Network data prediction analysis processing system based on electronic product

Abstract

The invention relates to the technical field of data prediction, in particular to a network data prediction analysis processing system based on an electronic product, which comprises a data acquisition unit, a data clustering unit, a modal decomposition unit and a prediction analysis unit, and is characterized in that the data acquisition unit is used for acquiring communication network data of a target electronic product; and the data clustering unit is used for clustering the communication network data of the target electronic product based on a clustering algorithm to obtain a clustering set corresponding to the communication network data, and dividing the communication network data into different data groups based on the clustering set. According to the method, the communication network data of the target electronic product is clustered, the system can perform grouping processing on the data with different characteristics, each data group can obtain finer local frequency components and time-frequency diagrams after being subjected to modal decomposition, the time-frequency characteristic expression of the data is enhanced, and the time-frequency characteristic expression of the data is improved. And the change rule of different communication data can be better reflected.

Description

Technical Field

[0001] The present invention relates to the technical field of data prediction, and in particular to a network data prediction analysis and processing system based on electronic products. Background Art

[0002] The network data of electronic products refers to various data transmitted, exchanged or processed through the network. These data may come from the communication process of electronic devices (such as smart phones, computers, routers, IoT devices, etc.) and are often time series data. In the communication network, many performance indicators (such as traffic, latency, bandwidth, packet loss rate, signal strength, etc.) will change over time. Therefore, these data are time-dependent and are often regarded as time series data.

[0003] Traditional systems often rely on simple statistical features or time series models, such as ARIMA and linear regression, to model data. These methods usually only focus on overall trends or simple historical data dependencies, but ignore the detailed features in the data. They cannot effectively capture the time-frequency characteristics and local changes of the data; traditional systems cannot deeply explore the local frequency components in the data and cannot fully express the time-frequency characteristics of the data, so their prediction capabilities for complex and time-varying communication data are weak; traditional systems often do not have the ability to cluster and personalize different types of data. The data they process is usually unclassified raw data, and data with different characteristics may be processed uniformly, making it impossible to optimize and analyze according to the inherent laws of the data; traditional systems may rely on offline models or relatively simple time series prediction methods, which cannot capture the time-varying characteristics of communication network data in real time, resulting in an inability to quickly respond to data changes and adjust prediction strategies. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a network data prediction, analysis and processing system based on electronic products.

[0005] The technical solution adopted to solve the above technical problems is: a network data prediction analysis processing system based on electronic products, including a data acquisition unit, a data clustering unit, a mode decomposition unit and a prediction analysis unit, specifically: The data acquisition unit is used to acquire the communication network data of the target electronic product, and transmit the communication network data of the target electronic product to the data clustering unit; The data clustering unit is used to receive the communication network data of the target electronic product transmitted by the data acquisition unit, and cluster the communication network data of the target electronic product based on a clustering algorithm to obtain a cluster set corresponding to the communication network data, divide the communication network data into different data groups based on the cluster set, and transmit the different data groups to the modal decomposition unit; The modal decomposition unit is used to receive the different data groups transmitted by the data clustering unit, and perform modal decomposition on the communication network data in the different data groups based on the integrated empirical mode decomposition method to obtain the local frequency components of the communication network data in the different data groups, obtain the spectral components of the local frequency components of the communication network data in the different data groups at each moment based on the short-time Fourier transform, sort the spectral components at each moment according to the time series to obtain a time-frequency diagram, and transmit the time-frequency diagram to the prediction and analysis unit.

[0006] Preferably, the prediction and analysis unit is used to receive the time-frequency diagram transmitted by the modal decomposition unit, and perform prediction and analysis on the time-frequency diagram based on a trained prediction and analysis model to obtain the communication network data at the next preset moment.

[0007] Preferably, clustering the communication network data of the target electronic product based on a clustering algorithm to obtain a cluster set corresponding to the communication network data includes: Calculate the division coefficient of the communication network data of the target electronic product, wherein the calculation formula of the division coefficient is as follows: ; in, Indicates the division factor of the communication network data of the target electronic product, represents the number of samples of communication network data, represents the number of clusters, Represents a data sample in the communication network data In a category The degree of affiliation in represents the fuzzy weighted index; Calculate the segmentation entropy of the communication network data of the target electronic product, wherein the calculation formula of the segmentation entropy is as follows: ; in, Represents the segmentation entropy of the communication network data of the target electronic product; Calculate the distance between the segmentation coefficient and segmentation entropy of the communication network data of the target electronic product and the cluster center and the sample mean, wherein the calculation formula of the distance is as follows: ; in, Represents the distance between the partition coefficient and partition entropy and the cluster center and sample mean. represents the cluster center, represents the sample mean; Based on the segmentation coefficient and segmentation entropy of the communication network data of the target electronic product and the distance between the cluster center and the sample mean as constraints, the target electronic product's communication network data has the minimum segmentation coefficient and the maximum segmentation entropy as the goal, and the optimal number of clusters is obtained.

[0008] Preferably, clustering the communication network data of the target electronic product based on a clustering algorithm to obtain a cluster set corresponding to the communication network data further includes: Construct a clustering objective function of the communication network data of the target electronic product, wherein the clustering objective function is as follows: ; in, represents the clustering objective function, Represents data samples To cluster center distance; Iterate from the initial point in the communication network data of the target electronic product until the clustering objective function converges to a local minimum point to obtain multiple local optimal solutions, wherein the solution formula for the local minimum point is as follows: ; in, Indicates local minimum points; A cluster set corresponding to the communication network data is obtained based on multiple local optimal solutions.

[0009] Preferably, performing modal decomposition on the communication network data in the different data groups based on an integrated empirical mode decomposition method to obtain local frequency components of the communication network data in the different data groups includes: Sliding in the communication network data in the different data groups based on a preset first sliding window to obtain sub-communication network data of the communication network data in the first sliding window; Performing EEMD decomposition on the sub-communication network data of the communication network data to obtain a plurality of IMF components corresponding to the sub-communication network data; Based on a preset second sliding window, slide on multiple IMF components corresponding to the sub-communication network data to obtain sub-IMF components of the IMF components within the second sliding window, and the sub-IMF components are local frequency components of the communication network data within the different data groups.

[0010] Preferably, performing EEMD decomposition on the sub-communication network data of the communication network data to obtain a plurality of IMF components corresponding to the sub-communication network data includes: Add a white noise sequence to the sub-communication network data to obtain new sub-communication network data, wherein the expression of the new sub-communication network data is as follows: ; in, Indicates new sub-communication network data, Represents sub-communication network data, represents a white noise sequence; The new sub-communication network data is decomposed based on the EMD algorithm to obtain the sum of multiple intrinsic mode functions and residuals, wherein the sum of multiple intrinsic mode functions and residuals is expressed as follows: ; in, represents the number of iterations of the EMD algorithm, Indicates the maximum number of iterations of the EMD algorithm, represents the number of integrations, represents the intrinsic mode function, represents the residual; The multiple intrinsic mode functions are averaged to obtain a final intrinsic mode function, wherein the final intrinsic mode function is the IMF component, wherein the expression of the final intrinsic mode function is as follows: ; in, represents the final intrinsic mode function, Represents the total number of times the white noise sequence is added.

[0011] Preferably, the prediction analysis model includes a feature extraction module, a feature connection module and a prediction module, the feature extraction module includes an initial convolution layer, a multi-size feature extraction block and an Add layer, and the initial convolution layer, the multi-size feature extraction block and the Add layer are connected in sequence.

[0012] Preferably, the multi-size feature extraction block includes three convolutional networks of different sizes, the Add layer is used to perform additive fusion of features extracted at different scales, the feature connection module is used to connect features of different sizes, and the feature connection module includes three identical convolutional networks stacked in parallel, and the extracted features are fused through a concat operation.

[0013] Preferably, the prediction module includes a position encoding layer, a multi-head self-attention mechanism, a feedforward neural network and a regression layer. The position encoding layer is used to use position encoding to add the position information of the input data. After converting the input data into an embedded vector through a linear layer, the position information is added using position encoding. The multi-head self-attention mechanism is used to input the input sequence with position information as Q, K, and V into the multi-head self-attention mechanism, and process the long-term dependency of time series data through parallel computing. The feedforward neural network layer is composed of two linear transformation layers, which are connected by a ReLU activation function in the middle, and are used to map the output of the multi-head self-attention mechanism and the residual connection to a high-dimensional space. The regression layer is used to map the features learned by the feedforward neural network layer into predicted values ​​of communication network data to obtain the communication network data at the next preset moment.

[0014] Preferably, the fusion calculation formula of the Add layer is as follows: ; in, represents the output of the Add layer, , and Represents the first of three convolutional networks of different sizes in the multi-scale feature extraction block. The feature map of each channel.

[0015] The beneficial effects of the present invention are as follows: (1) By clustering the communication network data of the target electronic product, the system can group data with different characteristics for processing. After modal decomposition, each data group can obtain more refined local frequency components and time-frequency diagrams, thereby enhancing the time-frequency feature expression of the data. This refined processing method can better reflect the changing laws of different communication data and improve the accuracy of the prediction model for future network data. (2) By integrating the empirical mode decomposition (EEMD) method and the short-time Fourier transform (STFT), the system can extract the spectral components at each moment and convert them into time-frequency diagrams, which makes the model more The invention can better capture the time-varying characteristics in the data, thereby improving the prediction ability of future data, and group the communication network data through the data clustering unit through the clustering algorithm, so that different data types can be processed separately according to their inherent laws during processing. This processing method adapts to the complexity of different communication network data, so that the system can optimize the analysis of specific types of data; (3) The present invention uses the time-frequency diagram obtained by the prediction and analysis unit through the received modal decomposition, and uses the trained model to perform prediction and analysis. This method can make efficient predictions based on historical data and timely predict future communication network data, thereby helping the system to achieve real-time monitoring and optimize decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of the system architecture of an overall system in an embodiment of the present invention.

[0017] Figure numerals: 1. Data acquisition unit; 2. Data clustering unit; 3. Modal decomposition unit; 4. Prediction and analysis unit. DETAILED DESCRIPTION

[0018] Embodiment 1, as Figure 1 As shown, the present invention proposes a network data prediction and analysis processing system based on electronic products, including a data acquisition unit 1, a data clustering unit 2, a modal decomposition unit 3 and a prediction and analysis unit 4, specifically: The data acquisition unit 1 is used to obtain the communication network data of the target electronic product and transmit the communication network data of the target electronic product to the data clustering unit 2; The data clustering unit 2 is used to receive the communication network data of the target electronic product transmitted by the data acquisition unit 1, and cluster the communication network data of the target electronic product based on the clustering algorithm to obtain a cluster set corresponding to the communication network data, divide the communication network data into different data groups based on the cluster set, and transmit the different data groups to the modal decomposition unit 3; The modal decomposition unit 3 is used to receive different data groups transmitted by the data clustering unit 2, and perform modal decomposition on the communication network data in different data groups based on the integrated empirical mode decomposition method to obtain the local frequency components of the communication network data in different data groups, obtain the spectral components of the local frequency components of the communication network data in different data groups at each moment based on the short-time Fourier transform, sort the spectral components at each moment according to the time series to obtain a time-frequency diagram, and transmit the time-frequency diagram to the prediction and analysis unit 4.

[0019] In the present invention, a clustering algorithm is an unsupervised learning method, which divides data into multiple categories by analyzing the similarity of data. Common clustering algorithms include K-Means, DBSCAN, hierarchical clustering, etc. The data are grouped based on certain features (such as the change pattern of network data) by using a clustering algorithm, with the purpose of grouping data with similar features. Communication network data is usually time series data. In a communication network, many performance indicators (such as traffic, delay, bandwidth, packet loss rate, signal strength, etc.) will change over time. Therefore, these data are time-dependent and are often regarded as time series data. A cluster set refers to a set of data groups divided according to a clustering algorithm. For example, the communication data collected by a data acquisition unit may have different characteristics. After clustering, there may be a data group with high traffic and a data group with low delay. Empirical Mode Decomposition (EMD) EEMD is an improved version of this method, which achieves a more robust decomposition process by introducing noise. The purpose of EEMD is to decompose complex signals into multiple intrinsic mode functions (IMFs), each of which represents the change in a certain frequency band of the signal. Short-time Fourier transform (STFT) is a method for time-frequency analysis, which obtains the spectral components of the signal at each moment by segmenting the signal and applying Fourier transform to each segment. STFT is particularly suitable for analyzing time-varying signals because it can capture both the time and frequency information of the signal. The time-frequency diagram is a graph obtained by methods such as STFT, which shows the changes of the signal in two dimensions: time and frequency. Through the time-frequency diagram, we can see more clearly how the frequency of the signal changes over time, which is particularly suitable for analyzing dynamic data such as communication signals.

[0020] In an optional embodiment, the prediction and analysis unit 4 is used to receive the time-frequency diagram transmitted by the modal decomposition unit 3, and perform prediction and analysis on the time-frequency diagram based on a trained prediction and analysis model to obtain the communication network data at the next preset moment.

[0021] Embodiment 2, a network data prediction analysis and processing system based on electronic products proposed by the present invention, compared with embodiment 1, this embodiment also includes: clustering the communication network data of the target electronic product based on a clustering algorithm to obtain a cluster set corresponding to the communication network data, including: Calculate the division coefficient of the communication network data of the target electronic product, where the calculation formula of the division coefficient is as follows: ; in, Indicates the division factor of the communication network data of the target electronic product, represents the number of samples of communication network data, represents the number of clusters, Represents a data sample in the communication network data In a category The degree of affiliation in represents the fuzzy weighted index; Calculate the segmentation entropy of the communication network data of the target electronic product, where the calculation formula of the segmentation entropy is as follows: ; in, Represents the segmentation entropy of the communication network data of the target electronic product; Calculate the distance between the segmentation coefficient and segmentation entropy of the communication network data of the target electronic product and the cluster center and the sample mean, where the distance calculation formula is as follows: ; in, Represents the distance between the partition coefficient and partition entropy and the cluster center and sample mean. represents the cluster center, represents the sample mean; Based on the segmentation coefficient and segmentation entropy of the communication network data of the target electronic product and the distance between the cluster center and the sample mean as constraints, the minimum segmentation coefficient and the maximum segmentation entropy of the communication network data of the target electronic product are set as the goals to obtain the optimal number of clusters.

[0022] In this embodiment, the partition coefficient is used to measure the degree of membership of a sample in a cluster, especially in a fuzzy clustering algorithm. Fuzzy clustering is different from traditional hard clustering. Samples can belong to multiple categories. The partition coefficient can be used to measure the degree of membership of a data sample in a certain category; the partition entropy is used to measure the uncertainty or information content of the clustering result, which is similar to the concept of information entropy. In fuzzy clustering, the smaller the entropy, the clearer the cluster, and the larger the entropy, the more vague the cluster; the cluster center is a core concept in the clustering algorithm, which refers to the "center" or "mean" of all samples in a cluster; the sample mean refers to the arithmetic mean of a group of data samples, which is a common indicator for measuring the central tendency of a sample data set; when optimizing the clustering results, it is hoped that by selecting an appropriate number of clusters (i.e., the number of clusters), the partition coefficient of the cluster is minimized (the tighter the cluster) and the partition entropy is maximized (the more uncertain the cluster, the higher the amount of information). This goal is used to balance the stability and information richness of the cluster in some applications.

[0023] In an optional embodiment, clustering the communication network data of the target electronic product based on a clustering algorithm to obtain a cluster set corresponding to the communication network data further includes: Construct a clustering objective function for the communication network data of the target electronic product, where the clustering objective function is as follows: ; in, represents the clustering objective function, Represents data samples To cluster center distance; Iterate from the initial point in the communication network data of the target electronic product until the clustering objective function converges to the local minimum point to obtain multiple local optimal solutions, where the solution formula for the local minimum point is as follows: ; in, Indicates local minimum points; A cluster set corresponding to the communication network data is obtained based on multiple local optimal solutions.

[0024] It should be noted that a local minimum point refers to a point in a certain neighborhood where the objective function value is less than or equal to the function values ​​of other points around the point, but it is not necessarily the global minimum. The objective function in the clustering algorithm may contain multiple local minimum points, so they do not necessarily represent the optimal solution; a local optimal solution means that under certain initial conditions, the clustering objective function reaches the minimum value, but this minimum value may not be the global minimum value. Since clustering algorithms are usually affected by initial conditions (such as the selection of initial cluster centers), the clustering process may converge to different local optimal solutions; in the clustering algorithm, a clustering set (i.e., multiple clustering division results) can be obtained by running the algorithm multiple times and extracting the local optimal solution from each run. These clustering results can be used to evaluate different clustering schemes for the data set, thereby determining the most appropriate number of clusters and cluster divisions.

[0025] In an optional embodiment, performing modal decomposition on the communication network data in different data groups based on an integrated empirical mode decomposition method to obtain local frequency components of the communication network data in different data groups includes: Sliding in the communication network data in different data groups based on a preset first sliding window to obtain sub-communication network data of the communication network data in the first sliding window; Performing EEMD decomposition on the sub-communication network data of the communication network data to obtain a plurality of IMF components corresponding to the sub-communication network data; Based on the preset second sliding window, multiple IMF components corresponding to the sub-communication network data are slid to obtain sub-IMF components of the IMF components in the second sliding window, and the sub-IMF components are local frequency components of the communication network data in different data groups.

[0026] It should be noted that the sliding window is a commonly used technique in signal processing and time series analysis, and is used to perform local analysis on a data set. The core idea of ​​the sliding window is to define a window of a fixed size and slide it along the data sequence to gradually obtain subsets of the data for local processing; the first sliding window refers to sliding on the original communication network data, with the goal of extracting different sub-data sets (i.e., sub-communication network data) from the entire data sequence; the second sliding window refers to sliding on the IMF components obtained after EEMD decomposition, with the goal of extracting local frequency components from the IMF components, i.e., the local characteristics of each IMF component; the sub-IMF component refers to the subset extracted from the IMF components obtained by EEMD decomposition through the second sliding window. Since each IMF component decomposed by EEMD may contain multiple frequency components, each IMF component can be further decomposed into smaller parts (i.e., sub-IMF components) through sliding window analysis. These sub-IMF components correspond to the changes in the IMF components in the local time period, reflecting the local frequency characteristics of the signal.

[0027] In an optional embodiment, EEMD decomposition is performed on the sub-communication network data of the communication network data to obtain a plurality of IMF components corresponding to the sub-communication network data, including: A white noise sequence is added to the sub-communication network data to obtain new sub-communication network data, wherein the expression of the new sub-communication network data is as follows: ; in, Indicates new sub-communication network data, Represents sub-communication network data, represents a white noise sequence; The new sub-communication network data is decomposed based on the EMD algorithm to obtain the sum of multiple intrinsic mode functions and residuals, wherein the expression of the sum of multiple intrinsic mode functions and residuals is as follows: ; in, represents the number of iterations of the EMD algorithm, Indicates the maximum number of iterations of the EMD algorithm, represents the number of integrations, represents the intrinsic mode function, represents the residual; The multiple intrinsic mode functions are averaged to obtain the final intrinsic mode function, wherein the final intrinsic mode function is the IMF component, wherein the expression of the final intrinsic mode function is as follows: ; in, represents the final intrinsic mode function, Represents the total number of times the white noise sequence is added.

[0028] It should be noted that white noise is a random signal containing all frequency components. Its characteristic is that the power density of each frequency is equal. The white noise sequence is added to the original sub-communication network data to increase the complexity of the signal or to suppress the influence of mode aliasing. The EMD algorithm is usually based on successive local extreme points, and performs repeated extraction and decomposition until certain stopping criteria are met. The number of iterations represents the number of repeated processing of the data in an EMD decomposition process. The final output of the EMD decomposition process is the residual, which represents the part of the signal that cannot be extracted by IMF decomposition. The final intrinsic mode function is a frequency component in the signal, representing a local feature of the signal. The total number of times white noise sequences are added refers to the number of white noise sequences added in each iteration during the EEMD process. Adding different white noise sequences each time can increase the diversity and stability of the EEMD decomposition, avoid the decomposition results being affected by a single noise, and finally obtain a more accurate IMF by integrating multiple decomposition results.

[0029] In an optional embodiment, the predictive analysis model includes a feature extraction module, a feature connection module and a prediction module, the feature extraction module includes an initial convolution layer, a multi-size feature extraction block and an Add layer, and the initial convolution layer, the multi-size feature extraction block and the Add layer are connected in sequence.

[0030] It should be noted that the Add layer is a layer that adds multiple feature maps. In deep learning models, especially when using network structures such as deep residual networks (ResNet), the Add layer is usually used for skip connections, that is, adding the input signal directly to the output of certain layers in the network. This structure can effectively alleviate the gradient vanishing problem in deep neural networks and improve the training efficiency and performance of the network.

[0031] In an optional embodiment, the multi-size feature extraction block includes three convolutional networks of different sizes, the Add layer is used to perform additive fusion of features extracted at different scales, the feature connection module is used to connect features of different sizes, the feature connection module includes three identical convolutional networks stacked in parallel, and the extracted features are fused through a concat operation.

[0032] It should be noted that "multi-size" refers to extracting features at multiple scales (or sizes). Multi-size feature extraction blocks usually use multiple convolution kernels of different sizes to capture local patterns of different sizes; "parallel stacking" means stacking multiple identical convolutional network layers in parallel at the same stage. These networks are the same, that is, they share structures (convolution kernel size, number, stride, etc.) and weights, but they operate independently. This design allows the network to perform parallel calculations on different features, thereby accelerating information extraction and fusion; concat is an abbreviation for "concatenation" or "connection", which means concatenating multiple tensors together in a specific dimension. In deep learning, it is common to concatenate feature maps from different sources in the feature dimension (usually the channel dimension).

[0033] In an optional embodiment, the prediction module includes a position encoding layer, a multi-head self-attention mechanism, a feedforward neural network and a regression layer. The position encoding layer is used to use position encoding to add the position information of the input data. After the input data is converted into an embedded vector through a linear layer, the position information is added using the position encoding. The multi-head self-attention mechanism is used to input the input sequence with position information as Q, K, and V into the multi-head self-attention mechanism, and the long-term dependencies of time series data are processed through parallel computing. The feedforward neural network layer is composed of two linear transformation layers, which are connected by a ReLU activation function in the middle, and are used to map the output of the multi-head self-attention mechanism and the residual connection to a high-dimensional space. The regression layer is used to map the features learned by the feedforward neural network layer into predicted values ​​of the communication network data to obtain the communication network data at the next preset moment.

[0034] It should be noted that the position encoding layer is used to add position information to the input sequence. Deep learning models, especially models like Transformer, do not have the ability to process the order of input data when processing sequence data, because standard neural network architectures (such as convolutional neural networks) usually process two-dimensional structured data such as images, and sequence data (such as text, time series, etc.) must be able to retain their timing or order information; Self-Attention is a technology that calculates the correlation between different positions in sequence data. The multi-head self-attention mechanism is an extension of self-attention. It enhances the model's learning ability for different information by calculating multiple "attention heads" in parallel; Q, K, V In the self-attention mechanism, the input sequence is linearly transformed to generate three vectors: query vector (Query, Q), key vector (Key, K) and value vector (Value, V), which are used to calculate the correlation between different positions in the sequence. Each attention head calculates the attention score based on these three vectors; Feedforward neural network is a network structure composed of two or more layers of linear transformation (usually fully connected layers); The regression layer is the final layer of the model, which is usually used to map the features learned in the deep learning network to actual numerical outputs.

[0035] In an optional embodiment, the fusion calculation formula of the Add layer is as follows: ; in, represents the output of the Add layer, , and Represents the first of three convolutional networks of different sizes in the multi-scale feature extraction block. The feature map of each channel.

[0036] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A network data prediction, analysis and processing system based on electronic products, characterized in that: The system comprises a data collection unit (1), a data clustering unit (2), a modal decomposition unit (3) and a prediction analysis unit (4), specifically: The data acquisition unit (1) is used to acquire communication network data of a target electronic product, and transmit the communication network data of the target electronic product to a data clustering unit (2); The data clustering unit (2) is used to receive the communication network data of the target electronic product transmitted by the data acquisition unit (1), cluster the communication network data of the target electronic product based on a clustering algorithm to obtain a cluster set corresponding to the communication network data, divide the communication network data into different data groups based on the cluster set, and transmit the different data groups to the modal decomposition unit (3); The modal decomposition unit (3) is used to receive the different data groups transmitted by the data clustering unit (2), and perform modal decomposition on the communication network data in the different data groups based on an integrated empirical mode decomposition method to obtain local frequency components of the communication network data in the different data groups, obtain the spectral components of the local frequency components of the communication network data in the different data groups at each moment based on short-time Fourier transform, sort the spectral components at each moment according to a time series to obtain a time-frequency diagram, and transmit the time-frequency diagram to the prediction and analysis unit (4).

2. The network data prediction, analysis and processing system based on electronic products according to claim 1, characterized in that: The prediction and analysis unit (4) is used to receive the time-frequency diagram transmitted by the mode decomposition unit (3), and to perform prediction and analysis on the time-frequency diagram based on a trained prediction and analysis model, so as to obtain communication network data at the next preset time.

3. The network data prediction, analysis and processing system based on electronic products according to claim 2, characterized in that: Clustering the communication network data of the target electronic product based on a clustering algorithm to obtain a cluster set corresponding to the communication network data includes: Calculate the division coefficient of the communication network data of the target electronic product, wherein the calculation formula of the division coefficient is as follows: ； in, Indicates the division factor of the communication network data of the target electronic product, represents the number of samples of communication network data, represents the number of clusters, Represents a data sample in the communication network data In a category The degree of affiliation in represents the fuzzy weighted index; Calculate the segmentation entropy of the communication network data of the target electronic product, wherein the calculation formula of the segmentation entropy is as follows: ； in, Represents the segmentation entropy of the communication network data of the target electronic product; Calculate the distance between the segmentation coefficient and segmentation entropy of the communication network data of the target electronic product and the cluster center and the sample mean, wherein the calculation formula of the distance is as follows: ； in, Represents the distance between the partition coefficient and partition entropy and the cluster center and sample mean. represents the cluster center, represents the sample mean; Based on the segmentation coefficient and segmentation entropy of the communication network data of the target electronic product and the distance between the cluster center and the sample mean as constraints, the target electronic product's communication network data has the minimum segmentation coefficient and the maximum segmentation entropy as the goal, and the optimal number of clusters is obtained.

4. The network data prediction, analysis and processing system based on electronic products according to claim 3 is characterized in that: Clustering the communication network data of the target electronic product based on a clustering algorithm to obtain a cluster set corresponding to the communication network data, further comprising: Construct a clustering objective function of the communication network data of the target electronic product, wherein the clustering objective function is as follows: ； in, represents the clustering objective function, Represents data samples To cluster center distance; Iterate from the initial point in the communication network data of the target electronic product until the clustering objective function converges to a local minimum point to obtain multiple local optimal solutions, wherein the solution formula for the local minimum point is as follows: ； in, Indicates local minimum points; A cluster set corresponding to the communication network data is obtained based on multiple local optimal solutions.

5. The network data prediction, analysis and processing system based on electronic products according to claim 4, characterized in that: Performing modal decomposition on the communication network data in the different data groups based on an integrated empirical mode decomposition method to obtain local frequency components of the communication network data in the different data groups includes: Sliding in the communication network data in the different data groups based on a preset first sliding window to obtain sub-communication network data of the communication network data in the first sliding window; Performing EEMD decomposition on the sub-communication network data of the communication network data to obtain a plurality of IMF components corresponding to the sub-communication network data; Based on a preset second sliding window, slide on multiple IMF components corresponding to the sub-communication network data to obtain sub-IMF components of the IMF components within the second sliding window, and the sub-IMF components are local frequency components of the communication network data within the different data groups.

6. The network data prediction, analysis and processing system based on electronic products according to claim 5, characterized in that: Performing EEMD decomposition on the sub-communication network data of the communication network data to obtain a plurality of IMF components corresponding to the sub-communication network data, including: Add a white noise sequence to the sub-communication network data to obtain new sub-communication network data, wherein the expression of the new sub-communication network data is as follows: ； in, Indicates new sub-communication network data, Represents sub-communication network data, represents a white noise sequence; The new sub-communication network data is decomposed based on the EMD algorithm to obtain the sum of multiple intrinsic mode functions and residuals, wherein the sum of multiple intrinsic mode functions and residuals is expressed as follows: ； in, represents the number of iterations of the EMD algorithm, Indicates the maximum number of iterations of the EMD algorithm, represents the number of integrations, represents the intrinsic mode function, represents the residual; The multiple intrinsic mode functions are averaged to obtain a final intrinsic mode function, wherein the final intrinsic mode function is the IMF component, wherein the expression of the final intrinsic mode function is as follows: ； in, represents the final intrinsic mode function, Represents the total number of times the white noise sequence is added.

7. The network data prediction, analysis and processing system based on electronic products according to claim 6, characterized in that: The prediction analysis model includes a feature extraction module, a feature connection module and a prediction module. The feature extraction module includes an initial convolution layer, a multi-size feature extraction block and an Add layer. The initial convolution layer, the multi-size feature extraction block and the Add layer are connected in sequence.

8. The network data prediction, analysis and processing system based on electronic products according to claim 7, characterized in that: The multi-size feature extraction block includes three convolutional networks of different sizes, the Add layer is used to perform additive fusion on features extracted at different scales, the feature connection module is used to connect features of different sizes, and the feature connection module includes three identical convolutional networks stacked in parallel, and the extracted features are fused through a concat operation.

9. The network data prediction, analysis and processing system based on electronic products according to claim 8, characterized in that: The prediction module includes a position encoding layer, a multi-head self-attention mechanism, a feedforward neural network and a regression layer. The position encoding layer is used to use position encoding to add the position information of the input data. After the input data is converted into an embedded vector through a linear layer, the position information is added using the position encoding. The multi-head self-attention mechanism is used to send the input sequence with position information as Q, K, and V into the multi-head self-attention mechanism, and process the long-term dependency of time series data through parallel calculation. The feedforward neural network layer is composed of two linear transformation layers, which are connected in the middle with a ReLU activation function, and are used to map the output of the multi-head self-attention mechanism and the residual connection to a high-dimensional space. The regression layer is used to map the features learned by the feedforward neural network layer into the predicted value of the communication network data to obtain the communication network data at the next preset moment.

10. The network data prediction, analysis and processing system based on electronic products according to claim 9, characterized in that: The fusion calculation formula of the Add layer is as follows: ； in, represents the output of the Add layer, , and Represents the first of three convolutional networks of different sizes in the multi-scale feature extraction block. The feature map of each channel.

Images