, and so on, the trend cycle part, the seasonal part output by the N de -1 decoding unit and the seasonal part output by the N en th encoding unit are taken as the input of the N de th decoding unit, and the frequency spectrum data prediction result is obtained according to the output of the N de th decoding unit.

[0122] The other steps and parameters are the same as those in specific implementation one or two.

[0123] To ensure the model can stably perform multi-step predictions, the present invention uses an improved Autoformer model structure as the spectrum prediction model. The improved Autoformer model is an improved model based on the Transformer, designed to address the limitations of the traditional Transformer model in time series processing. The improved Autoformer model decomposes the time series into a trend cycle component and a seasonal component, capturing long-term trends and short-term fluctuations through explicit modeling, making it more suitable for processing complex time series data. The number of encoding units and decoding units in the model of the present invention can be any integer.

[0124] Specific embodiment 4: This embodiment differs from any one of specific embodiments 1 to 3 in that the working process of the encoder is as follows:

[0125] First, the processed historical spectrum data sequence X′ is dimensionally expanded to d dimensions, and the dimension expansion result X is obtained. en for:

[0126]

[0127] Among them, Embedding(·) represents the linear mapping layer; the dimension expansion process is equivalent to the process before input encoder;

[0128] Then the output X of the linear mapping layer en As the input of the first coding unit, in the first coding unit, the input of the first coding unit is used as the input of the first correlation extraction module, then the output X of the first correlation extraction module is (1) for:

[0129] X (1) =Auto-Correlation1(X en )

[0130] Wherein, Auto-Correlation1(·) represents the operation of the first correlation extraction module; seasonal information can be extracted through the correlation extraction module;

[0131] The output of the first correlation extraction module is compared with X en Add, and the result is X (2) for:

[0132] X (2) =Auto-Correlation1(X en )+X en

[0133] Residual connections can alleviate the problems of gradient vanishing and gradient exploding, and promote information flow;

[0134] The addition result X (2) As the input of the first sequence decomposition module, the trend cycle part X t(3) and the seasonal part X s(3) are output by the first sequence decomposition module.

[0135] The encoder part mainly performs modeling of the seasonal part, therefore, the input of the forward propagation module only contains the seasonal part, and the seasonal part X s(3) is taken as the input of the first forward propagation module, and the output X (4) of the first forward propagation module is:

[0136] X (4) = FFN1(X s(3) )

[0137] where FFN1(·) represents the operation of the first forward propagation module.

[0138] The output of the first forward propagation module is added with the seasonal part X s(3) , and the addition result X (5) is obtained.

[0139] X (5) = X (4) + X s(3)

[0140] The trend cycle part and the seasonal part are further separated, the addition result X (5) is taken as the input of the second sequence decomposition module, the trend cycle part output by the second sequence decomposition module is denoted as X t(6) , the seasonal part output by the second sequence decomposition module is denoted as X s(6) , and X s(6) is taken as the seasonal part output by the first encoding unit (X s(6) , that is, the output of the first encoding unit);

[0141] The seasonal part X s(6) output by the first encoding unit is taken as the input of the second encoding unit (X s(6) , which is equivalent to X en for the first encoding unit); the seasonal part output by the second encoding unit is taken as the input of the third encoding unit again.

[0142] In this way, the seasonal part output by the N en -1th encoding unit is taken as the input of the N en th encoding unit, and the seasonal part output by the N en th encoding unit is obtained.

[0143] The other steps and parameters are the same as those in the first to third embodiments.

[0144] The present invention is to en By stacking coding units with the same structure, we can learn deeper seasonal fluctuations and separate deeper trend cycle components.

[0145] Specific embodiment 5: This embodiment differs from any one of specific embodiments 1 to 4 in that the cumulative structure of the trend cycle component and the cumulative structure of the seasonal component are obtained by:

[0146]

[0147] Among them, the processed historical spectrum data sequence X′ is passed through the sequence decomposition module to obtain the historical trend cycle sequence and historical seasonal sequence. It is the last H / 2 row of the historical trend cycle sequence. represents the last H / 2 rows of the historical seasonal series, mean(X′ H ) P Represents a vector of P rows and 1 column consisting of the mean of each row in the historical spectrum data sequence X′ (for any channel in the processed historical spectrum data sequence, the historical trend period data corresponding to the channel is equivalent to a vector of H rows and 1 column. First, select the data of the last H / 2 rows in the vector, then calculate the mean of the H rows of data corresponding to the channel, and use the selected H / 2 rows of data as the first H / 2 rows of the column vector, that is, and in A vector with P rows and 1 column is added after the supplementary vector, and the value of each element in the supplementary vector is the calculated mean. Similarly, each channel is processed separately, and each channel is processed in parallel). P Represents a 0 vector with P rows and 1 columns (the processing process is similar to the historical trend period data, the difference is The elements of the vector added after are all 0), and we get and After that, the linear mapping layer Embedding(·) is used to expand the dimension. represents the cumulative structure of the trend cycle component, Represents the cumulative structure of the seasonal component.

[0148] The other steps and parameters are the same as those in the first to fourth embodiments.

[0149] Specific embodiment 6: This embodiment differs from any one of specific embodiments 1 to 5 in that the working process of the decoder is as follows:

[0150] Will and Nth enThe seasonal part outputted by the encoding unit As the input of the first decoding unit, in the first decoding unit, As the input of the second correlation extraction module, the output of the second correlation extraction module is further added with The addition result is taken as the input of the third sequence decomposition module, and the output of the third sequence decomposition module is X s(7) and X t(7) :

[0151]

[0152] Wherein, Auto-Correlation2(·) represents the operation of the second correlation extraction module, and SeriesDecomp3(·) represents the operation of the third sequence decomposition module;

[0153] X s(7) and As the input of the third correlation extraction module, the output of the third correlation extraction module is added with and X s(7) The addition result is taken as the input of the fourth sequence decomposition module, and the output of the fourth sequence decomposition module is X s(8) and X t(8) :

[0154]

[0155] Wherein, Auto-Correlation3(·) represents the operation of the third correlation extraction module, and SeriesDecomp4(·) represents the operation of the fourth sequence decomposition module;

[0156] X s(8) As the input of the second forward propagation module, the output of the second forward propagation module is added with X s(8) The addition result is taken as the input of the fifth sequence decomposition module, and the output of the fifth sequence decomposition module is X s(9) and X t(9) :

[0157] X s(9) ,X t(9) = SeriesDecomp5(FFN2(X s(8) )+X s(8) )

[0158] Wherein, FFN2(·) represents the operation of the second forward propagation module, and SeriesDecomp5(·) represents the operation of the fifth sequence decomposition module;

[0159] X X t(7) , Xt(8) and X t(9) Add together to get the sum of the trend cycle components

[0160]

[0161] X s(9) As the seasonal part of the output of the first decoding unit, As the trend cycle part of the first decoding unit output, X s(9) and As the input of the second decoding unit (X s(9) Similar to the input of the first decoding unit Similar to the input of the first decoding unit );

[0162] And so on, we get the Nth de The seasonal part and trend cycle part of the decoding unit output; de The seasonal part of the decoding unit output is denoted as The Nth de The trend cycle part of the decoding unit output is recorded as Then the Nth de The trend cycle part of the decoding unit output and seasonal parts merge:

[0163]

[0164] Among them, Y de Indicates the merge result;

[0165] Then merge the result Y de Mapping is performed to obtain the spectrum data prediction result Y′:

[0166] Y′=W o Y de +b o

[0167] Among them, W o and b o are the learnable parameters of the model.

[0168] The other steps and parameters are the same as those in the first to fifth embodiments.

[0169] Restore the spectrum data prediction result Y′ to the original channel and get Prediction method such as Figure 4 As shown. By deStacking multiple identical decoding units can model more complex internal relationships. Each decoding unit contains two correlation extraction modules, one for modeling the future seasonal component and the other for extracting the relationship with the past seasonal component. Through the combination of the two, the time dependence of the future prediction time step is mined using the periodic nature based on the sequence. The information of the trend part is obtained by accumulating the trend of the decoding unit input to obtain the future trend period.

[0170] In the improved Autoformer model, a residual connection (i.e., adding the output of the current layer to the value input to the layer) is applied, which can ensure that the effect of a deep network is better than that of a shallow network.

[0171] Specific embodiment seven: different from one of the specific embodiments one to six, the working process of the first correlation extraction module (Auto-Correlation) is:

[0172] Step 1, X en After linear mapping, the sequences q, k and v are obtained, which can represent X en More information;

[0173] Step 2, initialize τ = 1;

[0174] Step 3, according to the time delay τ, the sequence v is right circularly shifted to obtain the right circularly shifted result Roll(v, τ);

[0175] Then calculate the autocorrelation function of X en

[0176]

[0177] Where q t represents the t-th element in the sequence q, and k t-τ represents the t-τ element in the sequence k;

[0178] Step 4, judge whether τ = H is satisfied;

[0179] If satisfied, continue to execute step 5;

[0180] If not satisfied, τ = τ + 1, return to execute step 3;

[0181] Step 5, according to Select the strongest k delays:

[0182]

[0183] ​wherein τ1, τ2,..., τ k represent the selected k delays; the first k delays with larger values of are selected, and the k delays represent the sequence X en the most likely period length, and the autocorrelation function corresponding to each delay, i.e., the likelihood of each period;

[0184] The autocorrelation functions corresponding to the selected k delays are processed:

[0185]

[0186] wherein respectively represent the k delays τ1, τ2,..., τ k corresponding weights, and softmax(·) is a normalized exponential function;

[0187] The output of the first correlation extraction module is Auto-Correlation(q, k, v):

[0188]

[0189] The other steps and parameters are the same as one of the first to sixth embodiments.

[0190] Embodiment Eight Figure 6 This embodiment is different from one of the first to seventh embodiments in that the first correlation extraction module is calculated according to FFT and IFFT, and the specific calculation method is:

[0191] Step 1, taking the sequence q as the input of the first linear layer, taking the sequence k as the input of the second linear layer, and taking the sequence v as the input of the third linear layer;

[0192] Step 2, the output of the first linear layer is passed through the first FFT module, the output of the second linear layer is passed through the second FFT module, the output of the first FFT module is multiplied by the output of the second FFT module, and the multiplication result is taken as the input of the first IFFT module, and the autocorrelation function values corresponding to each delay are output through the first IFFT module

[0193] The first k large autocorrelation function values are selected (according to the order from large to small of the autocorrelation function values, the k autocorrelation function values in the front are selected as the selected k delays), and the selected k delays are respectively denoted as τ1, τ2,..., τ k ;

[0194] ​Step 3: Delay the output of the third linear layer by τ1, τ2, ..., τ k Delay processing (i.e., performing a τ1-bit right cyclic shift on the output of the third linear layer, performing a τ2-bit right cyclic shift on the output of the third linear layer, ..., performing a τ1-bit right cyclic shift on the output of the third linear layer) k The rightward circular shift of bits is equivalent to obtaining Roll(v,τ1), Roll(v,τ2),…, Roll(v,τ k )), and we get the delays τ1, τ2, ..., τ k Corresponding delay processing results;

[0195] Step 4: multiply the delay processing result corresponding to delay τ1 by the autocorrelation function value corresponding to delay τ1, multiply the delay processing result corresponding to delay τ2 by the autocorrelation function value corresponding to delay τ2, ..., multiply the delay τ k Corresponding delay processing results and delay τ k The corresponding autocorrelation function values ​​are multiplied to obtain the delay τ1, τ2, ..., τ k The corresponding multiplication results are then accumulated to obtain the accumulated result;

[0196] The accumulated result is passed through the fourth linear layer, and the output of the fourth linear layer is used as the output of the first correlation extraction module.

[0197] The other steps and parameters are the same as those in the first to seventh embodiments.

[0198] For cross-autocorrelation The calculation is to convert X s(7) After linear mapping, the sequence q is obtained as Figure 6 The q in After linear mapping, the sequences k and v are obtained as Figure 6 The k and v in .

[0199] Specific embodiment 9: This embodiment differs from any one of specific embodiments 1 to 8 in that the working process of the first sequence decomposition module is as follows:

[0200] X t(3) =AvgPool(Padding(X (2) ))

[0201] X s(3) =X (2) -X t(3)

[0202] Among them, Padding(·) represents the data padding operation (padding is performed to keep the sequence length unchanged), AvgPool(·) represents the moving average operation, denotes a trend periodic part of the first sequence decomposition module output, denotes a seasonal part of the first sequence decomposition module output.

[0203] The other steps and parameters are the same as one of embodiments 1-8.

[0204] The first sequence decomposition module can adjust the moving average line to smooth the periodic fluctuations and highlight the long-term trend. The working processes of each sequence decomposition module in the present application are the same.

[0205] Embodiment 10: The difference between this embodiment and one of embodiments 1-9 is that the working process of the first forward propagation module is:

[0206] FFN1(X s(3) )=w2relu(w1X s(3) +b1)+b2

[0207] where FFN1(·) denotes the operation of the first forward propagation module, w1, w2, b1 and b2 are model learnable parameters, and relu(·) is an activation function.

[0208] The other steps and parameters are the same as one of embodiments 1-9.

[0209] The working processes of each forward propagation module in the present application are the same.

[0210] Experimental part

[0211] Below, we take the open-source spectrum dataset from Aachen University of Technology in Germany as an example, which contains data from different places and frequency bands. We selected data collected in a residential area in Maastricht, Netherlands, with a center frequency of 770 MHz and a bandwidth of 1500 MHz. These data are divided into 8192 channels with an interval of 200 kHz, with slight overlap between adjacent channels. In order to facilitate analysis and not to affect the problems existing in the actual environment, we selected 150 channels for description. The two-dimensional and three-dimensional power spectrum degrees of the data are shown in Figure 1 and Figure 2 .

[0212] The selected 150 channels are divided into training set, validation set and test set in the ratio of 7:2:1. In order to verify the effectiveness of the channel prediction method of the present application, the Seq-to-Seq model based on LSTM, the bidirectional LSTM model based on attention mechanism, the fusion model of CNN and LSTM and the improved Autoformer model (N en =1,N de=1) multi-channel joint prediction, single-channel prediction and the prediction of the method of the present application are performed, and the effects are compared. Among the four models, Adam, which is widely used and excellent, is used as the optimizer, and the initial learning rate is set to 10 -4 , the loss function is mean square error loss, and MAE and MSE are selected as evaluation indexes. In all models, the number of hidden units is set to 128, the learning rate of the optimizer is reduced by half every round, and the early stopping mechanism is set, with patience set to 10. In each model, the window length corresponding to the lowest MSE is selected, i.e. the required historical time length. Since the bidirectional LSTM model based on attention mechanism and the fusion model of CNN and LSTM can only perform single-step prediction, the present application takes single-step prediction as an example to verify the effectiveness of the prediction method of the present application.

[0213] The average MSE and MAE of all channels on the test set of the above models are compared, as shown in Figure 7 and Figure 8 . When the Seq-to-Seq model based on LSTM is used for multi-channel joint prediction, the average prediction MSE of all channels is 0.4007, and the average prediction MAE is 0.4750; when single-channel prediction is performed, the average prediction MSE of all channels is 0.0912, and the average prediction MAE is 0.2023; and when the prediction method of the present application is used, the average prediction MSE of all channels is 0.0895, and the average prediction MAE is 0.2015. When the bidirectional LSTM model based on attention mechanism is used for multi-channel joint prediction, the average prediction MSE of all channels is 0.3918, and the average prediction MAE is 0.4721; when single-channel prediction is performed, the average prediction MSE of all channels is 0.0918, and the average prediction MAE is 0.2033; and when the prediction method of the present application is used, the average prediction MSE of all channels is 0.0894, and the average prediction MAE is 0.2020. When the fusion model of CNN and LSTM is used for multi-channel joint prediction, the average prediction MSE of all channels is 0.3975, and the average prediction MAE is 0.4766; when single-channel prediction is performed, the average prediction MSE of all channels is 0.0959, and the average prediction MAE is 0.2089; and when the prediction method of the present application is used, the average prediction MSE of all channels is 0.0923, and the average prediction MAE is 0.2076. When the improved Autoformer model is used for multi-channel joint prediction, the average prediction MSE of all channels is 0.1221, and the average prediction MAE is 0.2514; when single-channel prediction is performed, the average prediction MSE of all channels is 0.0921, and the average prediction MAE is 0.2087; and when the prediction method of the present application is used, the average prediction MSE of all channels is 0.0898, and the average prediction MAE is 0.2037.

[0214] It can be seen that the loss of the four models when using the prediction method of the application is much lower than that of multi-channel joint prediction, indicating that the prediction method of the application can avoid the problem of mutual interference between channels and large difference in signal distribution of each channel. In addition, the loss of the method proposed in the application is further reduced than that of single-channel prediction, and the future information of all channels can be predicted at the same time, which shows that the prediction method of the application has stronger generalization ability and higher efficiency. Since the improved Autoformer model is more suitable for long-term prediction and is not prone to problems of gradient disappearance and gradient explosion, the improved Autoformer is selected as the prediction model, thereby verifying the effectiveness of the method of the application.

[0215] In summary, the application proposes a multi-channel joint prediction method that can avoid channel interference and large difference in channel data distribution. First, the channel dimension is split and fused into the batch dimension, and then an arbitrary spectrum prediction model is used for prediction. The method of the application can be regarded as training a common model for all channels, which will learn the most core conditional distribution P(Y|X) in each channel, avoiding the model being more biased towards any channel, thereby improving the generalization ability of the model. At the same time, the method of the application can predict the future information of all channels at once, thereby solving the problem that single-channel prediction can only predict one channel at a time, and the number of models that need to be trained and stored is proportional to the number of channels, which will seriously waste time and storage resources when the number of channels is large. Compared with multi-channel joint prediction, since the data distribution of each channel is significantly different, directly performing multi-channel joint prediction on the unprocessed data will result in a large increase in prediction loss. In addition, when there is a misalignment problem of channel correlation, the multi-channel joint prediction model will incorrectly learn these correlations, thereby interfering with the channel to be predicted. The method proposed in the application can well avoid these problems, and under the condition of sufficient spectrum data, it can make full use of time correlation to obtain a lower loss prediction result. The application uses an improved Autoformer that can perform stable long-term prediction as a spectrum prediction model, avoiding the problems of gradient disappearance and low efficiency of multi-step recursive prediction in LSTM in multi-step prediction. The low-loss prediction result can help new users identify idle or underutilized spectrum, thereby realizing dynamic spectrum access and improving overall spectrum utilization.

[0216] The above examples of the application are only used to illustrate the calculation model and calculation process of the application, and are not intended to limit the embodiments of the application. Based on the above description, those skilled in the art can make other different forms of changes or modifications, and it is impossible to exhaust all embodiments here. Any obvious changes or modifications derived from the technical solutions of the application still fall within the protection scope of the application.

Claims

1. A spectrum prediction method for channel separation, characterized in that: The method specifically comprises the following steps: Step 1: Define the spectrum data that needs to be predicted for the next P time slots, and build the corresponding spectrum prediction model based on the value of P; And define the number of historical time slots H∈O corresponding to the historical spectrum data sequence input by the spectrum prediction model, where O represents the set of all values ​​of H; Step 2: Record the historical spectrum data sequence input into the spectrum prediction model as in, represents a real number, M represents the number of channels that need to be predicted for spectrum data in each batch, and B represents the number of batches; The spectrum data of each channel in the historical spectrum data sequence X is normalized separately, and the channel dimension of the normalized input data is fused to the batch dimension to obtain the processed historical spectrum data sequence Step 3: Use the processed historical spectrum data sequence X′ to train the spectrum prediction model to obtain the trained spectrum prediction model corresponding to the current H value; Step 4: For each value of H in the set O, execute steps 2 and 3, that is, obtain the trained spectrum prediction model corresponding to each H value; Then, the spectrum prediction model with the smallest mean square error loss is selected from the trained spectrum prediction models corresponding to each H value; Step 5: After processing the historical spectrum data sequence of the channel to be predicted, the processed data is input into the selected spectrum prediction model to obtain the spectrum data prediction result of the channel to be predicted; The batch dimension of the spectrum data prediction result is then split and restored to the original channel dimension to obtain the final prediction result.

2. The spectrum prediction method for channel separation according to claim 1, characterized in that: The corresponding spectrum prediction model is constructed according to the value of P, specifically: When P=1, the constructed spectrum prediction model is a LSTM-based Seq2Seq model or an improved Autoformer model; When P>1, the constructed spectrum prediction model is an improved Autoformer model.

3. The spectrum prediction method for channel separation according to claim 2, characterized in that: The improved Autoformer model includes an encoder and a decoder, and the encoder includes N en The decoder includes N coding units. de decoding units; In the improved Autoformer model, the input of the improved Autoformer model is passed through the linear mapping layer, the output of the linear mapping layer is used as the input of the first encoding unit, and the seasonal part of the output of the first encoding unit is used as the input of the second encoding unit, and the seasonal part of the output of the second encoding unit is used as the input of the third encoding unit, and so on. en -1 The seasonal part of the output of the coding unit is used as the Nth en The input of the encoding unit is obtained en The seasonal component of the encoding unit output; The input of the improved Autoformer model is decomposed into the cumulative structure of the trend cycle component and the cumulative structure of the seasonal component, and the cumulative structure of the decomposed trend cycle component, the cumulative structure of the seasonal component and the Nth en The seasonal part output by the encoding unit is used as the input of the first decoding unit, and the trend cycle part, seasonal part and the Nth en The seasonal part of the output of the encoding unit is used as the input of the second decoding unit, and so on. de -1 The trend cycle part, seasonal part and Nth part of the decoding unit output en The seasonal part of the encoding unit output is used as the Nth de The input of the decoding unit is based on the Nth de The output of the decoding unit obtains the spectrum data prediction result.

4. The spectrum prediction method for channel separation according to claim 3, characterized in that: The working process of the encoder is: First, the processed historical spectrum data sequence X′ is dimensionally expanded to d dimensions, and the dimension expansion result X is obtained. en for: Among them, Embedding(·) represents the linear mapping layer; Then the output X of the linear mapping layer en As the input of the first coding unit, in the first coding unit, the input of the first coding unit is used as the input of the first correlation extraction module, then the output X of the first correlation extraction module is (1) for: X (1) =Auto-Correlation1(X en ) Wherein, Auto-Correlation1(·) represents the operation of the first correlation extraction module; The output of the first correlation extraction module is compared with X en Add, and the result is X (2) for: X (2) =Auto-Correlation1(X en )+X en Add the result X (2) As the input of the first sequence decomposition module, the trend cycle part X is output by the first sequence decomposition module t(3) and seasonal part X s(3) ; The seasonal part X s(3) As the input of the first forward propagation module, the output X of the first forward propagation module is (4) for: X (4) =FFN1(X s(3) ) Where FFN1(·) represents the operation of the first forward propagation module; The output of the first forward propagation module is combined with the seasonal part X s(3) Add and get the result X (5) : X (5) =X (4) +X s(3) Add the result X (5) As the input of the second sequence decomposition module, the trend cycle part of the second sequence decomposition module output is recorded as X t(6) , the seasonal part of the output of the second sequence decomposition module is recorded as X s(6) , and X s(6) The seasonal component as the output of the first coding unit; Then the seasonal part X output by the first encoding unit is s(6) As the input of the second coding unit; the seasonal part output by the second coding unit is used as the input of the third coding unit; And so on, the Nth en -1 The seasonal part of the output of the coding unit is used as the Nth en The input of the encoding unit is obtained en Seasonal component of the encoding unit output 5. The spectrum prediction method for channel separation according to claim 3, characterized in that: The cumulative structure of the trend cycle component and the cumulative structure of the seasonal component are obtained as follows: Among them, the processed historical spectrum data sequence X′ is passed through the sequence decomposition module to obtain the historical trend cycle sequence and historical seasonal sequence. It is the last H / 2 row of the historical trend cycle sequence. represents the last H / 2 rows of the historical seasonal series, mean(X′ H ) P represents a vector of P rows and 1 column consisting of the mean values ​​of each row in the historical spectrum data sequence X′, O P represents a 0 vector with P rows and 1 columns, represents the cumulative structure of the trend cycle component, Represents the cumulative structure of the seasonal component.

6. The spectrum prediction method for channel separation according to claim 4, characterized in that: The working process of the decoder is: Will and Nth en Seasonal component of the encoding unit output As the input of the first decoding unit, in the first decoding unit, As the input of the second correlation extraction module, the output of the second correlation extraction module is combined with Add, and use the addition result as the input of the third sequence decomposition module, and output X through the third sequence decomposition module s (7) and X t(7) : Wherein, Auto-Correlation2(·) represents the operation of the second correlation extraction module, and SeriesDecomp3(·) represents the operation of the third series decomposition module; X s(7) and As the input of the third correlation extraction module, the output of the third correlation extraction module and X s(7) Add, and use the addition result as the input of the fourth sequence decomposition module to obtain the output X of the fourth sequence decomposition module s(8) and X t(8) : Among them, Auto-Correlation3(·) represents the operation of the third correlation extraction module, and SeriesDecomp4(·) represents the operation of the fourth series decomposition module; X s(8) As the input of the second forward propagation module, the output of the second forward propagation module is combined with X s(8) Add, and use the added result as the input of the fifth sequence decomposition module to obtain the output X of the fifth sequence decomposition module s(9) and X t(9) : X s(9) ,X t(9) =SeriesDecomp5(FFN2(X s(8) )+X s(8) ) Where FFN2(·) represents the operation of the second forward propagation module; SeriesDecomp5(·) represents the operation of the fifth sequence decomposition module; Will X t(7) 、X t(8) and X t(9) Add together to get the sum of the trend cycle components X s(9) As the seasonal part of the output of the first decoding unit, As the trend cycle part of the first decoding unit output, X s(9) and as input to the second decoding unit; And so on, we get the Nth de The seasonal part and trend cycle part of the decoding unit output; de The seasonal part of the decoding unit output is denoted as The Nth de The trend cycle part of the decoding unit output is recorded as Then the Nth de The trend cycle part of the decoding unit output and seasonal parts merge: Among them, Y de Indicates the merge result; Then merge the result Y de Mapping is performed to obtain the spectrum data prediction result Y′: Y′=W o Y de +b o Among them, W o and b o are model parameters.

7. The spectrum prediction method for channel separation according to claim 4, characterized in that: The working process of the first correlation extraction module is as follows: Step 1. en After linear mapping, we get the sequences q, k and v; Step 2: Initialize τ=1; Step 3: cyclically shift the sequence v to the right according to the time delay τ to obtain the result Roll(v,τ) after the right cyclic shift. Calculate X again en The autocorrelation function Among them, q t represents the tth element in the sequence q, k t-τ represents the t-τth element in sequence k; Step 4: Determine whether τ=H is satisfied; If satisfied, proceed to step 5; If not, set τ = τ + 1 and return to step 3; Step 5: According to Select the k most correlated delays: Among them, τ1, τ2, ..., τ k represents the selected k delays; For the selected k delays corresponding to Processing: in, Represent k delays τ1, τ2, ..., τ k The corresponding weights, softmax(·) is the normalized exponential function; Then the output of the first correlation extraction module is Auto-Correlation(q,k,v):

8. The spectrum prediction method for channel separation according to claim 7, characterized in that: The first correlation extraction module performs calculations based on FFT and IFFT. The specific calculation method is: Step 1: Use sequence q as the input of the first linear layer, sequence k as the input of the second linear layer, and sequence v as the input of the third linear layer; Step 2: Pass the output of the first linear layer through the first FFT module, pass the output of the second linear layer through the second FFT module, multiply the output of the first FFT module by the output of the second FFT module, and use the multiplication result as the input of the first IFFT module. The autocorrelation function value corresponding to each delay is output through the first IFFT module. Then select the delays corresponding to the first k largest autocorrelation function values, and record the selected k delays as τ1, τ2, ..., τ k ; Step 3: Delay the output of the third linear layer by τ1, τ2, ..., τ k The delay processing is to obtain delays τ1, τ2, ..., τ k Corresponding delay processing results; Step 4: multiply the delay processing result corresponding to delay τ1 by the autocorrelation function value corresponding to delay τ1, multiply the delay processing result corresponding to delay τ2 by the autocorrelation function value corresponding to delay τ2, ..., multiply the delay τ k Corresponding delay processing results and delay τ k The corresponding autocorrelation function values ​​are multiplied to obtain the delays τ1, τ2, ..., τ k The corresponding multiplication results are then accumulated to obtain the accumulated result; The accumulated result is passed through the fourth linear layer, and the output of the fourth linear layer is used as the output of the first correlation extraction module.

9. The spectrum prediction method for channel separation according to claim 4, characterized in that: The working process of the first sequence decomposition module is as follows: X t(3) =AvgPool(Padding(X (2) )) X s(3) =X (2) -X t(3) Among them, Padding(·) represents the data filling operation, AvgPool(·) represents the moving average operation, Represents the trend cycle part of the first sequence decomposition module output, Represents the seasonal part of the output of the first series decomposition module.

10. The spectrum prediction method for channel separation according to claim 4, characterized in that: The working process of the first forward propagation module is: FFN1(X s(3) )=w2relu(w1X s(3) +b1)+b2 Where FFN1(·) represents the operation of the first forward propagation module, w1, w2, b1 and b2 are model parameters, and relu(·) is the activation function.

Images