Edge computing multi-service quality prediction method, system and device and storage medium

Abstract

The invention relates to the technical field of telecommunication, in particular to an edge computing multi-service quality prediction method, system and device and a storage medium, and the method comprises the steps: firstly, fusing user request information, network environment information and service operation information which affect the service quality of an edge server to obtain an edge server vector; carrying out feature extraction on the edge server vector based on multi-head attention processing, residual connection and normalization processing to obtain an edge server feature vector; then, the edge server feature vectors are subjected to multi-layer time sequence feature analysis processing, the relation between different scene factor features is obtained, the non-linear expression ability of the different scene factor features is enhanced, and edge server feature association vectors are obtained; and finally, performing multi-task quality prediction processing on the edge server feature association vector to obtain a plurality of service quality prediction values, thereby realizing high accuracy and high reliability of edge server prediction.

Description

Technical Field

[0001] The present invention relates to the field of electric communication technologies, and particularly to an edge computing multi-quality-of-service prediction method, system, device, and storage medium. Background Art

[0002] With the rapid development of mobile edge computing technology, a large number of edge computing services with diverse functions from different fields have emerged on the network, greatly improving the user experience and convenience. In practical applications, users often have specific quality requirements for the services they use, and these requirements are usually reflected in different dimensions of the quality of service (QoS), such as the expected response time not exceeding 60 minutes, the service availability being higher than 85%, the service reliability reaching more than 90%, etc. To ensure that user requirements are met, the service system needs to select or integrate the optimal edge computing service that meets the user's multi-faceted quality-of-service requirements from numerous service options. Therefore, in an increasingly complex network and service environment, accurately predicting the quality of service has become a crucial task for improving the user experience.

[0003] Existing technologies include linear regression, logistic regression, exponential smoothing, support vector machines, random forests, Bayesian networks, and artificial neural networks, etc. These technologies often predict the future quality of service by analyzing past service behaviors. However, these methods are difficult to effectively capture complex temporal patterns and non-linear relationships. In the context where changes in each context factor in the current mobile edge computing environment will cause fluctuations and impacts on the quality of service, the non-linear expression ability when dealing with quality-of-service issues is also relatively low, and it cannot fully explore the complex relationships between various factors and parameters of the quality of service, resulting in a decline in the quality-of-service prediction performance and a reduction in the user experience. Summary of the Invention

[0004] The purpose of the present invention is to provide an edge computing multi-quality-of-service prediction method, system, device, and storage medium.

[0005] The technical solution of the present invention is as follows:

[0006] An edge computing multi-quality-of-service prediction method, comprising:

[0007] S1. Obtain the user request information, network environment information, and service operation information of the edge server. After being respectively subjected to embedding processing and then spliced, an edge server vector is obtained; the edge server vector is subjected to multi-head attention processing, residual connection, and normalization processing to obtain an edge server feature vector; the user request information includes the number of user connections and the size of random task data; the network environment information includes network bandwidth, network latency, and computing speed; the service operation information includes the total amount of tasks to be processed, task load, and queued task volume;

[0008] S2. The edge server feature vector is processed through multi-layer time-series feature analysis to obtain an edge server feature correlation vector. During the multi-layer time-series feature analysis process: the output of the current layer's time-series feature analysis is fused with the edge server feature vector to obtain the current layer's initial fusion vector; the current layer's initial fusion vector is processed to enhance the non-linear expression ability to obtain the current layer's fusion feature vector, which is used as the input for the next layer's time-series feature analysis; the operation of enhancing the non-linear expression ability is specifically as follows: the current layer's initial fusion vector is multiplied by the current layer's weight and weighted to obtain the current layer's vector to be non-linearly processed; the current layer's vector to be non-linearly processed is mapped to obtain the current layer's initial fusion mapping value; it is determined whether the current layer's initial fusion mapping value is greater than the mapping value threshold; if it is greater, based on the mean of the edge server feature vector and the mean of the current layer's initial fusion vector, the current layer's vector to be non-linearly processed is subjected to the first non-linear processing to obtain the first non-linear enhancement vector, which is used as the current layer's fusion feature vector; if it is not greater, based on the variance of the edge server feature vector and the variance of the current layer's initial fusion vector, the current layer's vector to be non-linearly processed is subjected to the second non-linear processing to obtain the second non-linear enhancement vector, which is used as the current layer's fusion feature vector.

[0009] S3. The edge server feature correlation vector is processed through multi-task quality prediction to obtain several service quality prediction values.

[0010] In S2, the first non-linear processing is implemented through the following formula: y i,1 =(ω 1 (μ - μ 0 ) + γ)X i , and the second non-linear processing is implemented through the following formula: y i,1 , y i,2 are the first non-linear enhancement vector of the i-th layer and the second non-linear enhancement vector of the i-th layer respectively, ω 1 , ω 2 are the first learning rate and the second learning rate respectively, μ, μ 0 are the mean of the edge server feature vector and the mean of the current layer's initial fusion vector respectively, σ, σ 0 are the variance of the edge server feature vector and the variance of the current layer's initial fusion vector respectively, γ, ε are the first slope value and the second slope value respectively, γ > ε, X i is the initial fusion vector of the i-th layer.

[0011] The operations of multi-layer time series feature analysis and processing in S2 can be implemented by a time series analysis model composed of several time series feature analysis modules; in the time series analysis model, several time series feature analysis modules are distributed in a rectangular array, the data propagation direction of the time series feature analysis modules in the same column is from bottom to top, and the data propagation direction of the time series feature analysis modules in the same row and column is from left to right.

[0012] In the time series feature analysis module, the operations of time series feature analysis and processing are specifically as follows: after the inputs are respectively processed by the first activation function, hyperbolic tangent processing, and forgetting processing, fusion processing is performed to obtain memory features; the inputs are processed by the second activation function to obtain second non-linear features; the memory features are processed by hyperbolic tangent and then fused with the second non-linear features to obtain the output.

[0013] The total amount of tasks to be processed in S1 is the cumulative amount of all task data of the edge server in the first neighborhood time period; the task load is the sum of the number of tasks that are running and waiting to run on the edge server in the second neighborhood time period, and the second neighborhood time period is within the first neighborhood time period.

[0014] In S1, it also includes that after the average geographical location and average usage time period of all users connecting to the edge server are respectively embedded and then concatenated with the edge server vector, an edge server optimization vector is obtained, which is used to perform operations such as multi-head attention processing, residual connection, and normalization processing.

[0015] In the multi-task quality prediction processing in S3, each task quality prediction processing is implemented by a multi-layer fully connected neural network layer.

[0016] An edge computing multi-service quality prediction system for implementing the above-mentioned edge computing multi-service quality prediction method includes:

[0017] An edge server feature vector generation module, which is used to obtain the user request information, network environment information, and service operation information of the edge server, respectively perform embedding processing, and then perform concatenation to obtain an edge server vector; the edge server vector is processed by multi-head attention, residual connection, and normalization to obtain an edge server feature vector; the user request information includes the number of user connections and the size of random task data; the network environment information includes network bandwidth, network latency, and computing speed; the service operation information includes the total amount of tasks to be processed, task load, and queued task volume;

[0018] The edge server feature correlation vector generation module is used to obtain the edge server feature correlation vector through multi-layer time series feature analysis and processing of the edge server feature vector. During the multi-layer time series feature analysis and processing: the output of the current layer time series feature analysis and processing is fused with the edge server feature vector to obtain the current layer initial fusion vector; the current layer initial fusion vector is processed through enhanced non-linear expression ability to obtain the current layer fusion feature vector, which is used as the input for the next layer of time series feature analysis and processing. The operation of enhancing the non-linear expression ability is specifically as follows: the current layer initial fusion vector is multiplied by the current layer weight and weighted to obtain the current layer vector to be non-linearly processed; the current layer vector to be non-linearly processed is mapped to obtain the current layer initial fusion mapping value; it is judged whether the current layer initial fusion mapping value is greater than the mapping value threshold; if it is greater, based on the mean value of the edge server feature vector and the mean value of the current layer initial fusion vector, the first non-linear processing is performed on the current layer vector to be non-linearly processed to obtain the first non-linear enhanced vector, which is used as the current layer fusion feature vector; if it is not greater, based on the variance of the edge server feature vector and the variance of the current layer initial fusion vector, the second non-linear processing is performed on the current layer vector to be non-linearly processed to obtain the second non-linear enhanced vector, which is used as the current layer fusion feature vector.

[0019] The service quality prediction value generation module is used to obtain a plurality of service quality prediction values through multi-task quality prediction processing of the edge server feature correlation vector.

[0020] An edge computing multi-service quality prediction device includes a processor and a memory. Among them, when the processor executes the computer program stored in the memory, the above-mentioned edge computing multi-service quality prediction method is implemented.

[0021] A computer-readable storage medium is used to store a computer program. Among them, when the computer program is executed by a processor, the above-mentioned edge computing multi-service quality prediction method is implemented.

[0022] The beneficial effects of the present invention are as follows:

[0023] A method for edge computing multi-service quality prediction provided by the present invention first fuses the user request information, network environment information, and service operation information that affect the service quality of the edge server to obtain the edge server vector, and based on multi-head attention processing, residual connection, and normalization processing, feature extraction is performed on the edge server vector to obtain the edge server feature vector; then, the edge server feature vector is processed through multi-layer time series feature analysis to obtain the relationship between different scenario factor features and enhance the non-linear expression ability of different scenario factor features to obtain the edge server feature correlation vector; finally, the edge server feature correlation vector is processed through multi-task quality prediction to obtain a plurality of service quality prediction values, realizing high accuracy and high reliability of edge server prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] By reading the following detailed description of the preferred embodiments, the solutions and advantages of the present application will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention.

[0025] In the drawings:

[0026] Figure 1 is a schematic flowchart of the prediction method in this embodiment;

[0027] Figure 2 is a comparison chart of the service reliability accuracy rates of 5 prediction methods in the embodiment;

[0028] Figure 3 is a comparison chart of the service time accuracy rates of 5 prediction methods in the embodiment;

[0029] Figure 4 is a grouped comparison chart of the service reliability accuracy rates of 5 prediction methods in the embodiment;

[0030] Figure 5 is a grouped comparison chart of the service time accuracy rates of 5 prediction methods in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the drawings.

[0032] This embodiment provides an edge computing multi-service quality prediction method, see Figure 1 , including:

[0033] S1. Obtain the user request information, network environment information, and service operation information of the edge server. After embedding processing respectively, they are spliced to obtain an edge server vector; the edge server vector is processed by multi-head attention, residual connection, and normalization processing to obtain an edge server feature vector; the user request information includes the number of user connections and the size of random task data; the network environment information includes network bandwidth, network latency, and computing speed; the service operation information includes the total amount of tasks to be processed, task load, and queue task volume;

[0034] S2. The edge server feature vector is processed through multi-layer time-series feature analysis to obtain an edge server feature correlation vector. During the multi-layer time-series feature analysis process: the output of the current layer's time-series feature analysis processing is fused with the edge server feature vector to obtain the current layer's initial fusion vector; the current layer's initial fusion vector is processed to enhance its non-linear expression ability to obtain the current layer's fusion feature vector, which is used as the input for the next layer's time-series feature analysis processing. The operation of enhancing the non-linear expression ability is specifically as follows: the current layer's initial fusion vector is multiplied by the current layer's weight and weighted to obtain the current layer's vector to be non-linearly processed; the current layer's vector to be non-linearly processed is mapped to obtain the current layer's initial fusion mapping value; it is judged whether the current layer's initial fusion mapping value is greater than the mapping value threshold; if it is greater, based on the mean value of the edge server feature vector and the mean value of the current layer's initial fusion vector, the current layer's vector to be non-linearly processed is subjected to the first non-linear processing to obtain the first non-linearly enhanced vector, which is used as the current layer's fusion feature vector; if it is not greater, based on the variance of the edge server feature vector and the variance of the current layer's initial fusion vector, the current layer's vector to be non-linearly processed is subjected to the second non-linear processing to obtain the second non-linearly enhanced vector, which is used as the current layer's fusion feature vector.

[0035] S3. The edge server feature correlation vector is processed through multi-task quality prediction to obtain several service quality prediction values.

[0036] S1. Obtain the user request information, network environment information, and service operation information of the edge server. After being respectively embedded and spliced, an edge server vector is obtained. The edge server vector is processed through multi-head attention, residual connection, and normalization to obtain an edge server feature vector.

[0037] Fuse the user request information, network environment information, and service operation information that affect the service quality of the edge server to obtain an edge server vector, and based on multi-head attention processing, residual connection, and normalization, perform feature extraction on the edge server vector to obtain an edge server feature vector.

[0038] First, obtain the scenario factors that affect the computing effect of the edge server and the service quality of the edge server, including user request information reflecting the tasks to be processed by the edge server, network environment information reflecting the interaction efficiency between the user and the edge server, and service operation information reflecting the current usage situation of the edge server.

[0039] Among them, the user request information includes the number of user connections and the size of random task data. The number of user connections is the number of devices that use this edge server simultaneously, and the size of random task data is the size of any one task data among all task data of all users connected to the edge server, indicating the amount of data required for task transmission.

[0040] The network environment information includes network bandwidth, network latency, and computing speed. The network bandwidth is the transmission rate of the wireless local area network (WLAN), the network latency is the time required for data to travel from the sender to the receiver, and the computing speed is a performance metric for the edge server, usually expressed in millions of instructions per second (MIPS).

[0041] The service operation information includes the total number of tasks to be processed, task load, and queued task volume. The total number of tasks to be processed is the total amount of tasks that need to be processed, representing the cumulative amount of all task data on the edge server within the first neighborhood time period; the task load is the sum of the number of tasks that are currently running and waiting to run on the edge server within the second neighborhood time period, and the second neighborhood time period is within the first neighborhood time period; the queued task volume is the amount of tasks that need to be processed and are currently in the queue, representing the amount of tasks that have not yet started processing but are already in the queue.

[0042] Then, the user request information, network environment information, and service operation information, namely the number of user connections, the size of random task data, network bandwidth, network latency, computing speed, total number of tasks to be processed, task load, and queued task volume, are each subjected to embedding processing and then concatenated to obtain an edge server vector that can reflect both the processing performance characteristics of the edge server itself and the environmental characteristics of the edge server.

[0043] Finally, to focus on the characteristics of different context factors, the edge server vector is first subjected to multi-head attention processing to focus on different aspects of the input, then residual connection processing to focus on the differences in characteristics before and after multi-head attention processing, and finally normalization processing to obtain the edge server feature vector.

[0044] To further improve the richness of the edge server vector and the accuracy of subsequent prediction effects, it also includes embedding the average geographical location and average usage time period of all users connected to the edge server, and then concatenating them with the edge server vector to obtain an optimized edge server vector for performing operations such as multi-head attention processing, residual connection, and normalization processing.

[0045] S2. The edge server feature vector is processed through multi-layer time series feature analysis to obtain an edge server feature correlation vector.

[0046] The edge server feature vector is processed through multi-layer time series feature analysis to obtain the relationships between different context factor characteristics and enhance the non-linear expression ability of different context factor characteristics, thereby obtaining an edge server feature correlation vector.

[0047] The operations of multi-layer time series feature analysis and processing can be implemented by a time series analysis model composed of several time series feature analysis modules, thereby effectively capturing the long-term time dependence of data in the edge server feature vector, understanding the association of data at different time steps, and thus better analyzing and predicting the operating state of the edge server.

[0048] Among them, in the time series analysis model, several time series feature analysis modules are distributed in a rectangular array. The data propagation direction of the time series feature analysis modules in the same column is from bottom to top, and the data propagation direction of the time series feature analysis modules in the same row and column is from left to right. For example, if the time series analysis model consists of 9 time series feature analysis modules, the 9 time series feature analysis modules are arranged in a 3×3 array. The time series feature analysis module in the lower left corner is the module that performs the first layer of time series feature analysis and processing, which is used to process the edge server feature vector. The time series feature analysis module in the upper right corner is the module that performs the last layer of time series feature analysis and processing, which is used to output the edge server feature correlation vector.

[0049] In addition, in the time series analysis model, several time series feature analysis modules can also be connected in series, which can improve the feature analysis efficiency.

[0050] Among them, the operations of time series feature analysis and processing are specifically as follows: After the input is processed by the first activation function, hyperbolic tangent processing, and forgetting processing respectively, fusion processing is performed to obtain memory features; the input is processed by the second activation function to obtain second non-linear features; the memory features are processed by hyperbolic tangent and then fused with the second non-linear features (which can be achieved by vector multiplication) to obtain the output.

[0051] The operations of the current layer of time series feature analysis and processing can be implemented by the following formula:

[0052] f t =σ(W f ·[h t-1 ,x t +a f ),

[0053] i t =σ(W i ·[h t-1 ,x t +a i ),

[0054] c t =tanh(W c ·[h t-1 ,x t +a c ),

[0055] C t =f t ·Ct-1 +i t ·c t ,

[0056] O t = σ(W O ·[h t-1 , x t +a O ),

[0057] h t = O t ·tanh(C t ),

[0058] h t-1 is the output at time t-1, x t is the initial input at time t, is the fused feature vector of the previous layer, W f is the forgetting weight, a f is the forgetting compensation parameter, f t is the forgetting output at time t, σ is the sigmod function, i t is the first non-linear feature, W i is the weight of the first activation function, a i is the compensation parameter of the first activation function, c t is the first tangent feature (obtained by processing the input through hyperbolic tangent), W c is the first tangent weight, a c is the compensation parameter of the first tangent, C t is the memory feature at time t, C t-1 is the memory feature at time t-1, O t is the second non-linear feature, W O is the weight of the second activation function, a O is the compensation parameter of the second activation function, h t is the output at time t.

[0059] The operation of time series feature analysis and processing can also be: the input is normalized to obtain a standard feature vector; the covariance matrix of the standard feature vector is calculated, which reflects the linear correlation degree between features in the standard feature vector, and the covariance matrix is eigen-decomposed to obtain eigenvalues; based on the eigenvalues, the principal component vector of the input is obtained, which contains the main information of the original data, reduces the dimension of the data at the same time, and the principal components are independent of each other, eliminating the linear correlation between the original features; after multiplying the transpose of the standard feature vector by the principal component vector and adding it to the input, the output is obtained.

[0060] In the above process of multi-layer time series feature analysis and processing: The output of the current layer's time series feature analysis and processing is fused with the edge server feature vector (which can be achieved by horizontal splicing on a one-dimensional vector). This not only retains the key information of the original input but also effectively compensates for the losses in layer-by-layer information transmission, ensuring that important features are always retained and strengthened, resulting in the initial fusion vector of the current layer. The initial fusion vector of the current layer undergoes non-linear expression ability enhancement processing to improve the ability to capture complex features and enhance the non-linear expression ability of features, obtaining the fusion feature vector of the current layer, which serves as the input for the next layer's time series feature analysis and processing.

[0061] The operation of the non-linear expression ability enhancement processing is specifically as follows: The initial fusion vector of the current layer is multiplied by the weight of the current layer and undergoes weighted processing to obtain the vector to be non-linearly processed in the current layer. The vector to be non-linearly processed in the current layer is subjected to a mapping process to map it into specific values, obtaining the initial fusion mapping value of the current layer. It is judged whether the initial fusion mapping value of the current layer is greater than the mapping value threshold. If it is greater, based on the mean of the edge server feature vector and the mean of the initial fusion vector of the current layer, the first non-linear processing is performed on the vector to be non-linearly processed in the current layer to obtain the first non-linearly enhanced vector, which serves as the fusion feature vector of the current layer. If it is not greater, based on the variance of the edge server feature vector and the variance of the initial fusion vector of the current layer, the second non-linear processing is performed on the vector to be non-linearly processed in the current layer to obtain the second non-linearly enhanced vector, which serves as the fusion feature vector of the current layer.

[0062] The first non-linear processing is achieved through the following formula: y i,1 =(ω 1 (μ - μ 0 ) + γ)X i , and the second non-linear processing is achieved through the following formula: y i,1 、y i,2 are the first non-linearly enhanced vector of the i-th layer and the second non-linearly enhanced vector of the i-th layer respectively, ω 1 、ω 2 are the first learning rate and the second learning rate respectively, μ、μ 0 are the mean of the edge server feature vector and the mean of the initial fusion vector of the current layer respectively, σ、σ 0 are the variance of the edge server feature vector and the variance of the initial fusion vector of the current layer respectively, γ、ε are the first slope value and the second slope value (greater than 0 but less than 1), γ > ε, and X i is the vector to be non-linearly processed in the i-th layer.

[0063] S3. The edge server feature correlation vector undergoes multi-task quality prediction processing to obtain several service quality prediction values.

[0064] In multi-task quality prediction processing, the weights and bias values of different task quality predictions are different, and each task quality prediction processing is implemented through several layers of fully connected neural networks. Each layer of fully connected neural network is responsible for performing refined feature extraction on the input features to meet the requirements of specific tasks.

[0065] The operation of each layer of fully connected neural network processing can be implemented by the following formula:

[0066] Z p = σ p (β p Z p-1 + c p ),

[0067] Z p is the output of the p-th layer of fully connected neural network, σ p is the activation function of the p-th layer of fully connected neural network, β p is the weight matrix of the p-th layer of fully connected neural network, Z p-1 is the output of the p-th layer of fully connected neural network, c p is the bias term of the p-th layer of fully connected neural network.

[0068] Multi-task quality prediction processing includes running time prediction and service recommendation reliability prediction.

[0069] This embodiment also provides an edge computing multi-service quality prediction system for implementing the above-mentioned edge computing multi-service quality prediction method, including:

[0070] An edge server feature vector generation module, which is used to obtain the user request information, network environment information, and service operation information of the edge server. After being respectively embedded and then spliced, an edge server vector is obtained. The edge server vector is processed by multi-head attention, residual connection, and normalization to obtain an edge server feature vector. The user request information includes the number of user connections and the size of random task data. The network environment information includes network bandwidth, network latency, and computing speed. The service operation information includes the total amount of tasks to be processed, task load, and queue task volume.

[0071] The edge server feature correlation vector generation module is used to obtain the edge server feature correlation vector through multi-layer time series feature analysis and processing of the edge server feature vector. During the multi-layer time series feature analysis and processing: the output of the current layer time series feature analysis and processing is fused with the edge server feature vector to obtain the current layer initial fusion vector; the current layer initial fusion vector is processed to enhance the non-linear expression ability to obtain the current layer fusion feature vector, which is used as the input for the next layer time series feature analysis and processing. The operation of enhancing the non-linear expression ability is specifically as follows: the current layer initial fusion vector is multiplied by the current layer weight and weighted to obtain the current layer vector to be non-linearly processed; the current layer vector to be non-linearly processed is mapped to obtain the current layer initial fusion mapping value; it is judged whether the current layer initial fusion mapping value is greater than the mapping value threshold; if it is greater, based on the mean value of the edge server feature vector and the mean value of the current layer initial fusion vector, the first non-linear processing is performed on the current layer vector to be non-linearly processed to obtain the first non-linear enhanced vector, which is used as the current layer fusion feature vector; if it is not greater, based on the variance of the edge server feature vector and the variance of the current layer initial fusion vector, the second non-linear processing is performed on the current layer vector to be non-linearly processed to obtain the second non-linear enhanced vector, which is used as the current layer fusion feature vector.

[0072] The quality of service prediction value generation module is used to obtain a number of quality of service prediction values through multi-task quality prediction processing of the edge server feature correlation vector.

[0073] When the system implements the above method, the training sample set (formed by multiple scenario factors of several edge servers and corresponding multi-task quality of service label values) can be obtained, and the system is trained using the training data set. When the loss value obtained by weighted summation of the difference between the predicted value and the label value of each sample is less than the loss threshold, the training ends.

[0074] In addition, to improve the stability of system calculation, during each training process, the first-order moment deviation and second-order moment deviation of the edge server feature correlation vector are corrected, and parameter gradient backpropagation is performed to obtain the corrected training parameters, which are used to update the bias term of the fully connected neural network in the multi-layer time series feature analysis and processing.

[0075] The parameter gradient backpropagation is implemented through the following formula:

[0076]

[0077] β t is the corrected training parameter at time t, β t-1 is the corrected training parameter at time t - 1, η is the gradient backpropagation coefficient, s t is the corrected first-order moment deviation, r t is the corrected second-order moment deviation, and ε is the compensation value.

[0078] This embodiment also provides an edge computing multi-service quality prediction device, including a processor and a memory. When the processor executes the computer program stored in the memory, the above-mentioned edge computing multi-service quality prediction method is implemented.

[0079] This embodiment also provides a computer-readable storage medium for storing a computer program. When the computer program is executed by a processor, the above-mentioned edge computing multi-service quality prediction method is implemented.

[0080] For the edge computing multi-service quality prediction method provided in this embodiment, first, the user request information, network environment information, and service operation information that affect the service quality of the edge server are fused to obtain an edge server vector. Then, based on multi-head attention processing, residual connection, and normalization processing, feature extraction is performed on the edge server vector to obtain an edge server feature vector. Next, the edge server feature vector is processed through multi-layer time-series feature analysis to obtain the relationship between different scenario factor features and enhance the non-linear expression ability of different scenario factor features, resulting in an edge server feature correlation vector. Finally, the edge server feature correlation vector is processed through multi-task quality prediction to obtain several service quality prediction values, achieving high accuracy and high reliability in edge server prediction.

[0081] To verify the effectiveness of the method in this embodiment, the following experiments were conducted.

[0082] Experimental data. In the experiment, the edge computing simulation platform EdgeCloudSim was used to construct a scenario where the elderly in a simulated nursing home environment select a specific service S. During the experiment, the scenario information when the elderly select service S each time and multiple service quality (hereinafter referred to as QoS) indicators of the corresponding service S were recorded. The service quality indicators include running time and accuracy rate. A total of 200,000 service selection records were collected in the experiment. Each record details the specific situation when the elderly select service S and the service quality performance reflected by this selection. These data not only include the selection behavior of users but also cover various QoS parameters during the service execution process, providing a rich information basis for subsequent analysis.

[0083] Evaluation metrics. To evaluate the effectiveness of the method in this embodiment, the mean absolute error (MAE), mean square error (MSE), and accuracy rate (Acc) were used in the experiment to measure the deviation between the predicted value and the true value of QoS. Among them, the smaller the values of MAE and MSE, the higher the prediction accuracy, and the larger the value of Acc, the higher the prediction accuracy.

[0084] Experimental parameters. For the experimental parameter settings, refer to Table 1. The operating system is Windows 11, the graphics card is RTX3070Ti, the programming language is python3.6, and the deep learning framework is TensorFlow.

[0085] Table 1 Experimental parameter settings

[0086] Parameter Value Epoch 200 Batch size 256 Optimizer Adam Loss function Mean squared error Learning rate 0.001

[0087] Comparison with existing technologies. In the experiment, existing technologies such as the Cross-stitch method, U-MTL method,

[0088] MTDNN method, and IMTDNN method were selected for performance comparison with the method of this embodiment to further verify the effectiveness of the method of this embodiment.

[0089] Experimental result 1. To fully verify the effectiveness of the method of this embodiment and avoid the impact of prediction fluctuations caused by data differences, the experiment set different proportions of the test set (hereinafter referred to as TDS, with quantities of 10%, 20%, and 30%) to more comprehensively evaluate the performance of the method under different data scales, thereby ensuring the stability and reliability of the results. In each experiment, three key indicators, namely MAE, MSE, and Acc, were calculated. Finally, by taking the average of the results of 10 experiments, more robust and reliable evaluation results were obtained, as shown in Tables 2 and 3.

[0090] Table 2 Summary of the performance of different methods in reliability prediction on test sets of different quantities

[0091]

[0092] Table 3 Summary of the performance of different methods in running time on test sets of different quantities

[0093]

[0094] From the data results in Tables 2 and 3, it can be seen that the method of this embodiment is superior to the other 4 existing methods in the evaluation metrics Acc, MAE, and MSE when predicting two QoSs (reliability and running time). In particular, on the test data set with TDS = 30%, when the method of this embodiment predicts two QoSs, it is superior to other existing methods in the evaluation metric Acc, with a gain of up to 0.9%; it is also superior to other existing methods in the metric MAE, with a gain of up to 4%; and it is superior to other existing methods in MSE, with a gain of up to 13.3%. From the above experimental results, it can be concluded that the method of this embodiment can not only efficiently process shared features and specific task features, but also better reflect the QoS changes in different scenarios, enhance the robustness of feature representation, reduce information loss, and ultimately achieve a more accurate prediction effect.

[0095] Experimental result 2. To verify the convergence of the method in this embodiment, 70% of the experimental data was used as the training set, while the remaining 30% was used as the test set in the experiment to evaluate the performance of 5 different methods. In this way, not only can the effectiveness of model training be ensured, but also its generalization ability on unseen data can be accurately measured. The experimental results are as Figure 2 and Figure 3 shown, where Figure 2 shows the prediction results of service reliability, Figure 3 and Figure 2 and Figure 3 show the prediction results of service running time. In Figure 2 and Figure 3 , the ordinate represents the prediction accuracy, and the abscissa represents the number of algorithm iterations. It can be seen from Figure 2 and Figure 3 that as the number of iterations increases, the service accuracy and service time prediction accuracy of the 5 methods both show an upward trend. However, the method in this embodiment not only performs well in prediction accuracy, but also its convergence speed is significantly better than the other 4 methods. The method in this embodiment can quickly improve the prediction accuracy at the initial stage of iteration and maintain stable growth in subsequent iterations, which indicates that the method in this embodiment can find the optimal solution in a shorter time, thus improving the prediction efficiency.

[0096] Experimental result 3. To comprehensively evaluate the stability of the prediction ability of the method in this embodiment, 50% of the data was randomly selected from the test set (TDS = 30%) accounting for 30% of the total data set as the sub-test set, and a total of 10 independent samplings were carried out, thus forming 10 different sub-test sets. Subsequently, the method in this embodiment and the other 4 methods were used to test these 10 sub-test sets. The experimental results are as Figure 4 and Figure 5 shown, where Figure 4 shows the prediction accuracy of five different methods for service reliability, while Figure 5 focuses on the prediction accuracy of service running time. In the two figures, the ordinate represents the prediction accuracy, and the abscissa is the serial number of the test data group. It can be clearly seen from Figure 4 and Figure 5 that the method in this embodiment performs well in the prediction of QoS reliability and service time, and its prediction accuracies are respectively stable at about 0.89 and 0.96. This result indicates that no matter how the test data changes, the method in this embodiment can maintain a high prediction accuracy, showing its excellent stability and reliability. In contrast, the prediction accuracies of the other 4 algorithms fluctuate greatly, especially when dealing with different sub-test sets, and the performances of these algorithms are significantly different, further highlighting the advantages of the method in this embodiment.

Claims

1. A method for predicting multi-service quality of edge computing, characterized in that: include: S1. Obtain user request information, network environment information, and service operation information of the edge server, embed them respectively, and then splice them to obtain an edge server vector; The edge server vector is processed by multi-head attention, residual connection and normalization to obtain the edge server feature vector; user request information includes the number of user connections and the size of random task data; network environment information includes network bandwidth, network delay and computing speed; service operation information includes the total number of tasks to be processed, task load and queue task volume; S2, the edge server feature vector is processed by multi-layer time series feature analysis to obtain the edge server feature association vector; In the process of multi-layer time series feature analysis and processing: the output of the current layer time series feature analysis and processing is fused with the edge server feature vector to obtain the current layer initial fusion vector; the current layer initial fusion vector is processed by nonlinear expression capability enhancement to obtain the current layer fusion feature vector as the input of the next layer time series feature analysis and processing; The specific operation of the nonlinear expression capability enhancement processing is as follows: the initial fusion vector of the current layer is multiplied by the weight of the current layer, and a vector to be processed nonlinearly in the current layer is obtained after weighted processing; the vector to be processed nonlinearly in the current layer is mapped to obtain an initial fusion mapping value of the current layer; it is determined whether the initial fusion mapping value of the current layer is greater than the mapping value threshold; if greater, based on the mean value of the feature vector of the edge server and the mean value of the initial fusion vector of the current layer, the first nonlinear processing is performed on the vector to be processed nonlinearly in the current layer to obtain a first nonlinear enhancement vector as the fusion feature vector of the current layer; If it is not greater than, based on the variance of the feature vector of the edge server and the variance of the initial fusion vector of the current layer, perform a second nonlinear processing on the vector to be nonlinearly processed in the current layer to obtain a second nonlinear enhancement vector as the fusion feature vector of the current layer; S3. The edge server feature association vector is processed by multi-task quality prediction to obtain several service quality prediction values.

2. The edge computing multi-service quality prediction method according to claim 1, characterized in that: In S2, The first nonlinear processing is achieved by the following formula: i,1 =(ω1(μ-μ0)+γ)X i , The second nonlinear processing is achieved by the following formula: y i,1 ,y i,2 are the first nonlinear enhancement vector of the i-th layer and the second nonlinear enhancement vector of the i-th layer, ω1 and ω2 are the first learning rate and the second learning rate, μ and μ0 are the mean of the edge server feature vector and the mean of the initial fusion vector of the current layer, ρ and σ0 are the variance of the edge server feature vector and the variance of the initial fusion vector of the current layer, γ and ε are the first slope value and the second slope value, γ>ε, X i is the initial fusion vector of the i-th layer.

3. The edge computing multi-service quality prediction method according to claim 1, characterized in that: In S2, the multi-layer timing feature analysis processing operation can be implemented by a timing analysis model composed of a plurality of timing feature analysis modules; In the timing analysis model, several timing feature analysis modules are distributed in a rectangular array, the data propagation direction of the timing feature analysis modules in the same column is from bottom to top, and the data propagation direction of the timing feature analysis modules in the same row is from left to right.

4. The edge computing multi-service quality prediction method according to claim 3 is characterized in that: In the timing feature analysis module, the timing feature analysis processing operations are specifically as follows: The input is processed by the first activation function, the hyperbolic tangent and the forgetting respectively, and then fused to obtain the memory feature; The input is processed by the second activation function to obtain the second nonlinear feature; The memory feature is processed by hyperbolic tangent and then fused with the second nonlinear feature to obtain the output.

5. The edge computing multi-service quality prediction method according to claim 1, characterized in that: In S1, the total amount of tasks to be processed is the accumulated amount of all task data of the edge server in the first neighborhood time period; the task load is the sum of the number of tasks being run and waiting to be run by the edge server in the second neighborhood time period, and the second neighborhood time period is within the first neighborhood time period.

6. The edge computing multi-service quality prediction method according to claim 1, characterized in that: In the S1, the average geographic location and average usage time period of all users connected to the edge server are embedded and then concatenated with the edge server vector to obtain an edge server optimization vector for performing the multi-head attention processing, residual connection and normalization processing operations.

7. The edge computing multi-service quality prediction method according to claim 1, characterized in that: In S3, in the multi-task quality prediction process, each task quality prediction process is implemented through a multi-layer fully connected neural network layer.

8. An edge computing multi-service quality prediction system, used to implement an edge computing multi-service quality prediction method according to claim 1, characterized in that: include: The edge server feature vector generation module is used to obtain the user request information, network environment information and service operation information of the edge server, and after embedding processing, they are spliced ​​to obtain the edge server vector; the edge server vector is processed by multi-head attention, residual connection and normalization to obtain the edge server feature vector; user request information includes the number of user connections and the size of random task data; network environment information includes network bandwidth, network delay and computing speed; service operation information includes the total number of tasks to be processed, task load and queue task volume; The edge server feature association vector generation module is used to process the edge server feature vector through multi-layer time series feature analysis to obtain the edge server feature association vector; in the process of multi-layer time series feature analysis: the output of the current layer time series feature analysis is fused with the edge server feature vector to obtain the current layer initial fusion vector; the current layer initial fusion vector is processed by nonlinear expression capability enhancement to obtain the current layer fusion feature vector as the input of the next layer time series feature analysis; the operation of the nonlinear expression capability enhancement is specifically as follows: the current layer initial fusion vector is multiplied by the current layer weight and weighted to obtain the current layer nonlinear processing vector; the current layer nonlinear processing vector is mapped to obtain the current layer initial fusion mapping value; it is determined whether the current layer initial fusion mapping value is greater than the mapping value threshold; if greater, based on the edge server feature vector mean and the current layer initial fusion vector mean, the current layer nonlinear processing vector is subjected to the first nonlinear processing to obtain the first nonlinear enhancement vector as the current layer fusion feature vector; If it is not greater than, based on the variance of the feature vector of the edge server and the variance of the initial fusion vector of the current layer, perform a second nonlinear processing on the vector to be nonlinearly processed in the current layer to obtain a second nonlinear enhancement vector as the fusion feature vector of the current layer; The service quality prediction value generation module is used to process the edge server feature association vector through multi-task quality prediction to obtain a number of service quality prediction values.

9. An edge computing multi-service quality prediction device, characterized in that: It includes a processor and a memory, wherein when the processor executes the computer program stored in the memory, it implements the edge computing multi-service quality prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein when the computer program is executed by a processor, the edge computing multi-service quality prediction method as described in any one of claims 1-7 is implemented.

Images