CN118233035A - Multiband spectrum prediction method and system based on graph convolution inversion transform - Google Patents
Multiband spectrum prediction method and system based on graph convolution inversion transform Download PDFInfo
- Publication number
- CN118233035A CN118233035A CN202410658449.0A CN202410658449A CN118233035A CN 118233035 A CN118233035 A CN 118233035A CN 202410658449 A CN202410658449 A CN 202410658449A CN 118233035 A CN118233035 A CN 118233035A
- Authority
- CN
- China
- Prior art keywords
- multiband
- prediction
- data
- convolution
- graph
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001228 spectrum Methods 0.000 title claims abstract description 66
- 238000000034 method Methods 0.000 title claims abstract description 44
- 239000011159 matrix material Substances 0.000 claims abstract description 39
- 230000003595 spectral effect Effects 0.000 claims abstract description 29
- 238000005096 rolling process Methods 0.000 claims abstract description 13
- 238000004364 calculation method Methods 0.000 claims abstract description 12
- 238000007781 pre-processing Methods 0.000 claims abstract description 11
- 238000000605 extraction Methods 0.000 claims abstract description 8
- 230000006870 function Effects 0.000 claims description 14
- 230000007246 mechanism Effects 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims description 8
- 239000013598 vector Substances 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 230000010354 integration Effects 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 description 12
- 238000004891 communication Methods 0.000 description 10
- 230000000694 effects Effects 0.000 description 5
- 230000008901 benefit Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 238000007405 data analysis Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- 230000008521 reorganization Effects 0.000 description 2
- 238000013468 resource allocation Methods 0.000 description 2
- 238000012731 temporal analysis Methods 0.000 description 2
- 230000036962 time dependent Effects 0.000 description 2
- 238000000700 time series analysis Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000000969 carrier Substances 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 238000013524 data verification Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008846 dynamic interplay Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/391—Modelling the propagation channel
- H04B17/3913—Predictive models, e.g. based on neural network models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/382—Monitoring; Testing of propagation channels for resource allocation, admission control or handover
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Electromagnetism (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Complex Calculations (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The invention relates to the technical field of spectrum prediction, in particular to a multiband spectrum prediction method and a multiband spectrum prediction system based on a graph convolution inversion transform. The method includes acquiring multiband spectral data; carrying out data preprocessing on the multiband spectrum data; constructing a graph rolling network, and performing feature extraction on multiband spectrum data by using the graph rolling network to obtain an adjacent matrix; performing self-attention calculation on the adjacent matrix combined with the multiband spectrum data by using an inverted transducer model, and outputting a prediction result; and verifying the output prediction result to obtain a final prediction result. The method not only improves the accuracy of prediction, but also enhances the prediction capability of the model on future spectrum use trend.
Description
Technical Field
The invention relates to the technical field of spectrum prediction, in particular to a multiband spectrum prediction method and a multiband spectrum prediction system based on a graph convolution inversion transform.
Background
Graph convolution (Graph Convolution): graph convolution aims to update a feature representation of a node by aggregating neighborhood information of the node (i.e., the direct connection points of the node), thereby capturing the local structural features in the graph and the relationships between the nodes. Such operations typically involve a weighted sum of features of each node and its neighbors, which weights may be fixed or learned. The graph rolling network (GCN) can learn complex inter-node dependency relationships by overlapping multiple graph convolution layers, and is widely applied to tasks such as node classification, graph classification, link prediction and the like.
Transformer: a deep learning model architecture based on self-attention mechanisms is initially used to handle Natural Language Processing (NLP) tasks. The transducer is capable of capturing long-range dependencies in input data through its attention mechanism, and is now widely used for a variety of sequential processing tasks, including language, image and time-series analysis.
Inversion of the transducer: based on a variation of the traditional Transformer architecture, the ability of the model to process multivariate time series data is enhanced by changing the strategy of the processing dimension (e.g., treating the time points in the time series as independent variables). This structure is particularly suitable for time series prediction tasks where there is a high degree of correlation between those variables.
Multiband: refers to a scenario in which a plurality of frequency bands or channels are simultaneously considered in wireless communication. The multi-band technique may increase the efficiency of spectrum usage, allowing different communication activities to proceed in parallel on different frequency bands, thereby optimizing overall communication performance and resource allocation.
Radio spectrum prediction: in the field of wireless communications, spectrum prediction involves using historical spectrum usage data to predict spectrum occupancy over a period of time in the future. Accurate spectrum prediction can help to more effectively allocate spectrum resources, reduce interference, and improve communication quality.
In modern communication systems, efficient spectrum management is critical to ensure communication quality and efficiency. With the rapid development of wireless communication technology and the substantial increase in the number of wireless devices, spectrum resources have become invaluable. Therefore, the development of spectrum prediction technology becomes an important research direction for optimizing spectrum resource allocation and utilization.
Spectrum prediction techniques are mainly used to predict spectrum usage over a certain period of time in the future, helping communications carriers and device manufacturers to plan and manage spectrum resources more efficiently. Such predictions are typically implemented in conjunction with various algorithmic models, based on historical spectrum usage data. Conventional predictive models include statistical methods and machine learning techniques, each of which has application scenarios and processing capabilities.
In addition, multi-band spectrum prediction involves processing data for multiple bands or channels simultaneously, which places higher demands on the processing power and data analysis capabilities of the prediction model. In recent years, with the development of artificial intelligence technology, a deep learning model, particularly a neural network-based method, has shown good potential and advantages in the field of multiband spectrum prediction.
However, in the multiband spectrum prediction process, there is a problem that the spatial correlation analysis of spectrum data is insufficient: in multiband spectral prediction, the conventional method often ignores spatial correlation between different frequency bands, resulting in that the prediction model cannot fully exploit potential interactions between the frequency bands. Efficient capture and exploitation of this spatial correlation is critical to improving prediction accuracy.
Limitation of the transducer in time series prediction: while the Transformer model has significant advantages in processing sequence data, its standard architecture often suffers from inherent design limitations (e.g., inadequate processing for long-term dependencies) in the face of complex multi-band time-series data, making it difficult to achieve the desired predictive effect. In particular, in multivariate time series predictions, standard transformers may not be able to effectively distinguish and exploit dynamic relationships between variables in a time series.
Disclosure of Invention
In order to solve the above-mentioned problems, the present invention provides a multiband spectrum prediction method and system based on a graph convolution inversion transform.
In a first aspect, the present invention provides a multiband spectrum prediction method based on a graph convolution inversion transform, which adopts the following technical scheme:
a graph convolution inversion transform-based multiband spectral prediction method, comprising:
acquiring multiband spectrum data;
carrying out data preprocessing on the multiband spectrum data;
constructing a graph rolling network, and performing feature extraction on multiband spectrum data by using the graph rolling network to obtain an adjacent matrix;
Performing self-attention calculation on the adjacent matrix combined with the multiband spectrum data by using an inverted transducer model, and outputting a prediction result;
And verifying the output prediction result to obtain a final prediction result.
Further, the data preprocessing of the multiband spectrum data comprises denoising and signal intensity normalization of the obtained multiband spectrum data.
Further, the building of the graph-convolution network includes defining a relationship between nodes and edges, the nodes representing frequency bands, the edges representing correlations between the frequency bands, wherein the pearson correlation coefficient is used to measure the relationship between the frequency bands, and when the pearson correlation coefficient is higher than a correlation threshold, a value representing the edge in the adjacency matrix is set to 1, and otherwise, is set to 0, by comparing the pearson correlation coefficient with a correlation threshold, and the pearson correlation coefficient is expressed as:
Wherein, Representing Power Spectral Density (PSD) vectors of i and j frequency bands, respectively; /(I)Representing the pearson correlation coefficient,/>Representing covariance function,/>Representing standard deviation.
Further, the constructing of the graph rolling network further includes constructing an adjacency matrix and a feature matrix, and learning the frequency band correlation through the two-layer graph rolling network composed of the adjacency matrix and the feature matrix is expressed as:
Wherein, Is a process combining self-information and completing standardization,/>And/>Representing the weight matrix in the first layer and the second layer, respectively,/>Is a ReLU activation function.
Further, the feature extraction of the multiband spectral data by using the graph rolling network includes capturing spatial correlation between the band data and band features obtained by combining a self-attention mechanism by using the graph rolling network, and performing dimension conversion by combining the band original time sequence data as an input of an inverse transform model.
Further, the self-attention calculation of the extracted features by using the inverted transducer model comprises the steps of carrying out dimension adjustment on the extracted features through a linear transformation layer, matching input dimensions required by an inverted transducer encoder, and obtaining high-order feature representation through feature integration to serve as input of the inverted transducer encoder layer.
Further, the method includes performing self-attention computation on the extracted features by using an inverted transducer model, and further includes taking input features of an inverted transducer encoder layer as queries, keys and values of a multi-head self-attention mechanism, wherein each single-head self-attention mechanism performs self-attention computation on the input queries, keys and value matrices through respective linear transformation to obtain output of each self-attention head.
Further, after verifying the output prediction result, obtaining a final prediction result, which includes using a mean square error as a loss function to measure a difference between a model prediction value and a true value, wherein the output of each self-attention head is subjected to layer normalization through residual connection, and then the output is output through a transducer block after the frequency band is embedded into a full-connection layer through a feedforward network, so as to obtain the prediction result.
In a second aspect, a graph-convolution-inversion-transform-based multiband spectral prediction system includes:
A data acquisition module configured to acquire multiband spectral data;
a preprocessing module configured to perform data preprocessing on multiband spectral data;
the convolution module is configured to construct a graph convolution network, and the graph convolution network is utilized to perform feature extraction on the multiband spectrum data to obtain an adjacent matrix;
the prediction module is configured to perform self-attention calculation on the adjacent matrix combined with the multiband spectrum data by using an inverted transducer model and output a prediction result;
and the verification module is configured to obtain a final prediction result after verifying the output prediction result.
In a third aspect, the present invention provides a computer readable storage medium having stored therein a plurality of instructions adapted to be loaded by a processor of a terminal device and to perform the method of graph-convolution-inversion-transform-based multiband spectral prediction.
In a fourth aspect, the present invention provides a terminal device, including a processor and a computer readable storage medium, where the processor is configured to implement instructions; the computer readable storage medium is for storing a plurality of instructions adapted to be loaded by a processor and to perform the multi-band spectrum prediction method based on a graph convolution inversion transform.
In summary, the invention has the following beneficial technical effects:
1. Enhanced inter-band relationship capture: the convolution enhanced inverted transducer significantly enhances the ability to capture complex relationships between frequencies in multi-band data by combining the spatial relationship modeling ability of the convolution network (GCN) with the time series analysis advantages of the transducer. The method can be used for deeply understanding dynamic interaction between different frequency bands, and provides a basis for more accurate data analysis.
2. Significant improvement of the prediction effect: on the basis of integrating the graph-convolution network, the inverted transducer structure further optimizes the time-dependent processing, so that the overall prediction model shows excellent performance in multiband spectrum prediction. This not only improves the accuracy of the predictions, but also enhances the predictive ability of the model to future spectrum usage trends.
Drawings
Fig. 1 is a schematic diagram of a multiband spectrum prediction method based on a convolution inversion transform according to embodiment 1 of the present invention.
Fig. 2 is a schematic diagram of a multiband spectrum prediction system based on a convolution inversion transform according to embodiment 1 of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
Example 1
Referring to fig. 1, a multiband spectrum prediction method based on a convolution inversion transform of the present embodiment includes:
S1, acquiring data of different frequency bands,
The public radio spectrum data set is cooperatively used by the voice communication service provider, or the spectrum analyzer is deployed to collect spectrum data in a specified frequency bandwidth and time period to record signal intensity with fixed frequency band resolution, the collected data is required to be preprocessed through denoising and signal intensity normalization, and the specific data comprises signal intensity of each frequency band at each time point, namely, the signal intensity sequence of each frequency band in the same time range.
S2, carrying out data preprocessing on the multiband spectrum data, including denoising and signal intensity normalization on the obtained multiband spectrum data, wherein,
1. Denoising signals:
The denoising aim is to reduce noise in data and improve signal quality and signal-to-noise ratio. The invention uses a moving average filtering method to remove noise from data, the moving average filtering smoothes signals by averaging data points in a window, and reduces high-frequency noise, and the formula is as follows:
Wherein, For input multiband spectral data,/>For the filter window size,/>For the denoised signal, i and j are time stamps.
2. Signal intensity normalization:
The purpose of the normalization is to normalize the signal strength to a uniform range to reduce the impact of excessive differences between different frequency bands on subsequent processing. The invention uses maximum-minimum normalization, and the formula is:
Wherein, Is the input multiband spectral data; /(I)Is the minimum value of the signal; /(I)Is the maximum value of the signal; Is the normalized signal.
S3, constructing a graph convolutional network
First, when constructing a graph rolling network, the relationship between nodes and edges needs to be defined: each node represents a specific frequency band, and the edges are determined by correlation among the frequency bands, and the construction result is represented by using an adjacency matrix A. In this context, the pearson correlation coefficient is used to measure the relationship between frequency bands, expressed as:
Wherein, Representing the Power Spectral Density (PSD) vectors for i and j bins, respectively. /(I)Representing the pearson correlation coefficient,/>Representing covariance function,/>Representing standard deviation. By comparing the correlation coefficient with a correlation threshold, the correlation coefficient is set to 1 when it is higher than the threshold, and otherwise set to 0.
Second, the core infrastructure of the GCN includes two matrices: an adjacency matrix a and a feature matrix X. The general graph convolution formula is:
Wherein, Is a degree matrix,/>，/>Is an adjacency matrix,/>Solves the problem of the lack of self-information,Solves the problem of non-standardization,/>Is the output of layer I,/>Is the weight matrix of the first layer,Is a Sigmoid activation function. We use the graph convolution formula to extract the bin correlation. Generally, the number of layers in the GCN should not be too large, because the deep network structure would complicate the information flow and reduce the efficiency of the information transfer in each layer. Therefore, the invention adopts two layers of GCNs to learn the frequency band correlation, and is expressed as follows:
Wherein, Is a process combining self-information and completing standardization,/>And/>Representing the weight matrix in the first layer and the second layer, respectively,/>Is a ReLU activation function.
S4, capturing the frequency band correlation among the multiband data by using the GCN and combining the rich time sequence characteristics obtained by a self-attention mechanism, and modeling the dependency relationship among various time points on a time sequence.
Step 1, feature after graph convolution processingDimension adjustment is performed through a linear transformation layer to match the input dimension required by the inverted transform encoder:
Wherein the method comprises the steps of Is a weight matrix,/>Is an offset term used to convolve the output of the graph with/>Mapping to New feature space/>。
Step 2, feature integration converts the graph feature G obtained by GCN combined with the frequency band original time sequence data into a high-dimensional feature representation which can be processed by a model:
Wherein the method comprises the steps of For inputting time series vectors,/>For the transformed high-dimensional spectral embedding, n represents a specific spectral sequence.
Step 3 of the method, in which the step 3,The embedded tokens with N dimensions D are used directly as input to the transducer encoder layer with the mapped features H. In an encoder, these input features will first be used for the query (Q), key (K) and value (V) of the multi-head self-attention mechanism, expressed in the implementation of multi-head self-attention as:
step 4, the self-attention calculation formula is as follows:
；
MultiHead represents the multi-headed attention mechanism, Q, K, V represent the query, key, and value matrix, respectively, the contents of which are given above. Representing the calculation of the ith attention header, each independently processes Q, K, V. Concat splice together a plurality of attention heads. /(I): And the linear transformation matrix is used for transforming the spliced result to obtain the final multi-head attention output.
Each single-head self-attention mechanism carries out self-attention calculation on input inquiry, keys and value matrix after respective linear transformation to obtain the output of each attention head.A linear transformation matrix of queries, keys and values, respectively, is used to transform the input features into the appropriate dimensional space. The Attention represents a single head self-Attention calculation process.
Wherein each header head _ i corresponds to a portion of the processing input H,Is the dimension of the vector.
The self-attention mechanism decides the weighting of the value matrix by calculating the dot product similarity between the query and the key. The similarity results are normalized by a softmax function and then the value matrices are weighted and summed to obtain the final self-attention output.: Dot product similarity between the query and the key is calculated. /(I): Normalization is performed with the dimensions of the keys to prevent the dot product result from becoming too large. Softmax normalizes the similarity results to a probability distribution. V: a matrix of values representing the actual information to be weighted averaged.
S5, verifying the prediction result
The loss function uses the Mean Square Error (MSE), which is the most commonly used loss function in regression tasks, for measuring the difference between the model predicted and true values, as:
Is the true value of the ith sample,/> And n is the number of samples, namely the length of the predicted sequence, which is the predicted value of the ith sample.
Gradient calculations and parameter updates are automatically performed pytorch.
To improve training stability and efficiency, each self-attention output would be connected by a residual and then layer normalized:
each self-attention layer then goes through a feed-forward network, which embeds the same fully-connected layer for each band independently,
The complete TrmBlock structure is that,
Wherein the method comprises the steps ofIs an input of layer I,/>Is an output and is passed to the next TrmBlock or for final output. Use/>I.e. the output of the last transducer block, performs the final prediction:
Here the number of the elements is the number, And/>Weights and offsets of projection layers respectively, will deep feature space/>Mapping to prediction output space/>And Y is future spectrum data obtained through historical spectrum data, namely a prediction result.
In an inverted transducer, historical spectral data basedPredicting future sequences of each particular spectral dataCan be described simply as:
Wherein the method comprises the steps of An embedded token of dimension D is included, with the superscript denoting the layer index. Embedding function/>And projection function/>There are multi-layer perceptron (MLP) implementations. The resulting variable tokens interact through a self-attention mechanism and are independently processed by the shared feedforward network in each TrmBlock, since the order of the spectral sequence is implicitly stored in the neuronal arrangement of the feedforward network, no location embedding in the standard fransformer is required.
Example 2
The present embodiment provides a multiband spectrum prediction system based on a graph convolution inversion transform, including:
A data acquisition module configured to acquire multiband spectral data;
a preprocessing module configured to perform data preprocessing on multiband spectral data;
the convolution module is configured to construct a graph convolution network, and the graph convolution network is utilized to perform feature extraction on the multiband spectrum data to obtain an adjacent matrix;
the prediction module is configured to perform self-attention calculation on the adjacent matrix combined with the multiband spectrum data by using an inverted transducer model and output a prediction result;
and the verification module is configured to obtain a final prediction result after verifying the output prediction result.
Such as the model shown in fig. 2, wherein,
1. Raw spectrogram data: the leftmost spectrogram represents a visualization of multi-band data, typically the representation of time series data at different frequencies. This is the input to the model, demonstrating the change in data over time over different frequency bands.
2. Graph roll-up network (GCN): the next modules are represented using multi-color blocks representing different band nodes, with the edges between the nodes represented by line connections. The GCN is used to capture the frequency domain relationship between the frequency bands. The connections in the figure represent dependencies or interactions between different frequency bands, through which the GCN learns the complex interactions between the frequency bands.
3. Inversion of the transducer: the modules following the GCN represent the structure of an inverted transducer, which contains a plurality of TramBlock blocks. The inverse transform is used to process features obtained through the GCN, enhancing the modeling ability of time series data. The GCN extracted features are further encoded, and the dependency relationship in the time dimension is extracted.
4. Data stream reorganization and feature fusion: the layered structure near the end of the figure represents the reorganization and integration of the features after the inverted transducer treatment. This section involves reordering or fusing the processed data for final prediction or classification.
5. Output layer: the rightmost module represents the final output layer for generating the prediction result. This part is responsible for mapping all processed features to the final output, e.g. class labels or predictors.
And (3) data verification:
table 1 comparison of predictive effects
As shown in table 1, the inverted transfomer structure further optimizes the time-dependent processing on the basis of integrating the graph-convolution network, such that the overall predictive model exhibits excellent performance in multi-band spectrum prediction. This not only improves the accuracy of the predictions, but also enhances the predictive ability of the model to future spectrum usage trends.
A computer readable storage medium having stored therein a plurality of instructions adapted to be loaded by a processor of a terminal device and to perform the method.
A terminal device comprising a processor and a computer readable storage medium, the processor configured to implement instructions; the computer readable storage medium is for storing a plurality of instructions adapted to be loaded by a processor and to perform the method.
The above embodiments are not intended to limit the scope of the present invention, so: all equivalent changes in structure, shape and principle of the invention should be covered in the scope of protection of the invention.
Claims (10)
1. A method for multiband spectral prediction based on a convolution inversion transform, comprising:
acquiring multiband spectrum data;
carrying out data preprocessing on the multiband spectrum data;
constructing a graph rolling network, and performing feature extraction on multiband spectrum data by using the graph rolling network to obtain an adjacent matrix;
Performing self-attention calculation on the adjacent matrix combined with the multiband spectrum data by using an inverted transducer model, and outputting a prediction result;
And verifying the output prediction result to obtain a final prediction result.
2. The method of claim 1, wherein the constructing a graph-rolling network comprises defining a relationship between nodes and edges, the nodes representing frequency bands, the edges representing correlations between the frequency bands, wherein the relationships between the frequency bands are measured by pearson correlation coefficients, wherein values representing edges in an adjacency matrix are set to 1 when the pearson correlation coefficients are higher than a correlation threshold by comparing the pearson correlation coefficients to a correlation threshold, and otherwise are set to 0, the pearson correlation coefficients being expressed as:
Wherein, Representing Power Spectral Density (PSD) vectors of i and j frequency bands, respectively; /(I)Representing the pearson correlation coefficient,Representing covariance function,/>Representing standard deviation.
3. The method for multiband spectral prediction based on the convolution inversion transform according to claim 2, wherein the constructing the convolution network further comprises constructing an adjacency matrix and a feature matrix, and learning the band correlation by the two-layer convolution network composed of the adjacency matrix and the feature matrix is expressed as:
Wherein, Is a process combining self-information and completing standardization,/>And/>Representing the weight matrix in the first layer and the second layer, respectively,/>Is a ReLU activation function.
4. A method for multiband spectral prediction based on a convolution inversion transform according to claim 3, wherein the feature extraction of multiband spectral data by using a convolution network includes capturing spatial correlation between band data and band features obtained by combining a self-attention mechanism by using the convolution network, and performing dimension conversion by combining band original time series data as input of the inversion transform model.
5. The method of claim 4, wherein the performing self-attention computation on the extracted features by using an inverse transform model includes performing dimension adjustment on the extracted features by using a linear transformation layer, matching input dimensions required by an inverse transform encoder, and obtaining a high-order feature representation by feature integration as an input of the inverse transform encoder layer.
6. The method of claim 5, wherein the performing self-attention computation on the extracted features by using an inverse transform model further comprises using input features of an inverse transform encoder layer as a query, a key, and a value of a multi-headed self-attention mechanism, and each single-headed self-attention mechanism performs self-attention computation on the input query, key, and value matrix through respective linear transformation to obtain an output of each self-attention head.
7. The method for multiband spectral prediction based on the convolution inversion transform according to claim 6, wherein the obtaining of the final prediction result after verifying the output prediction result includes measuring the difference between the model prediction value and the true value by using the mean square error as a loss function, wherein the output of each self-attention head is subjected to layer normalization through residual connection, and the band is embedded into the full connection layer through the feedforward network and then output through the transform block, so as to obtain the prediction result.
8. A graph-convolution-inversion-transform-based multiband spectral prediction system, comprising:
A data acquisition module configured to acquire multiband spectral data;
a preprocessing module configured to perform data preprocessing on multiband spectral data;
the convolution module is configured to construct a graph convolution network, and the graph convolution network is utilized to perform feature extraction on the multiband spectrum data to obtain an adjacent matrix;
the prediction module is configured to perform self-attention calculation on the adjacent matrix combined with the multiband spectrum data by using an inverted transducer model and output a prediction result;
and the verification module is configured to obtain a final prediction result after verifying the output prediction result.
9. A computer readable storage medium having stored therein a plurality of instructions adapted to be loaded by a processor of a terminal device and to perform a graph-convolution-inversion-transform-based multi-band spectrum prediction method according to claim 1.
10. A terminal device comprising a processor and a computer readable storage medium, the processor configured to implement instructions; a computer readable storage medium for storing a plurality of instructions adapted to be loaded by a processor and to perform a method of graph-convolution-inversion-transform-based multiband spectral prediction as claimed in claim 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410658449.0A CN118233035B (en) | 2024-05-27 | 2024-05-27 | Multiband spectrum prediction method and system based on graph convolution inversion transform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410658449.0A CN118233035B (en) | 2024-05-27 | 2024-05-27 | Multiband spectrum prediction method and system based on graph convolution inversion transform |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118233035A true CN118233035A (en) | 2024-06-21 |
| CN118233035B CN118233035B (en) | 2024-08-06 |
Family
ID=91506382
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410658449.0A Active CN118233035B (en) | 2024-05-27 | 2024-05-27 | Multiband spectrum prediction method and system based on graph convolution inversion transform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118233035B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113487088A (en) * | 2021-07-06 | 2021-10-08 | 哈尔滨工业大学（深圳） | Traffic prediction method and device based on dynamic space-time diagram convolution attention model |
| CN114741507A (en) * | 2022-03-25 | 2022-07-12 | 西北大学 | Method for establishing and classifying quotation network classification model of graph convolution network based on Transformer |
| CN115049169A (en) * | 2022-08-16 | 2022-09-13 | 国网湖北省电力有限公司信息通信公司 | Regional power consumption prediction method, system and medium based on combination of frequency domain and spatial domain |
| CN117278154A (en) * | 2023-10-23 | 2023-12-22 | 哈尔滨工业大学 | Spectrum prediction method based on attention mechanism |
| CN117375751A (en) * | 2023-06-28 | 2024-01-09 | 重庆邮电大学 | Dynamic prediction method for wireless spectrum state |
| CN117709540A (en) * | 2023-12-19 | 2024-03-15 | 广西电网有限责任公司 | Short-term bus load prediction method and system for identifying abnormal weather |
| CN117834061A (en) * | 2023-12-13 | 2024-04-05 | 重庆邮电大学 | Wireless spectrum sensing data prediction method based on graph convolution network |
-
2024
- 2024-05-27 CN CN202410658449.0A patent/CN118233035B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113487088A (en) * | 2021-07-06 | 2021-10-08 | 哈尔滨工业大学（深圳） | Traffic prediction method and device based on dynamic space-time diagram convolution attention model |
| CN114741507A (en) * | 2022-03-25 | 2022-07-12 | 西北大学 | Method for establishing and classifying quotation network classification model of graph convolution network based on Transformer |
| CN115049169A (en) * | 2022-08-16 | 2022-09-13 | 国网湖北省电力有限公司信息通信公司 | Regional power consumption prediction method, system and medium based on combination of frequency domain and spatial domain |
| CN117375751A (en) * | 2023-06-28 | 2024-01-09 | 重庆邮电大学 | Dynamic prediction method for wireless spectrum state |
| CN117278154A (en) * | 2023-10-23 | 2023-12-22 | 哈尔滨工业大学 | Spectrum prediction method based on attention mechanism |
| CN117834061A (en) * | 2023-12-13 | 2024-04-05 | 重庆邮电大学 | Wireless spectrum sensing data prediction method based on graph convolution network |
| CN117709540A (en) * | 2023-12-19 | 2024-03-15 | 广西电网有限责任公司 | Short-term bus load prediction method and system for identifying abnormal weather |
Non-Patent Citations (4)
| Title |
|---|
| CHAOFEI GAO；TIAN WANG等: "3D Human Motion Prediction Based on Graph Convolution Network and Transformer", 《IEEE》, 14 March 2022 (2022-03-14) * |
| WEIQING YAN , YIQIU SUN , WUJIE ZHOU: "Deep Video Stabilization via Robust Homography Estimation", 《IEEE》, 1 November 2023 (2023-11-01) * |
| 陈海燕；欧阳林威；任保民: "一种基于频域注意力机制变量筛选的多变量时间 序列预测模型", 《华中科技大学学报（自然科学版）》, 9 March 2024 (2024-03-09) * |
| 陈瑞；徐金东；刘兆伟；阎维清等: "基于模糊空谱特征的高光谱图像分类", 《南京大学学报（自然科学）》, 28 February 2023 (2023-02-28) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118233035B (en) | 2024-08-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110502806B (en) | Wireless spectrum occupancy rate prediction method based on LSTM network | |
| CN114519469B (en) | Construction method of multivariable long-sequence time sequence prediction model based on transducer framework | |
| CN111401553A (en) | Missing data filling method and system based on neural network | |
| Mosavat-Jahromi et al. | Prediction and modeling of spectrum occupancy for dynamic spectrum access systems | |
| CN114500335B (en) | SDN network flow control method based on fuzzy C-means and mixed kernel least square support vector machine | |
| CN115856425B (en) | Spectrum anomaly detection method and device based on hidden space probability prediction | |
| CN112596016A (en) | Transformer fault diagnosis method based on integration of multiple one-dimensional convolutional neural networks | |
| CN116643949A (en) | Multi-model edge cloud load prediction method and device based on VaDE clustering | |
| CN117273188A (en) | Solar photovoltaic power generation power prediction method with incremental learning capability | |
| CN114494783A (en) | Pre-training method based on dynamic graph neural network | |
| CN117493930A (en) | Multi-element time sequence anomaly detection method based on contrast generation countermeasure network | |
| CN118233035B (en) | Multiband spectrum prediction method and system based on graph convolution inversion transform | |
| CN117496228A (en) | Knowledge distillation and graph model-based small sample increment radiation source individual identification method | |
| Peng et al. | Hmm-lstm for proactive traffic prediction in 6g wireless networks | |
| CN116861953A (en) | Searching method, searching device, predicting method and product of graph neural network architecture | |
| CN116614333A (en) | Modulation identification method based on Markov conversion field and deep learning | |
| Hassan et al. | On the kernel optimization of radial basis function using nelder mead simplex | |
| CN115174421A (en) | Network fault prediction method and device based on self-supervision unwrapping hypergraph attention | |
| CN111008584B (en) | Power quality measurement missing repair method for fuzzy self-organizing neural network | |
| CN115392102A (en) | Method and device for establishing energy consumption prediction model and method and system for predicting energy consumption | |
| Bidwai et al. | LSTM model for Channel Occupation Prediction in GSM Band | |
| CN117336187B (en) | Unmanned aerial vehicle communication network inference method based on inter-edge association | |
| CN115276856B (en) | Channel selection method based on deep learning | |
| JP7529334B1 (en) | Energy time series data forecasting system | |
| CN117692026B (en) | Link sensing method and device for power line communication |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |