Network resource optimization method and system based on user side computing power monitoring

Abstract

The invention relates to the technical field of network optimization, and provides a network resource optimization method and system based on user side computing power monitoring, and the method comprises the steps: collecting the use data of user computing resources through a lightweight plug-in installed on user equipment, and transmitting a time queue of the collected data to a central server; and in the central server, a data analysis and coding mode based on artificial intelligence is adopted to carry out full-connection embedded coding on the use data of each user computing resource, so that the user activity type is automatically identified according to the time sequence characteristic dynamic propagation representation between the full-connection embedded characteristics of the use condition of each computing resource. And extracting recommended network resources and assigning the recommended network resources to the user equipment. Therefore, the computing power demand of the user can be accurately identified, and the change of the user activity can be responded in real time, so that the resource waste is avoided, and the intelligent management and optimal distribution of the computing power end resources of the user are realized.

Description

Technical Field

[0001] The present application relates to the field of network optimization technology, and more specifically, to a network resource optimization method and system based on user-side computing power monitoring. Background Art

[0002] With the rapid development of the internet and mobile communications, user demands for network services are becoming increasingly diverse and personalized. Whether it's online video streaming, online gaming, file downloading, or everyday web browsing, users' demands for network resources vary greatly. This diverse demand requires service providers to continuously optimize network architecture and improve service quality to meet users' specific needs in different scenarios, thereby enhancing user experience and satisfaction.

[0003] However, existing resource allocation methods often rely on static configuration and predetermined rules, which have obvious limitations. For example, in an intelligent computing cluster, some users may only use a single GPU for computing, with very low network requirements; while other users may use multiple GPUs on multiple servers to perform distributed computing tasks. These users have higher demands on network resources (such as bandwidth, latency, etc.). If a fixed method is used to allocate network-related hardware and software resources, it may lead to a huge waste of resources. At the same time, users who really need good network performance will not be able to obtain the optimal network environment, which in turn affects their operating experience and computing efficiency.

[0004] Therefore, a network resource optimization solution based on user-side computing power monitoring is needed. Summary of the Invention

[0005] In response to the shortcomings of the prior art, the present application provides a network resource optimization method and system based on user-side computing power monitoring, which collects user computing resource usage data through a lightweight plug-in installed on the user device, and transmits the time queue of the collected data to the central server. In the central server, a data analysis and encoding method based on artificial intelligence is used to perform full-connection embedded coding of the usage data of each user's computing resource, thereby automatically identifying the user activity type based on the dynamic propagation representation of the temporal characteristics between the fully connected embedded features of each computing resource usage, and extracting recommended network resources to assign to the user device. In this way, by collecting the computing resource usage data of the user device in real time, the user's computing power requirements can be accurately identified. At the same time, it can respond to changes in user activities in real time and dynamically adjust resource allocation according to the type of user activity, thereby avoiding resource waste and realizing intelligent management and optimized allocation of user-side computing power resources.

[0006] According to one aspect of the present application, a network resource optimization method based on user-side computing power monitoring is provided, which includes: installing a lightweight plug-in on a user device, the lightweight plug-in being used to collect usage data of user computing resources; the lightweight plug-in transmitting the collected time queue of the user computing resource usage data to a central server; on the central server, performing full-connection embedded coding on the usage data of each user computing resource in the time queue of the user computing resource usage data to obtain a time queue of fully-connected embedded features of computing resource usage; performing dynamic propagation of computing resource features on the time queue of the fully-connected embedded features of computing resource usage to obtain a temporal coding feature of a computing resource occupancy pattern, including: extracting a fully-connected embedded feature of current computing resource usage from the time queue of the fully-connected embedded features of computing resource usage; using the fully-connected embedded feature of current computing resource usage as a feature prompt gate, performing gated dynamic propagation on the time queue of the fully-connected embedded features of computing resource usage to obtain a temporal coding feature of the computing resource occupancy pattern; obtaining an identification result based on the temporal coding feature of the computing resource occupancy pattern, and extracting recommended network resources based on the identification result; and assigning the recommended network resources to the user device.

[0007] In the above-mentioned network resource optimization method based on user-side computing power monitoring, the usage data of the user computing resources includes CPU utilization, memory utilization, storage space occupancy ratio, network bandwidth occupancy value and CPU load value.

[0008] In the above-mentioned network resource optimization method based on user-side computing power monitoring, the usage data of each user computing resource in the time queue of the user computing resource usage data is fully connected embedded encoded to obtain a time queue of fully connected embedded features of computing resource usage, including: using a computing resource usage embedding encoder based on a fully connected layer to perform fully connected embedded encoding on the usage data of each user computing resource in the time queue of the user computing resource usage data to obtain a time queue of fully connected embedded coding vectors of computing resource usage as the time queue of fully connected embedded features of computing resource usage.

[0009] In the above-mentioned network resource optimization method based on user-side computing power monitoring, the fully connected embedding feature of the current computing resource usage is extracted from the time queue of the fully connected embedding feature of the computing resource usage, including: extracting the fully connected embedding coding vector of the current computing resource usage from the time queue of the fully connected embedding coding vector of the computing resource usage as the fully connected embedding feature of the current computing resource usage.

[0010] In the above-mentioned network resource optimization method based on user-side computing power monitoring, the fully connected embedded feature of the current computing resource usage is used as a feature prompt gate, and the time queue of the fully connected embedded feature of the computing resource usage is gated and dynamically propagated to obtain the temporal coding feature of the computing resource occupancy mode, including: using the fully connected embedded coding vector of the current computing resource usage as a feature prompt gate, calculating the feature association strength prompt value of each computing resource usage fully connected embedded coding vector in the time queue of the fully connected embedded coding vector of the computing resource usage relative to the fully connected embedded coding vector of the current computing resource usage to obtain the sequence of the computing resource usage feature association strength prompt value column; extract the median of the sequence of the computing resource usage feature association strength prompt values ​​as the computing resource usage feature association strength gate threshold; use the computing resource usage feature association strength gate threshold as the switch threshold of the gating unit, input the time queue of the computing resource usage fully connected embedded coding vector and the sequence of the computing resource usage feature association strength prompt values ​​into the gating unit to obtain a sequence of modulated computing resource usage fully connected embedded coding vectors; perform feature dynamic propagation on the sequence of the modulated computing resource usage fully connected embedded coding vectors to obtain a computing resource occupancy pattern temporal coding feature vector as the computing resource occupancy pattern temporal coding feature.

[0011] In the above-mentioned network resource optimization method based on user-side computing power monitoring, the fully connected embedded coding vector of the current computing resource usage is used as a feature prompt gate, and the feature association strength prompt value of each computing resource usage fully connected embedded coding vector in the time queue of the computing resource usage fully connected embedded coding vector relative to the current computing resource usage fully connected embedded coding vector is calculated to obtain a sequence of computing resource usage feature association strength prompt values, including: calculating the computing resource usage fully connected embedded coding vector and the current computing resource usage fully connected embedded coding vector by dividing the position point by the computing resource usage. association strength vector; calculating the logarithmic function value with base two of the absolute value of each eigenvalue of the computing resource usage association strength vector to obtain the computing resource usage association strength logarithmic vector; calculating the computing resource usage fully connected embedded coding vector and the computing resource usage association strength logarithmic vector to perform positional point multiplication to obtain a point product vector, and calculating the positional point of each eigenvalue in the point product vector to obtain the computing resource usage association strength value; calculating the natural exponential function value with the natural constant e as the base and the computing resource usage association strength value as the exponential power to obtain the computing resource usage feature association strength prompt value.

[0012] In the above-mentioned network resource optimization method based on user-side computing power monitoring, the computing resource usage feature association strength gate threshold is used as the switch threshold of the gating unit, and the time queue of the computing resource usage fully connected embedded coding vector and the sequence of the computing resource usage feature association strength prompt value are input into the gating unit to obtain a sequence of modulated computing resource usage fully connected embedded coding vectors, including: comparing each computing resource usage feature association strength prompt value in the sequence of the computing resource usage feature association strength prompt value with the computing resource usage feature association strength gate threshold to obtain a sequence of weights; comparing the sequence of the weights Each value in the column is used as a weight to perform weighted modulation on the time queue of the computing resource usage fully connected embedded coding vector to obtain a sequence of the modulated computing resource usage fully connected embedded coding vector; wherein, in response to the computing resource usage feature association strength prompt value being greater than or equal to the computing resource usage feature association strength gate threshold, the value obtained by dividing the computing resource usage feature association strength prompt value by the computing resource usage feature association strength gate threshold is used as the weight; in response to the computing resource usage feature association strength prompt value being less than the computing resource usage feature association strength gate threshold, the weight is reset to zero.

[0013] In the above-mentioned network resource optimization method based on user-side computing power monitoring, the sequence of fully connected embedded coding vectors of the modulated computing resource usage is subjected to feature dynamic propagation to obtain the computing resource occupancy pattern time series coding feature vector, including: inputting the sequence of fully connected embedded coding vectors of the modulated computing resource usage into a feature dynamic propagation module based on the forward LTSM model to obtain the computing resource occupancy pattern time series coding feature vector.

[0014] In the above-mentioned network resource optimization method based on user-side computing power monitoring, an identification result is obtained based on the temporal coding characteristics of the computing resource occupancy pattern, and recommended network resources are extracted based on the identification result, including: inputting the temporal coding feature vector of the computing resource occupancy pattern into a classifier-based user activity identifier to obtain the identification result, and the identification result is used to represent the user activity type label; based on the identification result, recommended network resources are extracted from the user activity type-network resource recommendation table.

[0015] According to another aspect of the present application, a network resource optimization system based on user-side computing power monitoring is provided, which includes: a computing resource usage data acquisition module for installing a lightweight plug-in on a user device, wherein the lightweight plug-in is used to collect user computing resource usage data; a computing resource usage data transmission module for the lightweight plug-in to transmit the collected time queue of the user computing resource usage data to a central server; a computing resource usage data embedding coding module for performing full-connection embedding coding on the central server for the usage data of each user computing resource in the time queue of the user computing resource usage data to obtain a time queue of fully-connected embedded features of computing resource usage; a computing resource occupancy pattern timing coding feature extraction module for performing dynamic propagation of computing resource features on the time queue of the fully-connected embedded features of computing resource usage to obtain computing resource usage data; The computing resource occupancy pattern temporal coding feature is extracted, wherein the computing resource occupancy pattern temporal coding feature extraction module includes: a current computing resource usage fully connected embedded feature extraction unit, which is used to extract the current computing resource usage fully connected embedded feature from the time queue of the computing resource usage fully connected embedded feature; a computing resource occupancy pattern temporal coding feature generation unit, which is used to use the current computing resource usage fully connected embedded feature as a feature prompt gate, and perform gated dynamic propagation on the time queue of the computing resource usage fully connected embedded feature to obtain the computing resource occupancy pattern temporal coding feature; a recommended network resource extraction module, which is used to obtain an identification result based on the computing resource occupancy pattern temporal coding feature, and extract recommended network resources based on the identification result; a recommended network resource assignment module, which is used to assign the recommended network resources to the user equipment.

[0016] Due to the adoption of the above technical solutions, the present application has significant technical effects: the network resource optimization method and system based on user-side computing power monitoring provided by the present application collects the usage data of user computing resources through a lightweight plug-in installed on the user device, and transmits the time queue of the collected data to the central server, and adopts artificial intelligence-based data analysis and encoding methods in the central server to perform full-connection embedded coding of the usage data of each user's computing resources, thereby automatically identifying the user activity type based on the dynamic propagation representation of the temporal characteristics between the fully connected embedded features of each computing resource usage, and extracting recommended network resources to assign to the user device. In this way, by collecting the computing resource usage data of the user device in real time, the user's computing power requirements can be accurately identified. At the same time, it can respond to changes in user activities in real time and dynamically adjust resource allocation according to the type of user activity, thereby avoiding resource waste and realizing intelligent management and optimized allocation of user-side computing power resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 This is a flowchart of a network resource optimization method based on user-side computing power monitoring according to an embodiment of the present application.

[0019] Figure 2 Schematic diagram of data flow of a network resource optimization method based on user-side computing power monitoring according to an embodiment of the present application.

[0020] Figure 3 This is a flowchart of dynamically propagating computing resource features on a time queue of fully connected embedded features of computing resource usage to obtain timing coding features of computing resource occupancy patterns according to an embodiment of the present application.

[0021] Figure 4 This is a flowchart of performing gated dynamic propagation on the time queue of the fully connected embedded feature of the computing resource usage to obtain the temporal coding feature of the computing resource occupancy pattern according to an embodiment of the present application.

[0022] Figure 5 A flowchart is provided for obtaining an identification result based on the temporal coding characteristics of the computing resource occupancy pattern according to an embodiment of the present application, and extracting recommended network resources based on the identification result.

[0023] Figure 6 This is a system block diagram of a network resource optimization system based on user-side computing power monitoring according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0025] With the rapid development of the Internet and mobile communications, user demands for network services are becoming increasingly diverse and personalized. Whether it's online video streaming, online gaming, file downloading, or daily web browsing, users' demands for network resources vary significantly. This diversity of demands requires service providers to continuously optimize network architecture and improve service quality to meet users' specific needs in different scenarios, thereby enhancing user experience and satisfaction.

[0026] However, existing resource allocation methods often rely on static configurations and predetermined rules, which present significant limitations. For example, in an intelligent computing cluster, some users utilize a single GPU for computing, with virtually no network requirements, while other users utilize multiple GPUs across multiple servers for distributed computing, requiring certain network resources (bandwidth, latency, etc.). Fixed allocation of network-related hardware and software resources results in wasted resources, while users who require optimal network performance may not receive the best network environment. Consequently, traditional static resource allocation strategies are no longer able to meet the growing demand for personalized services.

[0027] Based on this, the technical concept of this application is to install a lightweight plug-in on the user device to collect the usage data of the user's computing resources, and then transmit the time queue of the collected data to the central server, and use artificial intelligence-based data analysis and encoding methods in the central server to perform full-connection embedded coding of the usage data of each user's computing resources, so as to automatically identify the user activity type based on the dynamic propagation representation of the temporal characteristics between the fully connected embedded features of each computing resource usage, and extract the recommended network resources to assign to the user device. In this way, by collecting the computing resource usage data of the user device in real time, the user's computing power needs can be accurately identified. At the same time, it can respond to changes in user activities in real time and dynamically adjust resource allocation according to the type of user activity, thereby avoiding resource waste and realizing intelligent management and optimized allocation of user-side computing resources.

[0028] Figure 1 This is a flowchart of a network resource optimization method based on user-side computing power monitoring according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the network resource optimization method based on user terminal computing power monitoring according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to an embodiment of the present application, a network resource optimization method based on user-side computing power monitoring includes: S110, installing a lightweight plug-in on a user device, wherein the lightweight plug-in is used to collect usage data of user computing resources; S120, the lightweight plug-in transmits the collected time queue of the user computing resource usage data to a central server; S130, on the central server, performing full-connection embedded coding on the usage data of each user computing resource in the time queue of the user computing resource usage data to obtain a time queue of fully-connected embedded features of computing resource usage; S140, performing dynamic propagation of computing resource features on the time queue of the fully-connected embedded features of computing resource usage to obtain a timing coding feature of a computing resource occupancy pattern; S150, obtaining an identification result based on the timing coding feature of the computing resource occupancy pattern, and extracting recommended network resources based on the identification result; S160, assigning the recommended network resources to the user device.

[0029] In step S110, a lightweight plug-in is installed on the user device. The lightweight plug-in is used to collect data on the user's computing resource usage. It should be understood that lightweight plug-ins are generally small in size and have a fast installation and update process, which users can complete in a short time. Lightweight plug-ins also consume very little system resources during operation and do not significantly affect the performance of the user device. By installing a lightweight plug-in on the user device, the user's computing resource usage can be monitored and collected in real time, so that the user's network resource usage can be analyzed later, thereby realizing dynamic adjustment of network resources.

[0030] In step S120, the lightweight plug-in transmits the collected time queue of the user computing resource usage data to the central server. Specifically, in an embodiment of the present application, the user computing resource usage data includes CPU utilization, memory utilization, storage space occupancy ratio, network bandwidth occupancy value and CPU load value. It should be understood that the CPU is the core processing unit of the device, and the CPU utilization reflects the busyness of the system when processing tasks; the memory utilization reflects the amount of memory occupied by the currently running applications; the storage space occupancy ratio reflects the usage of the device storage; the network bandwidth occupancy value reflects the usage of network resources; and the CPU load value represents the current workload of the CPU. By collecting these specific indicators, the performance and resource usage of the user device can be fully understood, and these data are also an important basis for evaluating the user activity type and the required network resources. Accordingly, considering that the user's computing resource usage is dynamically changing, in order to capture the changing trend of the user's computing resource usage, a lightweight plug-in is used to collect the time queue of the user's computing resource usage data. Then, considering that the central server has powerful computing power and storage resources and is able to process and analyze large amounts of data from user devices, the collected time queue data is transmitted to the central server to achieve centralized processing and analysis of the data, thereby improving processing efficiency and accuracy.

[0031] In step S130, the central server performs fully connected embedding coding on the usage data of each user computing resource in the time queue of the usage data of the user computing resources to obtain a time queue of fully connected embedding features of the computing resource usage. Specifically, in an embodiment of the present application, performing fully connected embedding coding on the usage data of each user computing resource in the time queue of the usage data of the user computing resources to obtain a time queue of fully connected embedding features of the computing resource usage includes: using a computing resource usage embedding encoder based on a fully connected layer to perform fully connected embedding coding on the usage data of each user computing resource in the time queue of the usage data of the user computing resources to obtain a time queue of fully connected embedding coding vectors of computing resource usage as the time queue of the fully connected embedding features of the computing resource usage. It should be understood that the usage data of each user computing resource in the time queue of the user computing resource usage data contains a series of information about the status of the computing resources of the user device, including key information and patterns about the user activity type. Therefore, in the technical solution of the present application, a computing resource usage embedding encoder based on a fully connected layer is used to perform fully connected embedding encoding on the usage data of each user computing resource in the time queue of the user computing resource usage data to capture and mine the complex relationships and patterns between the captured data and retain the most important information, thereby obtaining a time queue of fully connected embedded coding vectors of computing resource usage.

[0032] In step S140, the computing resource usage fully connected embedded feature time queue is dynamically propagated to obtain the computing resource occupancy pattern temporal coding feature. Specifically, Figure 3 This is a flow chart of dynamically propagating computing resource features to obtain timing coding features of computing resource occupancy patterns based on the time queue of fully connected embedded features of computing resource usage according to an embodiment of the present application. Figure 3 As shown, the time queue of the fully connected embedded feature of the computing resource usage is dynamically propagated to obtain the temporal coding feature of the computing resource occupancy pattern, including: S141, extracting the current computing resource usage fully connected embedded feature from the time queue of the fully connected embedded feature of the computing resource usage; S142, using the current computing resource usage fully connected embedded feature as a feature prompt gate, and performing gated dynamic propagation on the time queue of the fully connected embedded feature of the computing resource usage to obtain the temporal coding feature of the computing resource occupancy pattern.

[0033] It should be understood that there are different degrees of correlation between resource usage at different time points in the time queue of the fully connected embedded coding vector of the computing resource usage, and the dependencies between different time spans have different impacts. Based on this, in order to effectively capture the long-term and short-term dependencies of computing resource usage in time series data, learn the association between resource usage patterns at different time points, and thus better understand the user's usage habits and behavior patterns, in the technical solution of the present application, the computing resource characteristics of the time queue of the fully connected embedded coding vector of the computing resource usage are dynamically propagated to obtain the computing resource occupancy pattern temporal coding feature vector as the computing resource occupancy pattern temporal coding feature.

[0034] Specifically, first, the fully connected embedded coding vector of the current computing resource usage is extracted from the time queue of the fully connected embedded coding vectors of the computing resource usage. It can be understood that in the time queue of the fully connected embedded coding vectors of the computing resource usage, each fully connected embedded coding vector of the computing resource usage has different attributes or information, and these vectors describe the usage information represented by themselves. Extracting the fully connected embedded coding vector of the current computing resource usage (current node) means determining a reference point and using this node as a benchmark to measure the importance of other nodes. This is similar to selecting a center point in a feature sequence and then forming a focus range based on this center point to better identify which nodes or time points have a significant impact on the current resource usage pattern.

[0035] Then, the extracted fully connected embedding coding vector of the current computing resource usage is used as the feature prompt gate, and the feature association strength prompt between the fully connected embedding coding vectors of other computing resource usage and the fully connected embedding coding vector of the current computing resource usage is calculated to evaluate and quantify the importance and relevance of the data at each time point to the current node, and obtain a sequence of feature association strength prompt values ​​of computing resource usage, so that the subsequent model can dynamically adjust the attention to other data according to the status of the current node, laying the foundation for assigning dynamic weights to the fully connected embedding coding vectors of computing resource usage at each subsequent time point.

[0036] Next, the median of the sequence of computing resource usage feature correlation strength prompt values ​​is extracted as the threshold for the computing resource usage feature correlation strength gate. Using the median as the threshold ensures that the model is sensitive to feature correlation strengths at most time points, capturing the majority of highly correlated features. Furthermore, compared to the mean, the median is less affected by extreme values ​​(i.e., outliers), thus avoiding misjudgments caused by individual outliers and better reflecting the central tendency of the data.

[0037] Subsequently, the time queue of the fully connected embedded coding vector of computing resource usage and the time queue of the corresponding fully connected embedded coding vector of computing resource usage are input into the gating unit, and the computing resource usage feature correlation strength gate threshold is used as the switching threshold to achieve modulation of each fully connected embedded coding vector of computing resource usage. In other words, by setting the switching threshold of the gating unit to the median of the feature correlation strength, it is possible to effectively filter out features that are highly relevant to the current node. In other words, only data at time points with feature correlation strength greater than or equal to the median is retained, thereby reducing interference from irrelevant or weakly correlated features. In this way, the focus of attention on data from other nodes can be dynamically adjusted, providing data support for subsequent feature propagation.

[0038] Finally, the fully connected embedded coding vectors of the modulated computing resource usage are input into the feature dynamic propagation module based on the forward LTSM model to use the LTSM model to capture and extract the long-term dependency relationship between the node features between each time point, so as to effectively capture the temporal characteristics of the user resource usage pattern and obtain the temporal coding feature vector of the computing resource occupancy pattern.

[0039] Specifically, in an embodiment of the present application, the fully connected embedding feature of the current computing resource usage is extracted from the time queue of the fully connected embedding feature of the computing resource usage, including: extracting the fully connected embedding coding vector of the current computing resource usage from the time queue of the fully connected embedding coding vector of the computing resource usage as the fully connected embedding feature of the current computing resource usage.

[0040] Specifically, Figure 4 This is a flow chart of gated dynamic propagation of the time queue of the fully connected embedded feature of the computing resource usage to obtain the timing coding feature of the computing resource occupancy pattern according to an embodiment of the present application. Figure 4As shown, the fully connected embedding feature of the current computing resource usage is used as a feature prompt gate, and the time queue of the fully connected embedding feature of the computing resource usage is gated and dynamically propagated to obtain the temporal coding feature of the computing resource occupancy mode, including: S1421, using the fully connected embedding coding vector of the current computing resource usage as a feature prompt gate, calculating the feature association strength prompt value of each computing resource usage fully connected embedding coding vector in the time queue of the fully connected embedding coding vector of the computing resource usage relative to the fully connected embedding coding vector of the current computing resource usage to obtain a sequence of computing resource usage feature association strength prompt values; S1422, extracting the computing resource The median of the sequence of usage feature association strength prompt values ​​is used as the computing resource usage feature association strength gate threshold; S1423, the computing resource usage feature association strength gate threshold is used as the switching threshold of the gating unit, and the time queue of the computing resource usage fully connected embedded coding vector and the sequence of the computing resource usage feature association strength prompt values ​​are input into the gating unit to obtain the sequence of modulated computing resource usage fully connected embedded coding vectors; S1424, the sequence of modulated computing resource usage fully connected embedded coding vectors is dynamically propagated to obtain the computing resource occupancy pattern temporal coding feature vector as the computing resource occupancy pattern temporal coding feature.

[0041] More specifically, in an embodiment of the present application, the fully connected embedded coding vector of the current computing resource usage is used as a feature prompt gate, and the feature association strength prompt value of each computing resource usage fully connected embedded coding vector in the time queue of the computing resource usage fully connected embedded coding vector relative to the current computing resource usage fully connected embedded coding vector is calculated to obtain a sequence of computing resource usage feature association strength prompt values, including: calculating the computing resource usage fully connected embedded coding vector and the current computing resource usage fully connected embedded coding vector by dividing the position point by the computing resource usage association strength vector. Quantity; calculate the logarithmic function value with base two of the absolute value of each eigenvalue of the computing resource usage association strength vector to obtain the computing resource usage association strength logarithmic vector; calculate the computing resource usage fully connected embedded coding vector and the computing resource usage association strength logarithmic vector to perform position-by-position point multiplication to obtain a point product vector, and calculate the position-by-position point of each eigenvalue in the point product vector to obtain the computing resource usage association strength value; calculate the natural exponential function value with the natural constant e as the base and the computing resource usage association strength value as the exponential power to obtain the computing resource usage feature association strength prompt value.

[0042] More specifically, in an embodiment of the present application, the computing resource usage characteristic association strength gate threshold is used as the switch threshold of the gating unit, and the time queue of the computing resource usage fully connected embedded coding vector and the sequence of the computing resource usage characteristic association strength prompt value are input into the gating unit to obtain a sequence of modulated computing resource usage fully connected embedded coding vectors, including: comparing each computing resource usage characteristic association strength prompt value in the sequence of the computing resource usage characteristic association strength prompt value with the computing resource usage characteristic association strength gate threshold to obtain a sequence of weights; comparing each computing resource usage characteristic association strength prompt value in the sequence of the computing resource usage characteristic association strength prompt value with the computing resource usage characteristic association strength gate threshold to obtain a sequence of weights; comparing each computing resource usage characteristic association strength prompt value in the sequence of the weights The time queue of the computing resource usage fully connected embedded coding vector is weightedly modulated using the value as a weight to obtain a sequence of the modulated computing resource usage fully connected embedded coding vector; wherein, in response to the computing resource usage feature association strength prompt value being greater than or equal to the computing resource usage feature association strength gate threshold, the value obtained by dividing the computing resource usage feature association strength prompt value by the computing resource usage feature association strength gate threshold is used as the weight; in response to the computing resource usage feature association strength prompt value being less than the computing resource usage feature association strength gate threshold, the weight is reset to zero.

[0043] More specifically, in an embodiment of the present application, the sequence of fully connected embedded coding vectors of the post-modulation computing resource usage is subjected to feature dynamic propagation to obtain a temporal coding feature vector of the computing resource occupancy pattern, including: inputting the sequence of fully connected embedded coding vectors of the post-modulation computing resource usage into a feature dynamic propagation module based on a forward LTSM model to obtain the temporal coding feature vector of the computing resource occupancy pattern.

[0044] In the embodiment of the present application, specifically, the computing resource characteristics are dynamically propagated for the time queue of the fully connected embedded coding vector of the computing resource usage, which can be expressed as follows: .

[0045] in, is the time queue of the fully connected embedding encoding vector for the computing resource usage, They are the first, second, and third time queues of the fully connected embedded coding vectors of the computing resource usage. and The fully connected embedding encoding vector of computing resource usage, It is the The computing resource usage is fully connected and embedded in the feature value of each position in the encoding vector, It is The computing resource usage is fully connected and embedded in the feature value of each position in the encoding vector, represents the logarithmic function value with base 2, is the length of the fully connected embedding encoding vector for each computing resource usage, represents the exponential function value with the natural constant e as the base, yes The corresponding computing resource usage feature correlation strength prompt value, are the first, second and third in the sequence of the computing resource usage feature association strength prompt values. The characteristic correlation strength prompt value of computing resource usage, To get the median of a sequence, is the threshold value of the correlation strength of the computing resource usage feature, For gated mask operation, for Post-modulation computational resource usage after weighted modulation Fully connected embedding coding vector, They are the first, second and third sequences of fully connected embedded coding vectors after modulation respectively. The computational resource usage after modulation is fully connected and embedded into the coding vector, is the sequence of fully connected embedded coding vectors after the modulation, Encode for LTSM, A time-series encoding feature vector is provided for the computing resource occupancy pattern.

[0046] In step S150, based on the temporal coding characteristics of the computing resource occupancy pattern, an identification result is obtained, and based on the identification result, recommended network resources are extracted. Specifically, Figure 5 The flowchart of extracting recommended network resources based on the recognition result is obtained based on the temporal coding characteristics of the computing resource occupancy mode according to the embodiment of the present application. Figure 5 As shown, based on the temporal coding features of the computing resource occupancy pattern, an identification result is obtained, and based on the identification result, recommended network resources are extracted, including: S151, inputting the temporal coding feature vector of the computing resource occupancy pattern into a classifier-based user activity identifier to obtain the identification result, and the identification result is used to represent the user activity type label; S152, based on the identification result, extracting recommended network resources from the user activity type-network resource recommendation table.

[0047] In step S151, the computing resource usage pattern temporal coding feature vector is input into a classifier-based user activity identifier to obtain the recognition result, which is used to represent the user activity type label. Specifically, the computing resource usage pattern temporal coding features obtained by dynamically propagating the time queue of the fully connected embedded coding vector of the computing resource usage are classified to automatically identify the user activity type. User activity type labels can include a variety of different activity categories, common activity types include browsing the web, playing videos, online gaming, downloading / uploading files, editing documents, and system maintenance. In particular, in a specific implementation method of the embodiment of the present application, the computing resource occupancy pattern temporal coding feature vector is input into a classifier-based user activity identifier to obtain the recognition result, and the recognition result is used to represent the user activity type label, including: using the fully connected layer of the classifier to fully connect the computing resource occupancy pattern temporal coding feature vector to obtain a computing resource occupancy pattern temporal fully connected coding feature vector; inputting the computing resource occupancy pattern temporal fully connected coding feature vector into the Softmax classification function of the classifier to obtain the probability value of the computing resource occupancy pattern temporal coding feature vector belonging to each user activity type label; and determining the user activity type label corresponding to the largest of the probability values ​​as the recognition result.

[0048] In step S152, based on the identification result, recommended network resources are extracted from the user activity type-network resource recommendation table. It should be understood that the user activity type-network resource recommendation table is a key component that associates different user activity types with corresponding network resource configurations. Specifically, the horizontal axis of the recommendation table represents the various types of activities performed by users, such as video playback, online games, file downloads, etc.; the vertical axis of the recommendation table represents a set of recommended network resource configurations provided for each activity type, including the required bandwidth size, network latency requirements, network resource allocation priority, etc. Based on the identified user activity type, the system can dynamically extract the most appropriate network resource configuration from the predefined user activity type-network resource recommendation table to ensure that users can obtain the best performance experience when performing specific activities.

[0049] In step S160, the recommended network resources are assigned to the user device. That is, the system dynamically adjusts network bandwidth, computing resources, etc. according to the recommended resource configuration to ensure that the user can obtain the best performance when performing the current activity.

[0050] In particular, when each fully connected embedded coding vector of computing resource usage in the time queue of the fully connected embedded coding vector of computing resource usage respectively represents the fully connected embedded features of the user computing resource usage data in the local time domain, when the computing resource features are dynamically propagated, the sparsity of the local time domain prompts corresponding to the fully connected embedded feature representations in each local time domain relative to the current time series node will lead to the lack of dynamic propagation between local time domains, that is, the full time domain dynamic propagation representation of the computing resource occupancy pattern time series coding feature vector has a simple repetition based on the local time domain fully connected representation, which affects its temporal semantic logical dependency and reduces the accuracy of the classification results.

[0051] Therefore, preferably, when the computing resource occupancy pattern temporal coding feature vector is input into a classifier-based user activity identifier to obtain a recognition result, the computing resource occupancy pattern temporal coding feature vector is optimized, specifically including: determining a computing resource occupancy pattern temporal coding association response matrix and a computing resource occupancy pattern temporal coding distance response matrix based on the square root of the correlation value of the eigenvalue of the computing resource occupancy pattern temporal coding feature vector and the L2 distance value: wherein the process formula can be expressed as: .

[0052] in, A temporal encoding feature vector for the computing resource occupancy pattern, represents the number of eigenvalues ​​in the vector, The first character in the temporal coding feature vector of the computing resource occupancy mode is The eigenvalues ​​at the positions, The first character in the temporal coding feature vector of the computing resource occupancy mode is The eigenvalues ​​at the positions, The computing resource occupation mode temporal coding association response matrix The eigenvalues ​​at the positions, The computing resource occupation mode time sequence coding distance response matrix The eigenvalues ​​at each position.

[0053] The computing resource occupancy mode timing coding feature vector is matrix-multiplied with the computing resource occupancy mode timing coding association response matrix to obtain the computing resource occupancy mode timing coding association response vector; the computing resource occupancy mode timing coding distance response matrix is ​​matrix-multiplied with the transposed vector of the computing resource occupancy mode timing coding feature vector to obtain the computing resource occupancy mode timing coding distance response vector; the computing resource occupancy mode timing coding association response matrix and the computing resource occupancy mode timing coding distance response matrix are dot-multiplied, and then the matrix-multiplied with the transposed vector of the computing resource occupancy mode timing coding feature vector to obtain the computing resource occupancy mode timing coding bias vector; the computing resource occupancy mode timing coding association response vector, the computing resource occupancy mode timing coding distance response vector and the computing resource occupancy mode timing coding bias vector are dot-multiplied to obtain the optimized computing resource occupancy mode timing coding representation vector.

[0054] Here, the optimized computing resource occupancy mode temporal coding representation vector is expressed as: .

[0055] in, A temporal encoding feature vector for the computing resource occupancy pattern, represents the number of eigenvalues ​​in the vector to which it belongs, represents matrix multiplication, A time-series encoding correlation response matrix for the computing resource occupancy mode, A time-series encoding distance response matrix for the computing resource occupancy mode, is the computing resource occupancy mode temporal coding associated response vector, express The transposed vector of The computing resource occupation mode time sequence encoding distance response vector, Indicates point multiplication by position, is the timing coding bias vector of the computing resource occupancy mode, Indicates adding by position point, and Respectively and The transposed vector of It is the temporal coding representation vector of the optimized computing resource occupancy pattern.

[0056] That is, by using the computing resource occupancy pattern temporal coding feature vector as its statistical reference-free distribution response framework based on the auto-correlation matrix and the self-distance matrix, and constructing a retrieval-enhanced reverse response for the computing resource occupancy pattern temporal coding feature vector to avoid simple repetition of feature distribution, and by ensuring the intrinsic mapping logic of the computing resource occupancy pattern temporal coding feature vector based on the retrieval-response context to avoid its surface combination mapping, so as to achieve the semantic logic dependency mapping of the computing resource occupancy pattern temporal coding feature vector while maintaining the intuitive response correlation, thereby improving the accuracy of the recognition result obtained by the user activity identifier based on the classifier when the computing resource occupancy pattern temporal coding feature vector is input.

[0057] In summary, a network resource optimization method based on user-side computing power monitoring based on an embodiment of the present application is illustrated, which collects user computing resource usage data through a lightweight plug-in installed on the user device, and transmits the time queue of the collected data to the central server, and uses artificial intelligence-based data analysis and encoding methods in the central server to perform full-connection embedded coding of the usage data of each user's computing resources, thereby automatically identifying the user activity type based on the dynamic propagation representation of the temporal features between the fully-connected embedded features of each computing resource usage, and extracting recommended network resources to assign to the user device. In this way, by collecting the computing resource usage data of the user device in real time, the user's computing power requirements can be accurately identified. At the same time, it can respond to changes in user activities in real time and dynamically adjust resource allocation according to the type of user activity, thereby avoiding resource waste and realizing intelligent management and optimized allocation of user-side computing power resources.

[0058] Figure 6 : is a system block diagram of a network resource optimization system based on user-side computing power monitoring according to an embodiment of the present application. Figure 6As shown, according to the embodiment of the present application, the network resource optimization system 100 based on user-side computing power monitoring includes: a computing resource usage data acquisition module 110, which is used to install a lightweight plug-in on the user device, and the lightweight plug-in is used to collect the usage data of the user computing resources; a computing resource usage data transmission module 120, which is used for the lightweight plug-in to transmit the collected time queue of the user computing resource usage data to the central server; a computing resource usage data embedding coding module 130, which is used to perform full-connection embedding coding on the usage data of each user computing resource in the time queue of the user computing resource usage data on the central server to obtain a time queue of fully connected embedded features of computing resource usage; a computing resource occupancy pattern timing coding feature extraction module 140, which is used to perform computing resource feature dynamic propagation on the time queue of the fully connected embedded features of the computing resource usage to obtain to the temporal coding features of the computing resource occupancy pattern; a recommended network resource extraction module 150 is used to obtain an identification result based on the temporal coding features of the computing resource occupancy pattern, and extract recommended network resources based on the identification result; a recommended network resource assignment module 160 is used to assign the recommended network resources to the user equipment; wherein, the temporal coding feature extraction module of the computing resource occupancy pattern includes: a current computing resource usage fully connected embedded feature extraction unit, which is used to extract the current computing resource usage fully connected embedded feature from the time queue of the computing resource usage fully connected embedded feature; a computing resource occupancy pattern temporal coding feature generation unit, which is used to use the current computing resource usage fully connected embedded feature as a feature prompt gate, and perform gated dynamic propagation on the time queue of the computing resource usage fully connected embedded feature to obtain the temporal coding features of the computing resource occupancy pattern.

[0059] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the above-mentioned network resource optimization system 100 based on user-side computing power monitoring have been described in detail above. Figures 1 to 5 The network resource optimization method based on user-side computing power monitoring has been introduced in detail, and therefore, its repeated description will be omitted.

[0060] In summary, the network resource optimization system 100 based on user-side computing power monitoring according to the embodiment of the present application is explained. It collects the usage data of user computing resources through a lightweight plug-in installed on the user device, and transmits the time queue of the collected data to the central server. In the central server, a data analysis and encoding method based on artificial intelligence is used to perform full-connection embedded coding of the usage data of each user computing resource. In this way, the user activity type is automatically identified based on the dynamic propagation representation of the temporal features between the fully connected embedded features of each computing resource usage, and recommended network resources are extracted and assigned to the user device. In this way, resource allocation can be dynamically adjusted according to the user activity type, and changes in user activities can be responded to in real time to ensure that users obtain the best performance experience when performing specific activities.

[0061] The above description is merely a preferred embodiment of the present application and does not constitute any form of limitation to the present application. Although the present application has been disclosed as a preferred embodiment, it is not intended to limit the present application. Any technician familiar with the present profession can make slight changes or modifications to equivalent embodiments of the technical contents disclosed above without departing from the scope of the technical solution of the present application. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present application without departing from the content of the technical solution of the present application are still within the scope of the technical solution of the present application.

Claims

1. A network resource optimization method based on user-side computing power monitoring, characterized in that: include: A lightweight plug-in is installed on the user device, and the lightweight plug-in is used to collect the usage data of the user computing resources; the lightweight plug-in transmits the collected time queue of the usage data of the user computing resources to the central server; on the central server, the usage data of each user computing resource in the time queue of the usage data of the user computing resources is fully connected and embedded to obtain a time queue of fully connected embedded features of computing resource usage; The computing resource feature is dynamically propagated on the time queue of the fully connected embedded feature of the computing resource usage to obtain the computing resource occupancy pattern timing coding feature, including: extracting the current computing resource usage fully connected embedded feature from the time queue of the computing resource usage fully connected embedded feature; using the current computing resource usage fully connected embedded feature as a feature prompt gate, performing gated dynamic propagation on the time queue of the computing resource usage fully connected embedded feature to obtain the computing resource occupancy pattern timing coding feature; obtaining an identification result based on the computing resource occupancy pattern timing coding feature, and extracting recommended network resources based on the identification result; and assigning the recommended network resources to the user device.

2. The network resource optimization method based on user-side computing power monitoring according to claim 1 is characterized in that: The usage data of the user computing resources includes CPU utilization, memory utilization, storage space occupancy ratio, network bandwidth occupancy value and CPU load value.

3. The network resource optimization method based on user-side computing power monitoring according to claim 2 is characterized in that: The usage data of each user computing resource in the time queue of the usage data of the user computing resources is fully connected and embedded to obtain a time queue of fully connected embedded features of the computing resource usage, including: using a computing resource usage embedding encoder based on a fully connected layer to fully connected embed the usage data of each user computing resource in the time queue of the usage data of the user computing resources to obtain a time queue of fully connected embedded coding vectors of computing resource usage as the time queue of fully connected embedded features of the computing resource usage.

4. The network resource optimization method based on user-side computing power monitoring according to claim 3 is characterized in that: Extracting the fully connected embedding feature of current computing resource usage from the time queue of the fully connected embedding feature of computing resource usage, including: extracting the fully connected embedding coding vector of current computing resource usage from the time queue of the fully connected embedding coding vector of computing resource usage as the fully connected embedding feature of current computing resource usage.

5. The network resource optimization method based on user-side computing power monitoring according to claim 4 is characterized in that: The method uses the fully connected embedded feature of the current computing resource usage as a feature prompt gate, and performs gated dynamic propagation on the time queue of the fully connected embedded feature of the computing resource usage to obtain the temporal coding feature of the computing resource occupancy mode, including: using the fully connected embedded coding vector of the current computing resource usage as a feature prompt gate, calculating the feature association strength prompt value of each computing resource usage fully connected embedded coding vector in the time queue of the fully connected embedded coding vector of the computing resource usage relative to the fully connected embedded coding vector of the current computing resource usage to obtain a sequence of computing resource usage feature association strength prompt values; extracting the computing resource usage The median of the sequence of usage feature association strength prompt values ​​is used as the computing resource usage feature association strength gate threshold; the computing resource usage feature association strength gate threshold is used as the switch threshold of the gating unit, the time queue of the computing resource usage fully connected embedded coding vector and the sequence of the computing resource usage feature association strength prompt values ​​are input into the gating unit to obtain the sequence of modulated computing resource usage fully connected embedded coding vectors; the sequence of modulated computing resource usage fully connected embedded coding vectors is dynamically propagated to obtain the computing resource occupancy pattern temporal coding feature vector as the computing resource occupancy pattern temporal coding feature.

6. The network resource optimization method based on user-side computing power monitoring according to claim 5 is characterized in that: The fully connected embedded coding vector of the current computing resource usage is used as a feature prompt gate, and the feature association strength prompt value of each computing resource usage fully connected embedded coding vector in the time queue of the computing resource usage fully connected embedded coding vector relative to the current computing resource usage fully connected embedded coding vector is calculated to obtain a sequence of computing resource usage feature association strength prompt values, including: calculating the position point division of the computing resource usage fully connected embedded coding vector and the current computing resource usage fully connected embedded coding vector to obtain a computing resource usage association strength vector; calculating the logarithmic function value with base two of the absolute value of each eigenvalue of the computing resource usage association strength vector to obtain a computing resource usage association strength logarithmic vector; calculating the position point multiplication of the computing resource usage fully connected embedded coding vector and the computing resource usage association strength logarithmic vector to obtain a point product vector, and calculating the position point of each eigenvalue in the point product vector to obtain a computing resource usage association strength value; calculating the natural exponential function value with the natural constant e as the base and the computing resource usage association strength value as an exponential power to obtain the computing resource usage feature association strength prompt value.

7. The network resource optimization method based on user-side computing power monitoring according to claim 6 is characterized in that: The computing resource usage characteristic association strength gate threshold is used as the switch threshold of the gating unit, and the time queue of the computing resource usage fully connected embedded coding vector and the sequence of the computing resource usage characteristic association strength prompt value are input into the gating unit to obtain the sequence of the modulated computing resource usage fully connected embedded coding vector, including: comparing each computing resource usage characteristic association strength prompt value in the sequence of the computing resource usage characteristic association strength prompt value with the computing resource usage characteristic association strength gate threshold to obtain a sequence of weights; using each value in the sequence of weights as a weight to perform weighted modulation on the time queue of the computing resource usage fully connected embedded coding vector to obtain the sequence of the modulated computing resource usage fully connected embedded coding vector; wherein, in response to the computing resource usage characteristic association strength prompt value being greater than or equal to the computing resource usage characteristic association strength gate threshold, the value obtained by dividing the computing resource usage characteristic association strength prompt value with the computing resource usage characteristic association strength gate threshold is used as the weight; in response to the computing resource usage characteristic association strength prompt value being less than the computing resource usage characteristic association strength gate threshold, the weight is reset to zero.

8. The network resource optimization method based on user-side computing power monitoring according to claim 7 is characterized in that: The sequence of fully connected embedded coding vectors of the post-modulation computing resource usage is subjected to feature dynamic propagation to obtain a temporal coding feature vector of the computing resource occupancy pattern, including: inputting the sequence of fully connected embedded coding vectors of the post-modulation computing resource usage into a feature dynamic propagation module based on a forward LTSM model to obtain the temporal coding feature vector of the computing resource occupancy pattern.

9. The network resource optimization method based on user-side computing power monitoring according to claim 8 is characterized in that: Based on the temporal coding features of the computing resource occupancy pattern, a recognition result is obtained, and based on the recognition result, recommended network resources are extracted, including: inputting the temporal coding feature vector of the computing resource occupancy pattern into a classifier-based user activity identifier to obtain the recognition result, wherein the recognition result is used to represent a user activity type label; based on the recognition result, recommended network resources are extracted from a user activity type-network resource recommendation table.

10. A network resource optimization system based on user-side computing power monitoring, characterized in that: include: A computing resource usage data collection module, used to install a lightweight plug-in on a user device, wherein the lightweight plug-in is used to collect usage data of the user's computing resources; A computing resource usage data transmission module is used for the lightweight plug-in to transmit the collected time queue of the user computing resource usage data to the central server; a computing resource usage data embedding coding module is used for, on the central server, performing full-connection embedding coding on the usage data of each user computing resource in the time queue of the user computing resource usage data to obtain a time queue of fully connected embedded features of computing resource usage; a computing resource occupancy mode timing coding feature extraction module is used for dynamically propagating computing resource features on the time queue of fully connected embedded features of computing resource usage to obtain computing resource occupancy mode timing coding features, wherein the computing resource occupancy mode timing coding feature extraction module includes: current computing A resource usage fully connected embedded feature extraction unit is used to extract the current computing resource usage fully connected embedded feature from the time queue of the computing resource usage fully connected embedded feature; a computing resource occupancy pattern temporal coding feature generation unit is used to use the current computing resource usage fully connected embedded feature as a feature prompt gate, and perform gated dynamic propagation on the time queue of the computing resource usage fully connected embedded feature to obtain the computing resource occupancy pattern temporal coding feature; a recommended network resource extraction module is used to obtain an identification result based on the computing resource occupancy pattern temporal coding feature, and extract recommended network resources based on the identification result; a recommended network resource assignment module is used to assign the recommended network resources to the user device.

Images