Edge computing preprocessing communication method and system based on server performance indexes

Abstract

The invention provides an edge computing preprocessing communication method and system based on server performance indicators, and relates to the technical field of server performance indicators, and the method comprises the steps: collecting performance indicators and communication signals of a distributed edge server during operation, carrying out the time sequence analysis of the performance indicators to form a fluctuation analysis result, and carrying out the analysis of the fluctuation analysis result; the method comprises the following steps: extracting a spatial-temporal feature sequence by using a convolutional neural network, generating load trend prediction data in combination with a real-time service request type, preprocessing a communication signal, and finally processing the load trend prediction data by using an adaptive genetic algorithm to generate a task scheduling scheme matched with a current resource state. And the pre-processed communication signal is transmitted to the corresponding edge server. The task scheduling efficiency and the communication reliability in the edge computing environment are improved.

Description

Technical Field

[0001] This application relates to the field of server performance index technology, and in particular to an edge computing preprocessing communication method and system based on server performance index. Background Technology

[0002] In heterogeneous edge computing environments, multi-service parallel processing scenarios place higher demands on real-time communication and collaborative computing resources. Edge servers are characterized by hardware heterogeneity, dynamic resource fluctuations, and diverse service types. Therefore, it is necessary to build a preprocessing mechanism that can adapt to environmental changes, achieve efficient task scheduling, and ensure communication reliability. This mechanism needs to be able to perceive the status of each computing node in real time, predict its load change trends, and ensure the accurate transmission of scheduling instructions in complex electromagnetic environments.

[0003] One current targeted solution combines a dynamic weighted polling mechanism with signal filtering technology. This method first periodically collects server operating indicators and builds a resource assessment model based on historical data. Then, it uses a weighted polling algorithm to allocate computing tasks. At the communication level, a band-stop filter with fixed parameters is used to suppress noise in specific frequency bands, while redundant coding is used to improve transmission reliability. The task scheduling module dynamically adjusts the weight coefficients according to real-time load indicators, thereby achieving basic load balancing.

[0004] However, this scheme has several limitations, namely: the system exhibits lag in response when resources fluctuate drastically; the fixed filter parameters result in insufficient signal adaptability; the matching accuracy between the weight adjustment mechanism and real-time service requirements is also relatively limited; in addition, the depth of correlation analysis of multi-dimensional indicators is insufficient in dynamic environments, which affects the accuracy of load prediction; and when communication quality is affected by environmental interference, the overhead introduced by redundant coding will further reduce the effective transmission efficiency. Summary of the Invention

[0005] This application provides an edge computing preprocessing communication method and system based on server performance indicators to solve the problems of low task scheduling efficiency and poor communication reliability in the edge computing environment in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides an edge computing preprocessing communication method based on server performance metrics, comprising: Collect server performance metrics and communication signals during edge server operation; Perform time-series fluctuation analysis on the server performance indicators to generate fluctuation analysis results; The fluctuation analysis results are used to extract features using a convolutional neural network to extract spatiotemporal feature sequences. Based on the spatiotemporal feature sequences, combined with the business request type and priority, load trend prediction data is generated. The communication signal is preprocessed to obtain the preprocessed communication signal; An adaptive genetic algorithm is used to process the load trend prediction data, generate a task scheduling scheme, and transmit the task scheduling scheme to the corresponding edge server through the preprocessed communication signal.

[0007] Optionally, the step of using a convolutional neural network to extract features from the fluctuation analysis results, extracting a spatiotemporal feature sequence, and generating load trend prediction data based on the spatiotemporal feature sequence and in conjunction with the business request type and priority, includes: The fluctuation analysis results are input into a convolutional neural network, and a spatiotemporal feature sequence is generated through the alternating processing of multiple convolutional layers and pooling layers in the convolutional neural network. The temporal analysis module of the convolutional neural network processes the spatiotemporal feature sequence to generate regular features of computational load changes over time. Based on the type and priority of business requests, the aforementioned patterns are weighted and adjusted to generate load trend prediction data.

[0008] Optionally, the step of inputting the fluctuation analysis results into a convolutional neural network, and generating a spatiotemporal feature sequence through alternating processing of multiple convolutional layers and pooling layers in the convolutional neural network, includes: The fluctuation analysis results are input into a convolutional neural network, and the fluctuation analysis results are reorganized into a two-dimensional feature matrix according to the time dimension; The first feature is generated by extracting features from the two-dimensional feature matrix through the first convolutional layer of the convolutional neural network. The first feature is downsampled through the first pooling layer of the convolutional neural network to obtain the second feature. The second feature is extracted from the second feature through the second convolutional layer of the convolutional neural network to generate the third feature; The third feature is downsampled twice through the second pooling layer of the convolutional neural network to obtain the fourth feature; The fourth feature is flattened to generate a one-dimensional feature vector, and the one-dimensional feature vector is then compressed to generate a spatiotemporal feature sequence.

[0009] Optionally, the step of extracting features from the two-dimensional feature matrix to generate the first feature includes: A sliding window scan is performed along the row and column directions of the two-dimensional feature matrix using a set of convolutional kernels of a preset size; At each sliding window position, the product of the convolution kernel parameters and the data at the corresponding position within the sliding window is calculated to obtain multiple feature values; Based on each of the aforementioned feature values, a corresponding feature representation is generated using a nonlinear transformation function; All the aforementioned features are combined and arranged in a spatial dimension to form the first feature.

[0010] Optionally, the step of performing time-series fluctuation analysis on the server performance indicators to generate fluctuation analysis results includes: Arrange all server performance metrics in chronological order to form a performance metric time series; The performance index time series is divided into multiple time segments according to a fixed time window; Calculate the fluctuation amplitude and frequency for each time segment; The fluctuation amplitude and the fluctuation frequency are weighted and fused to form a quantitative indicator; Based on the changing trends of quantitative indicators across all time windows, volatility analysis results are constructed.

[0011] Optionally, the step of preprocessing the communication signal to obtain a preprocessed communication signal includes: Perform a Fourier transform on the communication signal to obtain a frequency domain signal; The frequency domain signal is segmented and scanned with a fixed bandwidth to form the power distribution of each frequency band; Based on the power distribution of each frequency band, frequency bands with signal power less than the preset background noise baseline or greater than the preset transmission threshold are marked as abnormal frequency bands. The abnormal frequency band is filtered, and the filtered frequency band is subjected to inverse Fourier transform to obtain the recovered communication signal. The recovered communication signal is segmented for signal-to-noise ratio detection, and based on the detection results, the substandard signal segments are subjected to secondary filtering to generate a preprocessed communication signal.

[0012] Optionally, the step of using an adaptive genetic algorithm to process the load trend prediction data and generate a task scheduling scheme includes: The load trend prediction data is encoded into chromosome gene sequences; Based on the server performance metrics, dynamically calculate the population initialization parameters, including population size and gene length; An initial population is generated based on the chromosome gene sequence and the population initialization parameters; The population is subjected to evolutionary operations. After the evolution, individuals with fitness values ​​greater than a preset fitness threshold are retained in the population through a roulette wheel selection mechanism to form a new generation of population. Repeat the evolution and selection operations until the last generation of the population reaches the preset number of generations or the preset fitness convergence threshold, then terminate the iteration process. The optimal chromosome is selected from the final population at the end of the iteration for decoding. The decoding result is mapped to a task scheduling instruction, and a task scheduling scheme is formed based on the task scheduling instruction.

[0013] Secondly, this application provides an edge computing preprocessing communication system based on server performance metrics, comprising: The acquisition module is used to collect server performance metrics and communication signals during the operation of the edge server; The analysis module is used to perform time-series fluctuation analysis on the server performance indicators and generate fluctuation analysis results; The extraction module is used to extract features from the fluctuation analysis results using a convolutional neural network, extract spatiotemporal feature sequences, and generate load trend prediction data based on the spatiotemporal feature sequences and in combination with business request types and priorities. The suppression module is used to preprocess the communication signal to obtain a preprocessed communication signal; The generation module is used to process the load trend prediction data using an adaptive genetic algorithm, generate a task scheduling scheme, and transmit the task scheduling scheme to the corresponding edge server through the preprocessed communication signal.

[0014] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor, used to execute the computer program to implement the steps of the edge computing preprocessing communication method based on server performance metrics as described in the first aspect above.

[0015] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the edge computing preprocessing communication method based on server performance indicators as described in the first aspect above.

[0016] This application provides an edge computing preprocessing communication method based on server performance indicators. The method includes: collecting server performance indicators and communication signals during edge server operation; performing time-series fluctuation analysis on the server performance indicators to generate fluctuation analysis results; using a convolutional neural network to extract features from the fluctuation analysis results, extracting spatiotemporal feature sequences; generating load trend prediction data based on the spatiotemporal feature sequences, combined with business request types and priorities; preprocessing the communication signals to obtain preprocessed communication signals; using an adaptive genetic algorithm to process the load trend prediction data, generating a task scheduling scheme, and transmitting the task scheduling scheme to the corresponding edge server via the preprocessed communication signals.

[0017] The technical solution provided in this application has the following beneficial effects: This application enables comprehensive perception of the operating status of multiple nodes in a distributed environment, providing a complete data foundation for subsequent analysis. Based on this, it further reveals the dynamic patterns of resource usage and extracts representative fluctuation characteristic patterns. By capturing the spatiotemporal correlation characteristics of resource usage to generate high-dimensional feature representations, it achieves accurate prediction of future load changes, providing a basis for resource scheduling decisions. Simultaneously, this scheme aims to improve signal transmission quality to reduce the impact of environmental interference on the communication process. Ultimately, it achieves optimal task allocation in dynamic environments, improves resource utilization efficiency, and ensures reliable delivery of scheduling commands, thus completing a full control loop.

[0018] Furthermore, this application extracts spatiotemporal feature sequences from the fluctuation analysis results through multi-layer convolution and pooling processing of convolutional neural networks, identifies load change patterns using a time-domain analysis module, and performs weighted corrections based on real-time business types and priorities to ultimately generate load trend prediction data.

[0019] Furthermore, this solution can deeply explore the spatiotemporal characteristics of resource usage, thereby accurately identifying the patterns of load changes; based on this, it generates accurate predictive data in conjunction with real-time business needs, thus providing a reliable basis for resource scheduling decisions.

[0020] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating an edge computing preprocessing communication method based on server performance metrics, provided in this application embodiment; Figure 2 This application provides a schematic diagram illustrating a specific implementation of an edge computing preprocessing communication method based on server performance metrics. Figure 3 This is a schematic diagram of the structure of an edge computing preprocessing communication system based on server performance indicators, provided in an embodiment of this application. Detailed Implementation

[0023] In heterogeneous edge computing environments, multi-service parallel processing scenarios place higher demands on communication real-time performance and resource coordination. To address this need, existing solutions employ a dynamic weighted polling mechanism combined with fixed-parameter filtering technology. While this approach can achieve basic load balancing and signal processing, it still has significant limitations in practical applications. Specifically, when resource fluctuations are drastic, the solution exhibits response lag; the fixed filter parameters result in insufficient signal adaptability; and the matching accuracy between its weight adjustment mechanism and real-time service requirements is relatively limited. Furthermore, the insufficient depth of multi-dimensional index correlation analysis in dynamic environments affects the accuracy of load prediction, and when communication quality is affected by environmental interference, the overhead of redundant coding further reduces effective transmission efficiency.

[0024] To address the aforementioned issues, this application proposes an edge computing preprocessing communication method based on server performance metrics. This method first collects performance metrics and communication signals from distributed servers in real time, and performs time-series analysis and fluctuation quantification on the performance data. Then, it utilizes a convolutional neural network to extract spatiotemporal features, thereby predicting load change trends. Simultaneously, adaptive filtering technology is employed at the communication layer to purify the communication signals. Finally, an intelligent optimization algorithm generates a task scheduling scheme that precisely matches the current resource state. This method, through collaborative analysis and dynamic optimization of multi-dimensional data, ultimately achieves accurate load prediction and reliable communication transmission in resource fluctuation scenarios, effectively solving problems such as response lag, insufficient signal adaptability, and limited resource scheduling accuracy in existing technologies, thereby improving resource utilization efficiency and overall communication quality in edge computing environments.

[0025] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] The core of this application is to provide an edge computing preprocessing communication method based on server performance indicators, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes: Step 101: Collect server performance metrics and communication signals during edge server operation.

[0027] In step 101, an edge server refers to a computing node distributed on the user side to provide localized computing services; server performance indicators include data reflecting the server's operating status, such as CPU utilization, GPU utilization, memory usage, and I / O latency; communication signals refer to control commands or task data signals that need to be sent to the edge server.

[0028] For example, taking three edge servers in locations A, B, and C as an example, the monitoring program collects performance metrics every 5 seconds. These performance metrics include CPU utilization, GPU utilization, memory usage, and I / O latency. At the same time, it collects task scheduling instruction signals that need to be sent to these servers. After continuous collection for 10 minutes, a total of 120 time points of performance metric data and corresponding communication signal data are obtained, forming the raw dataset.

[0029] Step 102: Perform time-series fluctuation analysis on the server performance indicators to generate fluctuation analysis results.

[0030] In step 102, fluctuation intensity quantification is a process that numerically represents the degree of data fluctuation. The fluctuation analysis result is a dataset that reflects the dynamic changes in computing resources.

[0031] In this embodiment, the collected performance indicators are first arranged in chronological order, and the time series is divided into multiple consecutive time segments using a sliding window method. Then, the fluctuation amplitude and fluctuation frequency of each time segment are calculated. The fluctuation amplitude is obtained by calculating the sum of squared differences between adjacent data points within the segment, while the fluctuation frequency is obtained by counting the number of fluctuations exceeding a set threshold per unit time. Then, the calculated fluctuation amplitude and fluctuation frequency are fused according to preset weights to form a unified quantitative indicator. Finally, a complete fluctuation analysis result is constructed based on the quantitative indicators of all time segments.

[0032] For example, performance metrics collected from 120 time points over 10 minutes from 3 servers are arranged chronologically and divided into 20 segments with 30-second time windows; the fluctuation amplitude is calculated for each segment using the formula: Fluctuation Amplitude = ∑( ,in Indicates the first The indicator value at each time point, Indicates the first Performance index values ​​at each time point; fluctuation frequency statistics of the number of times the fluctuation exceeds 10% per minute; the fluctuation amplitude and fluctuation frequency are weighted and merged with a weight of 0.6 and 0.4 respectively, and finally a fluctuation analysis result containing 80 values ​​is generated.

[0033] Step 103: Use a convolutional neural network to extract features from the fluctuation analysis results, extract spatiotemporal feature sequences, and generate load trend prediction data based on the spatiotemporal feature sequences and the business request type and priority.

[0034] In step 103, the spatiotemporal feature sequence is a data sequence that simultaneously contains temporal and spatial features; the service request type and priority are generated by the terminal devices accessing the edge computing environment and are dynamically determined based on the urgency of the service, service quality requirements, and resource demand intensity; the load trend prediction data is a prediction result of future changes in computing load.

[0035] In this embodiment, after the fluctuation analysis results are input into the convolutional neural network, features are first extracted by alternating processing through multiple convolutional layers and pooling layers. The convolutional layers use convolutional kernels to extract local features, while the pooling layers are responsible for dimensionality reduction of the feature map. Next, the feature map output by the network is flattened into a one-dimensional vector and dimensionality reduced to generate a spatiotemporal feature sequence. Then, the spatiotemporal feature sequence is processed by the temporal analysis module to identify the load change pattern, and the pattern is further weighted and corrected by combining the business request type and priority weight, finally generating load trend prediction data.

[0036] For example, the fluctuation analysis results containing 80 values ​​are first reorganized into a 20×4 two-dimensional matrix and input into a convolutional neural network. The first convolutional layer of the network uses 8 3×3 convolutional kernels to extract features, generating an 18×2×8 feature map. This feature map is then reduced in dimensionality by a pooling layer to obtain a 9×1×8 feature map. Subsequently, the second convolutional layer uses 16 3×3 convolutional kernels to further extract features. The final output features are flattened into a 144-dimensional vector and reduced to 36 dimensions to form a spatiotemporal feature sequence. By analyzing this sequence, a periodic fluctuation pattern in load every 15 minutes is identified. This pattern is then weighted and corrected by combining the weights of video analytics business priority (0.7) and IoT business priority (0.3), ultimately generating load trend prediction data.

[0037] Step 104: Preprocess the communication signal to obtain the preprocessed communication signal.

[0038] In step 104, the preprocessed communication signal is a signal that has undergone purification processing.

[0039] In this embodiment, the communication signal is first converted into a frequency domain representation and the power distribution of each frequency band is analyzed. Then, frequency bands with excessively low or high power are marked as frequency bands to be processed, and these specific frequency bands are filtered. Then, the signal in the filtered frequency band is inversely transformed to restore its time domain form, and then the signal-to-noise ratio is detected. For signal segments that do not meet the detection criteria, a second filtering process is performed, and finally the preprocessed communication signal is obtained.

[0040] For example, firstly, a Fourier transform is performed on the task scheduling command signal, and its power distribution in the 0-10MHz frequency band is analyzed. Then, a total of 12 frequency bands with power below -90dBm or above -30dBm are selected for filtering. Specifically, the frequency bands with excessively low power are compensated, while the frequency bands with excessively high power are attenuated. Next, the processed signal is inversely transformed to restore its time domain form, and its signal-to-noise ratio is detected in 10-millisecond units. Finally, for the three signal bands with a signal-to-noise ratio below 20dB found in the detection, a second filtering process is performed to obtain the preprocessed communication signal.

[0041] Step 105: Use an adaptive genetic algorithm to process the load trend prediction data, generate a task scheduling scheme, and transmit the task scheduling scheme to the corresponding edge server through the preprocessed communication signal.

[0042] In step 105, the current edge computing environment resource status is obtained by real-time collection and analysis of multi-dimensional performance indicators of each edge server, reflecting the distribution of available computing resources in the entire edge computing cluster; the task scheduling scheme is a specific plan that determines the task allocation method.

[0043] In this embodiment, the load prediction data is first encoded into chromosomes, and the population parameters of the genetic algorithm are determined according to the server performance indicators. After generating the initial population, evolutionary operations, including crossover and mutation, are performed on the population, wherein individuals with high fitness are retained through a roulette wheel selection strategy. This evolutionary process is repeated until a preset termination condition is met. Then, the optimal chromosome is decoded to generate the final task scheduling scheme, and finally, the scheme is transmitted to the edge server through a preprocessed communication signal.

[0044] For example, the load prediction data is first encoded into a chromosome containing 20 genes, and the population size of the genetic algorithm is set to 35 based on an average CPU utilization of 65%. After initializing the population, evolution is performed with a crossover probability of 0.8 and a mutation probability of 0.5, where individuals with fitness greater than 0.85 are retained using a roulette wheel selection strategy. After 50 generations of evolution, the optimal chromosome with fitness of 0.95 is selected, decoded to generate a specific task allocation scheme, and finally the scheme is sent to three edge servers through a preprocessed communication signal.

[0045] This method achieves accurate perception and intelligent scheduling of the edge computing environment through multi-step collaborative processing. Specifically, the method first accurately predicts the load change trend by collecting and analyzing server performance indicators in real time, and then combines an adaptive algorithm to generate the optimal task scheduling scheme to ensure the reliable transmission of communication signals. Ultimately, it effectively improves the resource utilization efficiency and overall task processing capability of the edge computing environment.

[0046] To address the issues of insufficient extraction of resource fluctuation features and inadequate load prediction accuracy in edge computing environments, some embodiments include step 103: utilizing a convolutional neural network to extract features from the fluctuation analysis results, extracting a spatiotemporal feature sequence, and generating load trend prediction data based on the spatiotemporal feature sequence, combined with business request type and priority, such as... Figure 2 As shown, it includes: Step 201: Input the fluctuation analysis results into a convolutional neural network, and generate a spatiotemporal feature sequence through the alternating processing of multiple convolutional layers and pooling layers in the convolutional neural network.

[0047] In step 201, a feature refers to a data set containing features in both time and space dimensions, obtained through multi-layer neural network processing.

[0048] In this embodiment, after the fluctuation analysis results are reorganized into a two-dimensional matrix and input into a convolutional neural network, the first convolutional layer first uses multiple convolutional kernels to extract features to generate preliminary features; then the first pooling layer performs dimensionality reduction on the preliminary features; subsequently, the second convolutional layer further extracts deep features; then the second pooling layer reduces the dimensionality of the deep features again to obtain the final features; finally, the final features are flattened and their dimensions are compressed to generate a spatiotemporal feature sequence.

[0049] Step 202: Process the spatiotemporal feature sequence through the temporal analysis module of the convolutional neural network to generate regular features of the computational load changing over time.

[0050] In step 202, the time-domain analysis module is a component in the convolutional neural network that is specifically designed to process time-series features. Regular features refer to the load change patterns identified from the spatiotemporal feature sequences.

[0051] In this embodiment, the spatiotemporal feature sequence is first input into the time domain analysis module. Then, the feature change trend in the time dimension is analyzed by sliding window to identify periodic fluctuation patterns and sudden change features. Finally, the regular features of the computing load changing over time are extracted.

[0052] Step 203: Based on the business request type and priority, the pattern characteristics are weighted and corrected to generate load trend prediction data.

[0053] In this embodiment of the application, firstly, priority weights are determined according to the types of service requests received in real time. Then, the indicators in the regularity features are weighted based on the aforementioned priority weights. After that, the weighted feature data is fused to finally generate load trend prediction data.

[0054] Here is a specific example: Following the fluctuation analysis results containing 80 values ​​generated in the previous embodiment, this information was obtained by processing 120 performance indicators collected over 10 minutes by three edge servers at locations A, B, and C. First, the 80 values ​​were reassembled into a 20x4 two-dimensional matrix in chronological order, where the 20 rows represent 20 time windows and the 4 columns represent 4 performance indicators. This matrix was then input into the first convolutional layer of a convolutional neural network, where eight 3x3 convolutional kernels were used for feature extraction. Each kernel slid across the matrix with a stride of 1. Within each scanning window, the sum of the products of the kernel parameters and the corresponding data was calculated, using the formula: the feature value equals k. 11 ×a 11 Add k 12 ×a 12 Add k 13 ×a 13 Add k 21 ×a 21 Add k 22 ×a 22 Add k 23 ×a 23 Add k 31 ×a 31 Add k 32 ×a 32 Add k 33 ×a 33 , where k is the convolution kernel parameter and a is the data value in the matrix. After obtaining the feature values, they are processed by the ReLU function to generate 8 first features with 18 rows and 2 columns.

[0055] Next, downsampling is performed through the first pooling layer, using a 2x2 scanning window with a stride of 2 to take the maximum value within each window, resulting in 8 second features of 9x1. Then, the second features are input into the second convolutional layer, and deep feature extraction is performed using 16 convolutional kernels of 3x3 to generate 16 third features of 7x1.

[0056] Then, downsampling is performed through the second pooling layer. Similarly, a 2x2 scanning window is used to slide with a stride of 2, and the maximum value in each window is taken to obtain 16 fourth features in 3x1 columns. The fourth features are then flattened into a one-dimensional vector to obtain 48 values. These 48 values ​​are obtained by using a 16×3×1 method. The dimensionality is reduced to 36 values ​​through a fully connected layer to form a spatiotemporal feature sequence.

[0057] Finally, the spatiotemporal feature sequence is input into the time-domain analysis module, which identifies a periodic fluctuation pattern in the computing load every 15 minutes, with the peak load being 2.3 times the normal value. At this time, the priority weight of the video analysis service request is 0.7 and the priority weight of the IoT data processing service request is 0.3. The pattern feature is then weighted and corrected to adjust the peak load of the video analysis service to 2.3 × 0.7 equals 1.61 and the peak load of the IoT service to 2.3 × 0.3 equals 0.69. Finally, load trend prediction data is generated, predicting that the peak load in the next cycle will reach 2.3 times the normal value, with the video analysis service contributing 1.61 times the load and the IoT service contributing 0.69 times the load.

[0058] In this embodiment, the spatiotemporal features of resource fluctuations are effectively extracted through multi-layer convolution processing, and the load change pattern is accurately identified by combining time-domain analysis. Then, targeted adjustments are made according to business priorities to generate accurate load prediction data, providing a reliable basis for resource scheduling and improving the resource utilization efficiency and service quality of the edge computing environment.

[0059] To address the issue of insufficient spatiotemporal feature extraction in fluctuation analysis results, in some embodiments, step 201 involves inputting the fluctuation analysis results into a convolutional neural network. Through alternating processing by multiple convolutional and pooling layers within the convolutional neural network, a spatiotemporal feature sequence is generated, including: Step 301: Input the fluctuation analysis results into a convolutional neural network and reorganize the fluctuation analysis results into a two-dimensional feature matrix according to the time dimension.

[0060] In step 301, the two-dimensional feature matrix refers to a two-dimensional array in which the fluctuation analysis results are arranged in chronological order, where rows represent the time dimension and columns represent the performance index dimension.

[0061] In this embodiment, the fluctuation analysis results are first arranged in the order of collection time. Then, the data of different performance indicators at the same time point are organized in the same row, and the data at different time points are organized in different rows, finally forming a two-dimensional matrix with time dimension and indicator dimension.

[0062] Step 302: Extract features from the two-dimensional feature matrix through the first convolutional layer of the convolutional neural network to generate the first feature.

[0063] In step 302, the first feature refers to the feature data set after processing by the first convolutional layer, which contains preliminary features extracted from the original matrix.

[0064] In this embodiment, after the two-dimensional feature matrix is ​​input into the first convolutional layer, multiple convolutional kernels are used to perform sliding scans on the matrix; then, at each scan position, the sum of the products of the convolutional kernel parameters and the corresponding local data is calculated, and the calculation result is activated by a nonlinear activation function to finally generate the first feature.

[0065] Step 303: The first feature is downsampled through the first pooling layer of the convolutional neural network to obtain the second feature.

[0066] In step 303, the second feature refers to the feature data after dimensionality reduction processing by the first pooling layer, which retains the main features while reducing the amount of data.

[0067] In this embodiment, after the first feature is input into the first pooling layer, the feature is downsampled using a sliding window method; within each sliding window, an output value representing the window is generated by selecting the maximum value or calculating the average value, thereby obtaining a second feature with half the dimension.

[0068] Step 304: Through the second convolutional layer of the convolutional neural network, perform deep feature extraction on the second feature to generate the third feature.

[0069] In step 304, the third feature refers to the feature data extracted from the deeper layers of the second convolutional layer, which contains more abstract and high-level feature representations.

[0070] In this embodiment, during the process of inputting the second feature into the second convolutional layer, the layer uses a larger number of convolutional kernels to extract features at a deeper level; by implementing convolution operations, the complex relationships between features can be effectively captured, thereby generating a third feature with a higher degree of abstraction.

[0071] Step 305: The third feature is downsampled twice through the second pooling layer of the convolutional neural network to obtain the fourth feature.

[0072] In step 305, the fourth feature refers to the feature data after the second pooling process, which further compresses the feature dimension.

[0073] In this embodiment, after the third feature is input into the second pooling layer, it is downsampled again. This downsampling process is performed in the same way as the first pooling layer to further reduce the feature dimension and finally obtain the fourth feature.

[0074] Step 306: Flatten the fourth feature to generate a one-dimensional feature vector, and compress the one-dimensional feature vector to generate a spatiotemporal feature sequence.

[0075] In step 306, a one-dimensional feature vector refers to a single-dimensional numerical sequence formed by flattening multi-dimensional features. Each element of the vector represents the feature intensity value at a specific location in the feature. This vector retains all the feature information of the original feature but is organized in the form of a linear sequence, which facilitates subsequent dimensionality reduction and feature extraction operations.

[0076] In this embodiment, the fourth feature is flattened into a one-dimensional vector, and the vector is compressed through a fully connected layer or dimensionality reduction processing to finally generate sequence data containing spatiotemporal features.

[0077] Here is a specific example: Following the fluctuation analysis results containing 80 values ​​generated in the previous embodiment, these results were obtained by processing 120 performance indicators collected over 10 minutes by three edge servers at locations A, B, and C. First, the fluctuation analysis results were reorganized into a two-dimensional feature matrix along the time dimension, forming a 20-row, 4-column matrix structure. The 20 rows represent 20 time windows, and the 4 columns represent 4 performance indicators. Then, this two-dimensional feature matrix was input into the first convolutional layer of a convolutional neural network, using eight 3x3 convolutional kernels for feature extraction. Each kernel slides across the matrix with a stride of 1. Then, within each scanning window, the sum of the products of the convolutional kernel parameters and the corresponding data position was calculated, with the formula being that the eigenvalue equals k. 11 ×a 11 Add k 12 ×a 12 Add k 13 ×a 13 Add k 21 ×a 21 Add k 22 ×a 22 Add k 23 ×a 23 Add k 31 ×a 31 Add k 32 ×a 32 Add k 33 ×a 33 , where k is the convolution kernel parameter and a is the data value in the matrix; after obtaining the feature values, they are processed by the ReLU function to finally generate 8 first features with 18 rows and 2 columns.

[0078] Next, the first feature is downsampled through the first pooling layer. A 2x2 scanning window is used to slide with a step size of 2, and the maximum value in each window is taken to obtain 8 second features with 9x1 rows.

[0079] The second feature is then input into the second convolutional layer, and deep feature extraction is performed using 16 convolutional kernels with 3 rows and 3 columns, generating 16 third features with 7 rows and 1 column.

[0080] The third feature is then downsampled again using a second pooling layer, employing a 2x2 scanning window with a stride of 2. The maximum value within each window is taken, resulting in 16 fourth features (3x1). These fourth features are then flattened, transforming them into a one-dimensional feature vector containing 16 × 3 × 1 equals 48 values, represented as [v1, v2, ..., v...]. 48 ].

[0081] Finally, the one-dimensional feature vector is compressed by using a fully connected layer to reduce the 48-dimensional vector to 36-dimensionality, generating a spatiotemporal feature sequence, represented as [s1, s2, ..., s]. 36 Each value represents the feature intensity in different spatiotemporal dimensions, completing the transformation process from fluctuation analysis results to temporal feature representation.

[0082] In this embodiment, by alternating multi-layer convolution and pooling processes, the spatiotemporal features in the fluctuation analysis results are effectively extracted. While retaining key features, the data dimension is gradually compressed, and finally, sequence data containing temporal correlation and spatial distribution features is generated, providing high-quality feature input for subsequent load prediction.

[0083] To address the issue of insufficient feature extraction from the two-dimensional feature matrix, in some embodiments, step 302: extracting features from the two-dimensional feature matrix to generate a first feature, includes: Step 401: Use a set of convolutional kernels of a preset size to perform a sliding window scan along the row and column directions of the two-dimensional feature matrix.

[0084] In this embodiment, a set of convolutional kernels of a preset size is used as a scanning window, which is moved sequentially along the row and column directions of the two-dimensional feature matrix with a specified step size to ensure that every possible position of the matrix is ​​covered.

[0085] Step 402: At each sliding window position, calculate the product of the convolution kernel parameters and the data at the corresponding position within the sliding window to obtain multiple feature values.

[0086] In step 402, the feature value refers to the result of the sum of the products of the convolution kernel parameters and the local data, reflecting the degree of local feature matching.

[0087] In the embodiments of this application, firstly, at each position where the sliding window stops, each parameter of the convolution kernel is multiplied with the data at the corresponding position in the window, and then all the product results are added together to obtain the feature value. It should be noted that in the process of generating the feature value, each convolution kernel generates one feature value.

[0088] Step 403: Based on each of the aforementioned feature values, generate the corresponding feature representation using a nonlinear transformation function.

[0089] In step 403, the feature representation refers to the feature representation obtained after performing a nonlinear transformation on the feature values.

[0090] In the embodiments of this application, a nonlinear transformation function is applied to each feature value to convert the feature value into a non-negative value, thereby enhancing the expressive power of the feature. It should be noted that in the feature representation generation process, each convolutional kernel generates a feature representation.

[0091] Step 404: Combine and arrange all the aforementioned features in a spatial dimension to form the first feature.

[0092] In this embodiment of the application, the feature representations generated by each convolution kernel are arranged and combined in channel order to form a feature containing multiple channels, wherein each channel corresponds to a feature extracted by a convolution kernel.

[0093] In this embodiment, the input matrix is ​​fully covered by a sliding window scan, and local features are extracted using convolution operations. Then, nonlinear transformations are applied to enhance the expressive power of the features, and finally, multi-channel features are combined to effectively capture the spatial features in the fluctuation analysis results, thus laying the foundation for subsequent feature extraction.

[0094] To address the issue of insufficient analysis of server performance metrics, in some embodiments, step 102: performing time-series fluctuation analysis on the server performance metrics to generate fluctuation analysis results includes: Step 501: Arrange all server performance metrics in chronological order to form a performance metric time series.

[0095] In step 501, all server performance metrics refer to the server performance metrics of multiple edge servers. That is, the server performance metrics include the CPU utilization, GPU utilization, memory usage, and I / O wait time of each of the distributed edge servers. The performance metric time series refers to the sequence data formed by arranging the collected performance metrics in chronological order.

[0096] In this embodiment of the application, the performance indicators of multiple servers collected in a distributed manner are sorted according to the collection timestamp, and the indicator data at the same time point are organized together to form continuous time series data.

[0097] Step 502: Divide the performance index time series into multiple time segments according to a fixed time window.

[0098] In step 502, a time segment refers to a subsequence segment obtained by dividing a complete time series into segments of fixed duration.

[0099] In this embodiment of the application, a fixed-duration time window is used to segment the time series of performance indicators to form multiple time segments, wherein each time window contains data from multiple consecutive time points.

[0100] Step 503: Calculate the fluctuation amplitude and fluctuation frequency for each time segment.

[0101] In step 503, fluctuation amplitude refers to the degree of drastic change in data, and fluctuation frequency refers to the frequency of data fluctuation.

[0102] In this embodiment of the application, when calculating the change amplitude between adjacent data points for each time segment, the fluctuation amplitude can be obtained by calculating the sum of squared differences, and the fluctuation frequency can be obtained by counting the number of fluctuations exceeding a set threshold within a unit time. It should be noted that this embodiment of the application does not specifically limit the size of the preset threshold, which can be set according to the actual situation.

[0103] Step 504: Weight and fuse the fluctuation amplitude and the fluctuation frequency to form a quantitative indicator.

[0104] In step 504, the quantitative indicator refers to the comprehensive indicator formed by quantifying the fluctuation characteristics.

[0105] In this embodiment of the application, the fluctuation amplitude and fluctuation frequency are weighted according to a preset weight to integrate the two features into a comprehensive value and form a quantitative index.

[0106] Step 505: Construct volatility analysis results based on the changing trends of quantitative indicators across all time windows.

[0107] In step 505, the trend of change is obtained by arranging the quantitative indicators calculated for each time window in chronological order and analyzing their numerical change patterns. It means the overall pattern and regularity of the evolution of resource fluctuation characteristics over time. The trend of change includes characteristics such as the periodic change of fluctuation intensity, the frequency of sudden peak occurrences, and long-term trends, which are used to comprehensively characterize the dynamic behavior of resources.

[0108] In this embodiment of the application, the original performance indicators can be converted into feature data with clear physical meaning through systematic time series analysis and fluctuation feature quantification, thereby accurately capturing the dynamic change pattern of resource use and providing reliable input data for subsequent feature extraction and load prediction.

[0109] To address the problem of poor communication signal transmission quality, in some embodiments, step 104, which involves preprocessing the communication signal to obtain a preprocessed communication signal, includes: Step 601: Perform a Fourier transform on the communication signal to obtain a frequency domain signal.

[0110] In step 601, the frequency domain signal refers to the signal representation after the time domain signal has been converted to the frequency domain.

[0111] In this embodiment, the communication signal in time domain form is converted into a frequency domain representation by Fourier transform to obtain the amplitude and phase information of the signal at each frequency component.

[0112] Step 602: The frequency domain signal is segmented and scanned with a fixed bandwidth to form the power distribution of each frequency band.

[0113] In step 602, the power distribution is used to reflect the power intensity distribution of the signal in different frequency bands.

[0114] In this embodiment, a sliding window is used to scan the frequency domain signal in segments, and the sum of the power values ​​of all frequency points in each frequency band is calculated to form the power distribution of each frequency band.

[0115] Step 603: Based on the power distribution of each frequency band, mark the frequency bands with signal power less than the preset background noise baseline or greater than the preset transmission threshold as abnormal frequency bands.

[0116] In step 603, the signal power is a value obtained by squaring and integrating the amplitude value of the signal in a specific frequency band, which represents the energy intensity of the signal in that frequency band; the abnormal frequency band refers to the frequency range in which the power value exceeds the normal range.

[0117] In this embodiment of the application, frequency bands with excessively low or high power values ​​are marked as abnormal frequency bands based on a preset background noise baseline and a preset transmission threshold. These frequency bands may have noise interference or signal distortion.

[0118] Step 604: Filter the abnormal frequency band and perform an inverse Fourier transform on the filtered frequency band to obtain the recovered communication signal.

[0119] In this embodiment, the marked abnormal frequency bands are processed in a graded manner. Specifically, the frequency bands with too low power are compensated for gain, and the frequency bands with too high power are attenuated and adjusted. Then, the time domain signal is recovered by inverse Fourier transform.

[0120] Step 605: Perform segmented signal-to-noise ratio detection on the recovered communication signal, and based on the detection results, perform secondary filtering on the substandard signal segments to generate a preprocessed communication signal.

[0121] In step 605, segmented signal-to-noise ratio detection refers to dividing the signal into small segments and measuring the signal-to-noise ratio; the preprocessed communication signal refers to the signal that meets the preset transmission quality conditions. The preset transmission quality conditions refer to the minimum quality standards that the communication signal needs to meet before transmission. The preset transmission quality conditions include specific indicators such as a signal-to-noise ratio of not less than 30dB, a bit error rate of less than 10^-6, and a signal distortion of less than 5%.

[0122] In this embodiment of the application, the recovered time-domain signal is divided into multiple small time periods, and then the signal-to-noise ratio (SNR) value of each time period is calculated. Then, secondary filtering is performed on the time periods with substandard SNR to ensure the overall signal quality.

[0123] In this embodiment, noise interference is effectively suppressed and signal quality is improved through frequency domain analysis and filtering. Then, secondary filtering ensures the reliability of signal transmission and provides a guarantee for the accurate transmission of task scheduling instructions.

[0124] To address the issue of poor matching between task scheduling schemes and resource status, in some embodiments, step 105: processing the load trend prediction data using an adaptive genetic algorithm to generate a task scheduling scheme includes: Step 701: Encode the load trend prediction data into chromosomal gene sequences.

[0125] In step 701, each chromosome gene corresponds to a task allocation strategy parameter, and the chromosome gene sequence refers to the serialized data structure formed by encoding the task allocation strategy parameter.

[0126] In this embodiment of the application, each parameter in the load trend prediction data is converted into a gene sequence according to a predetermined rule, and each gene represents a specific task allocation strategy parameter.

[0127] Step 702: Calculate the population initialization parameters dynamically based on the server performance indicators, including population size and gene length.

[0128] In step 702, the population initialization parameters refer to the basic parameters that need to be set at the beginning of the genetic algorithm.

[0129] In this embodiment of the application, the population size and gene length are dynamically calculated based on the real-time collected server performance indicators to ensure that the parameter settings match the current resource status.

[0130] Step 703: Generate an initial population based on the chromosome gene sequence and the population initialization parameters.

[0131] In step 703, the initial population refers to the first set of individuals generated at the start of the algorithm.

[0132] In this embodiment of the application, a specified number of individuals are randomly generated based on the chromosome gene sequence and population initialization parameters, and each individual represents a possible task allocation scheme.

[0133] Step 704: Perform evolutionary operations on the population. In the evolved population, individuals with fitness values ​​greater than a preset fitness threshold are retained through a roulette wheel selection mechanism to form a new generation of population.

[0134] In step 704, the fitness value is a numerical comparison result calculated by the fitness function. It means the comprehensive performance of the task allocation strategy represented by the chromosome gene sequence in terms of task response time and resource balance. Specifically, it is represented by the output value of the fitness function. The larger the value, the higher the fitness.

[0135] In this embodiment of the application, crossover and mutation operations are performed on the initial population, and then individuals with high fitness values ​​are selected through a roulette wheel selection mechanism to form a new generation of population.

[0136] Step 705: Repeat the evolution and selection operations until the last generation of the population reaches the preset number of generations or the preset fitness convergence threshold, then terminate the iteration process.

[0137] In step 705, the iteration termination condition refers to the criterion for determining when the algorithm stops evolving.

[0138] Step 706: Select the optimal chromosome from the final population at the end of the iteration for decoding, map the decoding result into a task scheduling instruction, and form a task scheduling scheme based on the task scheduling instruction.

[0139] In this embodiment of the application, the chromosome with the highest fitness is selected from the final population, its gene sequence is decoded, and mapped to specific task scheduling instructions to form a complete task scheduling scheme.

[0140] In this embodiment, an adaptive genetic algorithm is used to dynamically optimize the task allocation strategy and generate a scheduling scheme that is highly matched with the current resource status, thereby improving resource utilization efficiency and task execution performance and ensuring the stable operation of the edge computing environment.

[0141] Figure 3 A schematic diagram of an edge computing preprocessing communication system based on server performance metrics is provided in this application embodiment. The system includes: The acquisition module 31 is used to acquire server performance indicators and communication signals during the operation of the edge server.

[0142] Analysis module 32 is used to perform time-series fluctuation analysis on the server performance indicators and generate fluctuation analysis results.

[0143] The extraction module 33 is used to extract features from the fluctuation analysis results using a convolutional neural network, extract spatiotemporal feature sequences, and generate load trend prediction data based on the spatiotemporal feature sequences and in combination with business request types and priorities.

[0144] The suppression module 34 is used to preprocess the communication signal to obtain a preprocessed communication signal.

[0145] The generation module 35 is used to process the load trend prediction data using an adaptive genetic algorithm, generate a task scheduling scheme, and transmit the task scheduling scheme to the corresponding edge server through the preprocessed communication signal.

[0146] The edge computing preprocessing communication system based on server performance indicators in this application is used to implement the aforementioned edge computing preprocessing communication method based on server performance indicators. Therefore, the specific implementation of the edge computing preprocessing communication system based on server performance indicators can be found in the embodiment section of the edge computing preprocessing communication method based on server performance indicators above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0147] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described edge computing preprocessing communication methods based on server performance metrics.

[0148] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described edge computing preprocessing communication methods based on server performance metrics.

[0149] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0150] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the edge computing preprocessing communication method based on server performance indicators.

[0151] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0152] The foregoing has provided a detailed description of an edge computing preprocessing communication method and system based on server performance indicators provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. An edge computing preprocessing communication method based on server performance metrics, characterized in that, include: Collect server performance metrics and communication signals during edge server operation; Perform time-series fluctuation analysis on the server performance indicators to generate fluctuation analysis results; The fluctuation analysis results are used to extract features using a convolutional neural network to extract spatiotemporal feature sequences. Based on the spatiotemporal feature sequences, combined with the business request type and priority, load trend prediction data is generated. The communication signal is preprocessed to obtain the preprocessed communication signal; An adaptive genetic algorithm is used to process the load trend prediction data, generate a task scheduling scheme, and transmit the task scheduling scheme to the corresponding edge server through the preprocessed communication signal.

2. The method according to claim 1, characterized in that, The process involves using a convolutional neural network to extract features from the fluctuation analysis results, extracting a spatiotemporal feature sequence, and generating load trend prediction data based on the spatiotemporal feature sequence, combined with business request type and priority. This includes: The fluctuation analysis results are input into a convolutional neural network, and a spatiotemporal feature sequence is generated through the alternating processing of multiple convolutional layers and pooling layers in the convolutional neural network. The temporal analysis module of the convolutional neural network processes the spatiotemporal feature sequence to generate regular features of computational load changes over time. Based on the type and priority of business requests, the aforementioned patterns are weighted and adjusted to generate load trend prediction data.

3. The method according to claim 2, characterized in that, The step involves inputting the fluctuation analysis results into a convolutional neural network, and generating a spatiotemporal feature sequence through alternating processing of multiple convolutional layers and pooling layers in the convolutional neural network, including: The fluctuation analysis results are input into a convolutional neural network, and the fluctuation analysis results are reorganized into a two-dimensional feature matrix according to the time dimension; The first feature is generated by extracting features from the two-dimensional feature matrix through the first convolutional layer of the convolutional neural network. The first feature is downsampled through the first pooling layer of the convolutional neural network to obtain the second feature. The second feature is extracted from the second feature through the second convolutional layer of the convolutional neural network to generate the third feature; The third feature is downsampled twice through the second pooling layer of the convolutional neural network to obtain the fourth feature; The fourth feature is flattened to generate a one-dimensional feature vector, and the one-dimensional feature vector is then compressed to generate a spatiotemporal feature sequence.

4. The method according to claim 3, characterized in that, The step of extracting features from the two-dimensional feature matrix to generate the first feature includes: A sliding window scan is performed along the row and column directions of the two-dimensional feature matrix using a set of convolutional kernels of a preset size; At each sliding window position, the product of the convolution kernel parameters and the data at the corresponding position within the sliding window is calculated to obtain multiple feature values; Based on each of the aforementioned feature values, a corresponding feature representation is generated using a nonlinear transformation function; All the aforementioned features are combined and arranged in a spatial dimension to form the first feature.

5. The method according to claim 1, characterized in that, The step of performing time-series fluctuation analysis on the server performance indicators to generate fluctuation analysis results includes: Arrange all server performance metrics in chronological order to form a performance metric time series; The performance index time series is divided into multiple time segments according to a fixed time window; Calculate the fluctuation amplitude and frequency for each time segment; The fluctuation amplitude and the fluctuation frequency are weighted and fused to form a quantitative indicator; Based on the changing trends of quantitative indicators across all time windows, volatility analysis results are constructed.

6. The method according to claim 1, characterized in that, The step of preprocessing the communication signal to obtain a preprocessed communication signal includes: Perform a Fourier transform on the communication signal to obtain a frequency domain signal; The frequency domain signal is segmented and scanned with a fixed bandwidth to form the power distribution of each frequency band; Based on the power distribution of each frequency band, frequency bands with signal power less than the preset background noise baseline or greater than the preset transmission threshold are marked as abnormal frequency bands. The abnormal frequency band is filtered, and the filtered frequency band is subjected to inverse Fourier transform to obtain the recovered communication signal. The recovered communication signal is segmented for signal-to-noise ratio detection, and based on the detection results, the substandard signal segments are subjected to secondary filtering to generate a preprocessed communication signal.

7. The method according to claim 1, characterized in that, The process of using an adaptive genetic algorithm to process the load trend prediction data and generate a task scheduling scheme includes: The load trend prediction data is encoded into chromosome gene sequences; Based on the server performance metrics, dynamically calculate the population initialization parameters, including population size and gene length; An initial population is generated based on the chromosome gene sequence and the population initialization parameters; The population is subjected to evolutionary operations. After the evolution, individuals with fitness values ​​greater than a preset fitness threshold are retained in the population through a roulette wheel selection mechanism to form a new generation of population. Repeat the evolution and selection operations until the last generation of the population reaches the preset number of generations or the preset fitness convergence threshold, then terminate the iteration process. The optimal chromosome is selected from the final population at the end of the iteration for decoding. The decoding result is mapped to a task scheduling instruction, and a task scheduling scheme is formed based on the task scheduling instruction.

8. An edge computing preprocessing communication system based on server performance metrics, characterized in that, include: The acquisition module is used to collect server performance metrics and communication signals during the operation of the edge server; The analysis module is used to perform time-series fluctuation analysis on the server performance indicators and generate fluctuation analysis results. The extraction module is used to extract features from the fluctuation analysis results using a convolutional neural network, extract spatiotemporal feature sequences, and generate load trend prediction data based on the spatiotemporal feature sequences and in combination with business request types and priorities. The suppression module is used to preprocess the communication signal to obtain a preprocessed communication signal; The generation module is used to process the load trend prediction data using an adaptive genetic algorithm, generate a task scheduling scheme, and transmit the task scheduling scheme to the corresponding edge server through the preprocessed communication signal.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the edge computing preprocessing communication method based on server performance metrics as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the edge computing preprocessing communication method based on server performance metrics as described in any one of claims 1 to 7.

Images