An intelligent management method and system for computing and network resources

Abstract

The present invention relates to the technical field of resource management, and specifically to an intelligent management method and system for computing and network resources, including the following steps: Based on the computing power node traffic monitoring data, sample each node traffic one by one, calculate the traffic difference between adjacent nodes, and use time series analysis to analyze the change rate. In the present invention, by sampling the node traffic and combining time series to extract spatio-temporal feature values, the multi-dimensional understanding of node traffic is enhanced, the processing accuracy is improved, the fine dependence weights are obtained by using the node dependence relationship matrix and the multi-head dependence matrix, the description of the relationship between nodes is enhanced, the time and space features are fused in the traffic association prediction, ensuring the accuracy of the prediction. In resource scheduling, reasonable marking and priority allocation are carried out for future over-limit nodes through recursive calculation and load evaluation, making the allocation more dynamic. The ratio adjustment of core and edge nodes realizes the optimal resource allocation through priority accumulation, improving the system stability and utilization rate.

Description

Technical Field

[0001] The present invention relates to the technical field of resource management, and particularly to an intelligent management method and system for computing and network resources. Background Art

[0002] The technical field of resource management involves the effective allocation, scheduling, and optimization of various resources in a computer system to improve the overall performance and resource utilization rate of the system. These resources can include processors, memories, network bandwidths, storage spaces, etc., and are widely used in various systems such as computer operating systems, cloud computing, and distributed computing environments. The goal of resource management is to ensure that system resources are fully utilized, meet the requirements of various tasks, while optimizing performance, saving costs, and avoiding resource waste. Through intelligent resource management methods, the system can achieve automatic adjustment in complex and changing operating environments, thereby achieving efficient and flexible resource allocation.

[0003] Among them, the intelligent management method for computing and network resources refers to a technical means for intelligent scheduling and management between computing power and network resources, aiming to improve the overall service performance and efficiency of computing and network through optimizing the collaborative allocation of computing power and network resources. This method is often applied in fields such as cloud computing and edge computing to achieve dynamic and intelligent management of various computing and communication resources, ensuring the effective provision of computing and communication services in complex network environments and meeting the needs of different users and tasks.

[0004] The prior art lacks in-depth extraction of the spatio-temporal characteristics of node traffic in resource management. Usually, only simple analysis is carried out, which is difficult to comprehensively describe the dynamic characteristics of traffic, resulting in insufficient accuracy of the correlation prediction of node traffic and affecting the rationality of resource scheduling. The processing of node dependencies is relatively single and cannot fully reflect multi-dimensional dependencies, resulting in inaccurate resource scheduling in scenarios with complex node associations, which may cause resource waste or unbalanced allocation. The load assessment in the prior art is mainly based on the current state and lacks the ability to predict future loads, resulting in untimely responses and lagging scheduling of the system when facing load fluctuations. The resource allocation between core and edge nodes lacks a flexible adjustment mechanism, resulting in unreasonable resource scheduling in complex environments, with bottleneck nodes overloaded or resources idle, reducing the overall operating efficiency and resource utilization rate of the system. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art and propose an intelligent management method and system for computing and network resources.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: An intelligent management method for computing and network resources, comprising the following steps:

[0007] S1: Based on the traffic monitoring data of computing power nodes, sample the node traffic one by one, calculate the traffic difference between adjacent nodes, use time series analysis to analyze the change rate, merge the spatial feature values with the time features, and extract the traffic spatio-temporal feature values;

[0008] S2: Based on the traffic spatio-temporal feature values, select the node dependency relationship matrix, adopt the node spatial position and traffic sequence, obtain the weights through matrix operations, establish a multi-head dependency matrix using the head mean value, and obtain the node dependency weight values;

[0009] S3: Based on the node dependency weight values, fuse the spatial and time features, call the module to merge the feature values, accumulate the node features, perform weighted fusion to match the traffic relationship between nodes, establish an association matrix, and obtain the node traffic association prediction values;

[0010] S4: Based on the node traffic association prediction values, select the node traffic time period, perform recursive calculation for multiple future time periods, accumulate the intervals to evaluate the load status, set thresholds to mark over-limit nodes, and allocate resource weights according to the priorities to obtain the node resource scheduling plan;

[0011] S5: Based on the node resource scheduling plan, analyze the load status of multiple nodes, adjust the resource ratio between the core and edge nodes, accumulate the core node priorities using the priority parameters, re-allocate resources according to the ratio, and obtain the dynamic resource allocation results.

[0012] The traffic spatio-temporal feature values include traffic spatial features, traffic time features, and adjacent node traffic difference features. The node dependency weight values include node-to-node dependency relationship weights, node spatial position weights, and traffic sequence weights. The node traffic association prediction values include node association matrices, time feature fusion values, and traffic relationship matching values. The node resource scheduling plan includes over-limit node markings, resource weight allocations, and priority sorting schemes. The dynamic resource allocation results include core node resource allocations, edge node resource allocations, and priority parameter adjustments.

[0013] As a further solution of the present invention, the steps of extracting the traffic spatio-temporal feature values are specifically as follows:

[0014] S111: Based on the traffic monitoring data of computing power nodes, perform time series sampling on multiple nodes, record the traffic data of each node at each moment, and generate a node traffic time series matrix;

[0015] S112: Using the node traffic time series matrix, calculate the traffic difference between any two adjacent nodes, and extract the traffic difference feature values between adjacent nodes from the differences to obtain a node traffic difference feature matrix;

[0016] S113: Combine the node traffic difference feature matrix with the change rate obtained through time series analysis and use the composite formula:

[0017] ;

[0018] Calculate the traffic spatio-temporal value of each node at the differential moment, integrate the data of all time points, and extract the traffic spatio-temporal feature value;

[0019] Among them, represents the traffic spatio-temporal feature value, represents the total number of nodes, represents the traffic value of node i at time t, represents node at time the traffic value, that is, the traffic of the previous moment, represents the time interval adjustment coefficient of node i, represents the spatial weight coefficient of node i, represents the traffic adjustment variable of node i.

[0020] As a further solution of the present invention, the specific steps for obtaining the node dependence weight value are as follows:

[0021] S211: According to the traffic spatio-temporal feature value, select the node dependence relationship matrix, combine the node spatial position and traffic sequence data, adjust and verify the dependence relationship matrix, and generate an adjusted node dependence relationship matrix;

[0022] S212: Starting from the adjusted node dependence relationship matrix, through matrix operations, integrate the node spatial position data and traffic sequence, calculate the preliminary weight value of the mutual dependence between multiple nodes, and generate a preliminary dependence weight matrix between nodes;

[0023] S213: Combine the preliminary dependence weight matrix between nodes, use the average traffic feature between nodes, and use the formula:

[0024] ;

[0025] Integrate the weights between multiple nodes and calculate the node dependence weight value;

[0026] Among them, represents the dependence weight between node i and node j, represents the position parameter of node i and k, represents the traffic coefficient of node k and j, represents the initial dependence weight between node i and j, represents the adjustment coefficient of node i, represents the adjustment coefficient of node j, Represents the total number of intermediate nodes for averaging.

[0027] As a further aspect of the present invention, the steps for obtaining the predicted value of node traffic association are specifically as follows:

[0028] S311: Based on the node dependence weight value, call the spatial and temporal feature matrices, combine the spatial and temporal features of multiple nodes according to the weights, integrate the spatial position and temporal feature values, and obtain the fused node feature matrix;

[0029] S312: Use the fused node feature matrix to perform an accumulation process on the node features, calculate the weighted sum of the multiple node features based on the node dependence weight, and at the same time perform an inter-node matching analysis on the accumulation result of the node features to generate a weighted fused node feature value matrix;

[0030] S313: Through the weighted fused node feature value matrix, combine the traffic relationships of multiple nodes, perform a matching process on the traffic features between nodes, and use the formula:

[0031] ;

[0032] Calculate and generate the predicted value of node traffic association;

[0033] Wherein, represents the predicted value of traffic association between node i and node j, represents the spatial feature coefficient between node i and k, represents the temporal feature weight between node k and j, represents the initial traffic matching relationship between node i and j, represents the feature matching weight between node i and j, represents the dependence adjustment coefficient of node i, which is used to characterize the characteristic correction of node i in traffic dependence, represents the dependence adjustment coefficient of node j, which is used to characterize the characteristic correction of node j in traffic dependence, represents the total number of features between nodes, which is used to normalize the entire calculation result.

[0034] As a further aspect of the present invention, the steps for obtaining the node resource scheduling plan are specifically as follows:

[0035] S411: Based on the predicted value of node traffic association, select the traffic feature values of each node in the future time period, perform multi-time period recursive calculation for each node, and gradually accumulate the traffic features in multiple time periods to obtain the accumulated traffic value of the node in the future time period;

[0036] S412: Use the cumulative flow value of the nodes in the future time period described in the section to set the threshold range of the node load, and judge the load status of each node in turn. Mark all the over-limit nodes that exceed the set threshold to obtain the node load over-limit mark;

[0037] S413: Based on the node load over-limit mark, allocate resource weights according to the load priority of the nodes, calculate the resource scheduling amounts of multiple nodes, and use the formula:

[0038] ;

[0039] Allocate the resource scheduling amounts of the nodes to obtain the node resource scheduling plan;

[0040] Among them, represents the resource scheduling amounts of node i and node j, represents the number of nodes participating in the resource scheduling calculation or the number of types of load priority factors, represents the load priority of node i, represents the load influence factor between node p and node j, represents the resource demand relationship between node i and node j, represents the initial resource matching value between node i and node j, represents the load adjustment coefficient between node i and node j.

[0041] As a further solution of the present invention, the steps for obtaining the dynamic resource allocation result are specifically as follows:

[0042] S511: Based on the node resource scheduling plan, call the load status of multiple nodes, analyze the load data of the core nodes and the edge nodes, and at the same time calculate the resource ratio of the core nodes and the edge nodes. Make a preliminary ratio of the resources of the core and edge nodes to obtain the resource ratio result of the core and edge nodes;

[0043] S512: Use the resource ratio result of the core and edge nodes to calculate the priority parameters of the core nodes, and calculate the priority cumulative values of the multi-core nodes in a weighted cumulative manner;

[0044] S513: Based on the priority cumulative value of the core nodes, re-adjust the resource ratio of the core and edge nodes, and combine the load requirements of the core nodes and the edge nodes, and use the formula:

[0045] ;

[0046] Calculate and generate the dynamic resource allocation result;

[0047] Among them, represents the dynamic resource allocation value between node i and node j, represents the total number of nodes, represents the resource allocation ratio between nodes i and k, represents the load demand value of nodes k and j, represents the load correlation parameter between nodes i and j, represents the priority adjustment coefficient of node k, represents the load adjustment parameter between nodes i and j.

[0048] An intelligent management system for computing and network resources, which is used to execute the above-mentioned intelligent management method for computing and network resources. The system includes:

[0049] The traffic feature extraction module samples the traffic of multiple nodes one by one based on the computing power node traffic monitoring data, calculates the traffic difference between nodes, combines the time series to analyze the change rate, extracts the spatial and temporal features, and obtains the traffic spatio-temporal feature values;

[0050] The node dependence weight calculation module selects the node dependence relationship matrix based on the traffic spatio-temporal feature values, calculates the weights by multiplying and accumulating the spatial position and the traffic sequence, and establishes a multi-head dependence matrix by taking the average value between the heads to obtain the node dependence weight values;

[0051] The association prediction matrix establishment module fuses the spatial and temporal features based on the node dependence weight values, calls the module to accumulate the node features, and performs weighted fusion to match the traffic relationship between nodes, establishes an association matrix, and obtains the node traffic association prediction values;

[0052] The dynamic resource scheduling module recursively calculates the traffic in multiple future time periods based on the node traffic association prediction values, accumulates and evaluates the load status, marks the over-limit nodes, and reallocates the resource weights according to the priorities to obtain the dynamic resource allocation results.

[0053] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0054] In the present invention, by sampling the node traffic and combining the time series to extract the spatio-temporal feature values, the multi-dimensional understanding of the node traffic is enhanced, and the processing accuracy is improved. The fine dependence weights are obtained by using the node dependence relationship matrix and the multi-head dependence matrix, enhancing the description of the relationship between nodes. The time and space features are fused in the traffic association prediction to ensure the accuracy of the prediction. In the resource scheduling, the future over-limit nodes are reasonably marked and priority allocated through recursive calculation and load evaluation, making the allocation more dynamic. The ratio adjustment of the core and edge nodes realizes the optimal resource allocation through priority accumulation, improving the system stability and utilization rate. Description of the Drawings

[0055] Figure 1 is a schematic diagram of the working process of the present invention;

[0056] Figure 2 Flow chart of steps for extracting traffic spatio - temporal eigenvalue of the present invention;

[0057] Figure 3 Flow chart of steps for obtaining node dependency weight value of the present invention;

[0058] Figure 4 Flow chart of steps for obtaining node traffic correlation prediction value of the present invention;

[0059] Figure 5 Flow chart of steps for obtaining node resource scheduling plan of the present invention;

[0060] Figure 6 Flow chart of steps for obtaining dynamic resource allocation result of the present invention. Detailed implementation manners

[0061] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0062] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more unless otherwise specifically defined.

[0063] Embodiment 1: Please refer to Figure 1 , the present invention provides a technical solution: An intelligent management method for computing and network resources, including the following steps:

[0064] S1: Based on the computing power node traffic monitoring data, sample each node traffic one by one, calculate the traffic difference between adjacent nodes, use time - series analysis to analyze the change rate, combine the spatial eigenvalue with the time eigenvalue, and extract the traffic spatio - temporal eigenvalue;

[0065] S2: Based on the traffic spatio - temporal eigenvalue, select the node dependency relationship matrix, adopt the node spatial position and traffic sequence, obtain the weight through matrix operation, establish a multi - head dependency matrix using the head - to - head average value, and obtain the node dependency weight value;

[0066] S3: Based on the node dependency weight values, fuse the spatial and temporal features, call the module to merge the eigenvalue, accumulate the node features, weighted-fuse to match the traffic relationship between nodes, establish an association matrix, and obtain the predicted value of node traffic association;

[0067] S4: Based on the predicted value of node traffic association, select the node traffic time period, perform recursive calculation for multiple future time periods, accumulate and evaluate the load status in the interval, set a threshold to mark the over-limit nodes, and allocate resource weights according to the priority to obtain the node resource scheduling plan;

[0068] S5: Based on the node resource scheduling plan, analyze the load status of multiple nodes, adjust the resource ratio of the core and edge nodes, use the priority parameter to accumulate the priority of the core nodes, and re-allocate resources according to the ratio to obtain the dynamic resource allocation result.

[0069] The traffic spatio-temporal eigenvalue includes traffic spatial feature, traffic time feature, and adjacent node traffic difference feature. The node dependency weight value includes the dependency relationship weight between nodes, the node spatial position weight, and the traffic sequence weight. The predicted value of node traffic association includes the node association matrix, the time feature fusion value, and the traffic relationship matching value. The node resource scheduling plan includes the over-limit node mark, resource weight allocation, and priority sorting scheme. The dynamic resource allocation result includes the core node resource allocation, the edge node resource allocation, and the priority parameter adjustment.

[0070] Please refer to Figure 2 , and the steps for extracting the traffic spatio-temporal eigenvalue are specifically as follows:

[0071] S111: Based on the traffic monitoring data of computing power nodes, perform time series sampling on multiple nodes, record the traffic data of each node at each moment, and generate a node traffic time series matrix;

[0072] On the basis of monitoring the traffic of computing power nodes, during the process of time series sampling, the traffic data of each node at each moment is recorded. These data sets constitute the node traffic time series matrix. This process ensures the continuity and comparability of each node's traffic, thus allowing for the precise measurement of traffic changes in subsequent analysis. The creation of the node traffic time series matrix is achieved by measuring devices capturing traffic data per second and transmitting it through a data interface to a centralized processing system for integration and synchronization, ensuring real-time update and accuracy of the data, and obtaining the node traffic time series matrix.

[0073] S112: Using the node traffic time series matrix, calculate the traffic difference between any two adjacent nodes, extract the traffic difference eigenvalue between adjacent nodes from the difference, and obtain the node traffic difference feature matrix;

[0074] Based on the obtained node traffic time series matrix, the traffic difference eigenvalue is obtained by calculating the traffic difference between any adjacent nodes. This process not only involves simple subtraction operations numerically, but also uses advanced data processing techniques to identify and analyze those nodes with abnormal traffic changes. This data processing process includes smoothing of time series data, outlier detection, and trend analysis, so as to ensure the accuracy and reliability of the traffic difference eigenvalue and obtain the node traffic difference feature matrix.

[0075] S113: The node traffic difference feature matrix is combined with the change rate obtained through time series analysis, and the composite formula is used:

[0076] ;

[0077] Calculate the traffic spatio-temporal value of each node at the differential moment, integrate the data of all time points, and extract the traffic spatio-temporal eigenvalue;

[0078] Among them, represents the traffic spatio-temporal eigenvalue, represents the total number of nodes, represents the traffic value of node i at time t, represents node at time traffic value, that is, the traffic of the previous moment, represents the time interval adjustment coefficient of node i, represents the spatial weight coefficient of node i, represents the traffic adjustment variable of node i.

[0079] Formula:

[0080] ;

[0081] The advantage of the formula is that it not only considers the combination of time and space factors, but also introduces the square sum and square root operations, enhances the sensitivity to outliers, and increases the adaptability to different node characteristics through the adjustment coefficient

[0082] Detailed explanation of the formula and the derivation process of formula calculation:

[0083] Set the actual data as follows: , First, calculate the square of the time difference:

[0084] ;

[0085] Calculate the product of the square root of the denominator and the weight:

[0086] ​

[0087] Calculated :

[0088] ;

[0089] This result indicates that for Node 1, between time t1 and t2, the increase in traffic has a relatively high spatio-temporal eigenvalue, which indicates that the traffic has increased significantly during this time period. For the step result, it means that Node 1 needs to be monitored closely at this time point.

[0090] Please refer to Figure 3 , the steps for obtaining the node dependence weight value are specifically as follows:

[0091] S211: According to the traffic spatio-temporal eigenvalue, select the node dependence relationship matrix, combine the node spatial position and traffic sequence data, adjust and verify the dependence relationship matrix, and generate an adjusted node dependence relationship matrix;

[0092] Based on the selection of the traffic spatio-temporal eigenvalue and the adjustment of the node dependence relationship matrix, by integrating the node spatial position and traffic sequence data, the dependence relationship between nodes can be effectively adjusted and confirmed. This process first involves the extraction of node spatial position data, which comes from the actual geographical location information of node deployment, and the traffic sequence data comes from previous traffic monitoring records. After these data are processed by time series analysis, a time series graph of the traffic pattern is generated for each node. Then, these data are input into the dependence relationship matrix, and through matrix operation analysis, the mutual influence between nodes is determined, and the relationship strength and dependence degree between nodes are determined, so as to generate an adjusted node dependence relationship matrix. This matrix can provide a scientific basis for the further optimization and management of the network.

[0093] S212: Starting from the adjusted node dependence relationship matrix, through matrix operation, integrate the node spatial position data and traffic sequence, calculate the preliminary weight value of the mutual dependence between multiple nodes, and generate a preliminary dependence weight matrix between nodes;

[0094] Starting from the adjusted node dependence relationship matrix, through matrix operation to integrate the node spatial position data and traffic sequence. In this step, the integration of node spatial position data is mainly based on the spatial analysis technology of the geographical information system to determine the physical distance and connectivity between nodes, and the traffic sequence data is obtained through previous data analysis. The key lies in how to calculate the mutual dependence weight between nodes through these data. This not only involves simple data integration, but also requires the application of complex matrix operation formulas. Through these formulas, the dependence weight between each node can be calculated. This weight matrix helps to further understand the interaction dynamics of nodes in the network and provides data support for the optimization and adjustment of the network.

[0095] S213: Combine the preliminary dependence weight matrix between nodes, use the average traffic characteristics between nodes, and adopt the formula:

[0096] ;

[0097] Integrate the weights between multiple nodes and calculate the node dependence weight value;

[0098] Among them, represents the dependence weight between node i and node j, represents the position parameter of node i and k, represents the traffic coefficient of node k and j, represents the initial dependence weight between node i and j, represents the adjustment coefficient of node i, represents the adjustment coefficient of node j, represents the total number of intermediate nodes for averaging.

[0099] Formula:

[0100] ;

[0101] The benefit of the formula is that by introducing the product of the position and traffic coefficient between nodes, after weighted averaging and then standardizing through the square root summation of the node adjustment coefficients, it increases the accuracy of the calculation and the rationality of the node weight distribution.

[0102] Detailed explanation of the formula and the derivation process of the formula calculation:

[0103] Suppose there are 3 nodes, and the position parameter between nodes is the reciprocal of the distance. Suppose , the node traffic coefficient Suppose it is the traffic difference. Suppose , the initial dependence weight Based on the previous traffic analysis, suppose , the node adjustment coefficients and Suppose they are respectively :

[0104] ;

[0105] The result shows that the dependence weight value between node 1 and node 2 is -1.132, indicating that under the given parameter settings, the dependence of node 1 on node 2 is relatively low, and it may be necessary to re-evaluate the traffic configuration or physical connectivity between nodes to improve the overall efficiency of the network.

[0106] Please refer to Figure 4, the steps for obtaining the predicted value of node traffic association are specifically as follows:

[0107] S311: Based on the node dependence weight value, call the spatial and temporal feature matrices, combine the spatial and temporal features of multiple nodes according to the weights, integrate the spatial position and temporal feature values, and obtain the fused node feature matrix;

[0108] Based on the comprehensive data of the node dependence weight value and the spatio-temporal features, an in-depth analysis of the node feature integration is initiated, which involves the precise determination of the spatial position and the synchronous processing of the temporal data. By aggregating the feature data of multiple nodes, the system can provide a comprehensive view showing the performance of each node under different temporal and spatial conditions. In this process, the data of each node are accumulated and adjusted according to their weights, ensuring the integrity and representativeness of the data, and forming a matrix containing the comprehensive features of all nodes, which provides the basic data for subsequent analysis.

[0109] S312: Use the fused node feature matrix to perform an accumulation process on the node features. Based on the node dependence weights, calculate the weighted sum of the multi-node features, and at the same time perform an inter-node matching analysis on the accumulated results of the node features to generate a matrix of weighted fused node feature values;

[0110] Using the comprehensive feature matrix, further analyze the mutual dependence between nodes through weighted calculation. In this step, the system first matches the features of each node with the features of other nodes to analyze their mutual correlation degree. The relationship between nodes is strengthened through matrix operations, and the features between each pair of nodes are weighted, ensuring that the features of each node not only represent its own state but also reflect the interaction with other nodes. This weighted processing helps to accurately identify the potential connections between nodes and lays the foundation for constructing a more advanced network model.

[0111] S313: Through the matrix of weighted fused node feature values, combine the traffic relationships of multiple nodes, perform a matching process on the traffic features between nodes, and use the formula:

[0112] ;

[0113] Calculate and generate the predicted value of node traffic association;

[0114] Where, represents the predicted value of traffic association between node i and node j, represents the spatial feature coefficient between node i and k, represents the temporal feature weight between node k and j, represents the initial traffic matching relationship between node i and j, represents the feature matching weight between node i and j, Represents the dependency adjustment coefficient of node i, which is used to characterize the characteristic correction of node i in traffic dependency. Represents the dependency adjustment coefficient of node j, which is used to characterize the characteristic correction of node j in traffic dependency. Represents the total number of features between nodes, which is used to normalize the entire calculation result.

[0115] Formula:

[0116] ;

[0117] The benefit of the formula is that it allows for the precise calculation of the traffic association prediction value between nodes by combining specific node characteristics, time series data, and the dependency weights between nodes, improving the accuracy of prediction and the efficiency of network traffic management.

[0118] Detailed explanation of the formula and the derivation process of formula calculation:

[0119] Consider a simplified network that includes three nodes, and the spatial characteristic coefficients of each node are 1.2, 0.8, and 1.5 respectively, and the time characteristic weights are 0.5, 1.1, and 0.7 respectively, the initial traffic matching relationship is 1.0, 1.3, and 0.9, the feature matching weights are 0.9 and 1.1, and the node dependency adjustment coefficients are 1.0 and 1.2 respectively. The specific calculation process is as follows:

[0120] ;

[0121] The result shows that the traffic association prediction value between node 1 and node 2 is 0.645, indicating that the two nodes have a medium degree of association after considering spatial and time factors. This prediction result will help network administrators adjust resource allocation and traffic management strategies to optimize network performance.

[0122] Please refer to Figure 5 , and the specific steps for obtaining the node resource scheduling plan are as follows:

[0123] S411: Based on the node traffic association prediction value, select the traffic feature values of each node in the future time period, perform multi-time period recursive calculation for each node, and gradually accumulate the traffic features within multiple time periods to obtain the accumulated traffic value of the node in the future time period;

[0124] When formulating a node resource scheduling plan based on the predicted value of node traffic association, it is first necessary to evaluate the traffic characteristic value of each node in the future time period. This step requires the use of time series analysis methods to predict the possible traffic changes of each node. By performing regression analysis on past traffic data, the patterns and trends of traffic changes are obtained to predict future traffic. Then, weight allocation is carried out in combination with the operational importance and traffic carrying capacity of each node to ensure that important nodes can obtain necessary resource support first when traffic surges, thereby avoiding service interruptions caused by insufficient resources. This analysis process not only needs to consider the data of individual nodes but also the mutual influence among nodes in the entire network, and complex network analysis and optimization algorithms are used to ensure the reasonable allocation of resources in the network.

[0125] S412: Use the cumulative traffic value of nodes in the future time period to set the threshold range of node load, and judge the load status of each node in turn, mark all over-limit nodes that exceed the set threshold, and obtain the node load over-limit mark;

[0126] In the calculation of the cumulative traffic value of nodes in the future time period, the real-time traffic data of nodes is accumulated, and it is judged whether a node is overloaded according to the set threshold, and overloaded nodes are marked. This requires collecting the traffic data of each node at different time points and analyzing the data through mathematical modeling methods. For example, methods such as moving average or exponential smoothing are used to smooth the previous traffic data to reduce the influence of outliers and ensure the accuracy and reliability of traffic data. Then, by comparing the total traffic in different time periods with the threshold, nodes at risk are identified to provide a basis for subsequent resource scheduling. This step not only requires data processing technology but also an in-depth understanding of the characteristics of network operation and the dependency relationship among nodes.

[0127] S413: Based on the node load over-limit mark, allocate resource weights according to the load priority of nodes, calculate the resource scheduling volume of multiple nodes, and use the formula:

[0128] ;

[0129] Allocate the resource scheduling volume of nodes to obtain the node resource scheduling plan;

[0130] Among them, represents the resource scheduling volume between node i and node j, represents the number of nodes participating in resource scheduling calculation or the number of types of load priority factors, represents the load priority of node i, represents the load influence factor between node p and node j, represents the resource demand relationship between node i and node j, represents the initial resource matching value between node i and node j, Represents the load adjustment coefficient between node i and node j.

[0131] Formula:

[0132] ;

[0133] The advantage of the formula is that, by comprehensively considering the load differences between nodes, the resource demand relationship, and the load adjustment coefficient, it allows the system to dynamically adjust resource allocation when facing different node demands, enhancing the system's response ability to traffic fluctuations and resource utilization efficiency.

[0134] Detailed explanation of the formula and the derivation process of formula calculation:

[0135] Set the load value between node i and node j as units, represents the load impact factor, units represents the resource demand relationship, units represents the initial resource matching, is the load adjustment coefficient. Substitute into the formula:

[0136] ;

[0137] ;

[0138] ;

[0139] ;

[0140] ;

[0141] The result shows that the resource scheduling volume between node i and node j is 147.35 units, which reflects how resources between nodes are calculated and allocated under the given node priorities, load impact factors, and resource demand relationships, thus supporting key support for nodes with high traffic.

[0142] Please refer to Figure 6 for the specific steps to obtain the dynamic resource allocation result:

[0143] S511: Based on the node resource scheduling plan, call the load status of multiple nodes, analyze the load data of core nodes and edge nodes, and at the same time calculate the resource ratio of core nodes and edge nodes, and make a preliminary ratio of the resources of core and edge nodes to obtain the resource ratio result of core and edge nodes;

[0144] Based on the detailed analysis of the node resource scheduling plan, according to the specific resource allocation strategy, first call the real-time load data of each node. These data are from the monitoring module of the system and are updated in real time to ensure the freshness and accuracy of the data, and analyze the load status of the core nodes and edge nodes. Next, according to the preset resource ratio model, combined with the operation efficiency and response time data of each node, calculate a preliminary resource ratio result. This calculation process considers the mutual influence and dependency relationships between nodes and uses a load balancing algorithm to improve the overall operation efficiency of the system.

[0145] S512: Utilize the resource ratio results of the core and edge nodes to calculate the priority parameters of the core nodes. Calculate the priority accumulation value of the core nodes by calculating the priority parameters of multiple core nodes in a weighted accumulation manner.

[0146] By refining the resource ratio results of the core and edge nodes, quantify the priorities of each core node. Adopt a dynamic priority algorithm, which dynamically adjusts the priority parameters according to the task urgency of the node, the previous data processing efficiency, and its position in the network to ensure that critical tasks can be processed first. The accumulated core node priority is expressed as a numerical index, which will directly affect the resource allocation decision of the node.

[0147] S513: Based on the priority accumulation value of the core nodes, readjust the resource ratio of the core and edge nodes. Combine the load requirements of the core nodes and edge nodes and use the formula:

[0148] ;

[0149] Calculate and generate the dynamic resource allocation result;

[0150] Among them, represents the dynamic resource allocation value between node i and node j, represents the total number of nodes, represents the resource ratio value between node i and k, represents the load demand value between node k and j, represents the load correlation parameter between node i and j, represents the priority adjustment coefficient of node k, represents the load adjustment parameter between node i and j.

[0151] Formula:

[0152] ;

[0153] The advantage of the formula is that it comprehensively considers the differences between nodes and priority adjustments, and optimizes the overall network performance by balancing the resource allocation between nodes.

[0154] Detailed Explanation of Formulas and Derivation Process of Formula Calculation:

[0155] First, set the resource ratio between node i and node k as , the load requirements between node k and node j as , the load correlation parameter between node i and j as , the priority adjustment coefficient of node k as , and the load adjustment parameter between node i and j as ; Assume: , , , , , calculate:

[0156] ;

[0157] This result indicates that the dynamic resource allocation value between node i and node j is 2, indicating that the resource configuration of these two nodes is relatively balanced and can effectively support task processing and data exchange between nodes.

[0158] An intelligent management system for computing and network resources, which is used to execute the above-mentioned intelligent management method for computing and network resources. The system includes:

[0159] The traffic feature extraction module samples the traffic of multiple nodes one by one based on the computing power node traffic monitoring data, calculates the traffic difference between nodes, combines the time series to analyze the change rate, extracts spatial and temporal features, and obtains the traffic spatio-temporal feature values;

[0160] The node dependence weight calculation module selects the node dependence relationship matrix based on the traffic spatio-temporal feature values, calculates the weights by multiplying and accumulating the spatial position and the traffic sequence, and establishes a multi-head dependence matrix by taking the average value between the heads to obtain the node dependence weight values;

[0161] The correlation prediction matrix establishment module fuses spatial and temporal features based on the node dependence weight values, calls the module to accumulate node features, performs weighted fusion to match the traffic relationship between nodes, establishes a correlation matrix, and obtains the node traffic correlation prediction values;

[0162] The dynamic resource scheduling module recursively calculates the traffic in multiple future time periods based on the node traffic correlation prediction values, accumulatively evaluates the load status, marks the over-limit nodes, and reallocates the resource weights according to the priority to obtain the dynamic resource allocation results.

[0163] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for intelligent management of computing network resources, characterized in that: The following steps are involved: Based on the traffic monitoring data of computing nodes, the node traffic is sampled one by one, the traffic difference between adjacent nodes is calculated, the change rate is analyzed by time series, the spatial feature value is combined with the time feature, and the spatiotemporal feature value of the traffic is extracted; Based on the spatiotemporal characteristic values ​​of the traffic, a node dependency matrix is ​​selected, the node spatial position and traffic sequence are taken, weights are obtained through matrix operations, a multi-head dependency matrix is ​​established using the average value between heads, and the node dependency weight value is obtained; Based on the node dependency weight value, the spatial and temporal features are integrated, the module is called to merge the feature values, the node features are accumulated, the weighted fusion matching node traffic relationship is performed, the association matrix is ​​established, and the node traffic association prediction value is obtained; Based on the node traffic correlation prediction value, select the node traffic time period, recursively calculate the future multiple time periods, evaluate the load status by interval accumulation, set the threshold to mark the over-limit node, allocate resource weights according to priority, and obtain the node resource scheduling plan; Based on the node resource scheduling plan, the load status of multiple nodes is analyzed, the resource ratio of core and edge nodes is adjusted, the priority parameters are used to accumulate the core node priorities, and the resources are reallocated according to the ratio to obtain the dynamic resource allocation result.

2. The method for intelligent management of computing network resources according to claim 1, characterized in that: The traffic spatiotemporal characteristic values ​​include traffic space characteristics, traffic time characteristics, and adjacent node traffic difference characteristics. The node dependency weight values ​​include node dependency weights, node spatial position weights, and traffic sequence weights. The node traffic association prediction values ​​include node association matrices, time feature fusion values, and traffic relationship matching values. The node resource scheduling plan includes over-limit node marking, resource weight allocation, and priority sorting scheme. The dynamic resource allocation results include core node resource allocation, edge node resource allocation, and priority parameter adjustment.

3. The method for intelligent management of computing network resources according to claim 2, characterized in that: The steps of extracting the spatiotemporal characteristic values ​​of traffic flow are specifically as follows: Based on the computing power node traffic monitoring data, time series sampling is performed on multiple nodes, the traffic data of each node at each moment is recorded, and the node traffic time series matrix is ​​generated; Using the node traffic time series matrix, the traffic difference between any adjacent nodes is calculated, and the traffic difference characteristic values ​​between the adjacent nodes are extracted from the difference to obtain the node traffic difference characteristic matrix; The node traffic difference characteristic matrix is ​​combined with the change rate obtained through time series analysis, and a composite formula is used: ； Calculate the spatiotemporal value of traffic flow at each node at the differentiated moment, integrate the data of all time points, and extract the spatiotemporal characteristic value of traffic flow; in, represents the spatiotemporal characteristic value of traffic, Represents the total number of nodes, represents the flow value of node i at time t, Representative Node At the moment The flow value, that is, the flow at the previous moment, represents the time interval adjustment coefficient of node i, represents the spatial weight coefficient of node i, Represents the traffic adjustment variable of node i.

4. The method for intelligent management of computing network resources according to claim 3, characterized in that: The steps for obtaining the node dependency weight value are specifically as follows: According to the spatiotemporal characteristic values ​​of the traffic flow, a node dependency matrix is ​​selected, and the dependency matrix is ​​adjusted and verified in combination with the spatial position of the node and the traffic sequence data to generate an adjusted node dependency matrix; Starting from the adjusted node dependency matrix, integrating the node spatial position data and the flow sequence through matrix operations, calculating the preliminary weight values ​​of the mutual dependence between multiple nodes, and generating a preliminary dependency weight matrix between nodes; Combined with the preliminary dependency weight matrix between nodes, using the average traffic characteristics between nodes, the formula is adopted: ； Integrate the weights between multiple nodes and calculate the node dependency weight value; in, represents the dependency weight between node i and node j, represents the position parameters of nodes i and k, represents the flow coefficient of nodes k and j, represents the initial dependency weight of nodes i and j, represents the adjustment coefficient of node i, represents the adjustment coefficient of node j, Represents the total number of intermediate nodes used for averaging.

5. The method for intelligent management of computing network resources according to claim 4, characterized in that: The steps for obtaining the node traffic correlation prediction value are specifically as follows: Based on the node dependency weight value, the spatial and temporal feature matrix is ​​called, the spatial and temporal features of multiple nodes are combined according to the weights, and the spatial position and temporal feature values ​​are integrated to obtain a fused node feature matrix; Utilizing the fused node feature matrix, performing accumulation processing on the node features, calculating the weighted sum of multiple node features based on the node dependency weights, and performing node matching analysis on the accumulation results of the node features to generate a weighted fused node feature value matrix; Through the node feature value matrix after weighted fusion, combined with the traffic relationship of multiple nodes, the traffic characteristics between nodes are matched and processed, using the formula: ； Calculate and generate node traffic correlation prediction values; in, represents the traffic correlation prediction value between node i and node j, represents the spatial characteristic coefficient of nodes i and k, represents the time feature weight of nodes k and j, represents the initial traffic matching relationship between nodes i and j, represents the feature matching weight between nodes i and j, Represents the dependency adjustment coefficient of node i, which is used to characterize the characteristic correction of node i in traffic dependency. Represents the dependency adjustment coefficient of node j, which is used to characterize the characteristic correction of node j in traffic dependency. Represents the total number of features between nodes and is used to normalize the entire calculation result.

6. The method for intelligent management of computing network resources according to claim 5, characterized in that: The steps for obtaining the node resource scheduling plan are specifically as follows: Based on the node traffic association prediction value, the traffic characteristic value of each node in the future time period is selected, and multi-time period recursive calculation is performed for each node, and the traffic characteristics in multiple time periods are gradually accumulated to obtain the accumulated traffic value of the node in the future time period; The accumulated flow value of the node in the future time period is used to set the threshold range of the node load, and the load status of each node is judged in turn, and all the nodes exceeding the set threshold are marked to obtain the node load overlimit mark; Based on the node load overlimit mark, resource weights are allocated according to the node load priority, and the resource scheduling amount of multiple nodes is calculated using the formula: ； Allocate node resource scheduling amounts and obtain node resource scheduling plans; in, represents the resource scheduling amount of node i and node j, Represents the number of nodes or types of load priority factors involved in resource scheduling calculations. represents the load priority of node i, represents the load impact factor between node p and node j, represents the resource demand relationship between nodes i and j, represents the initial resource matching value of node i and node j, represents the load adjustment coefficient of node i and node j.

7. The method for intelligent management of computing network resources according to claim 6, characterized in that: The steps for obtaining the dynamic resource allocation result are specifically as follows: Based on the node resource scheduling plan, the load status of multiple nodes is called, the load data of core nodes and edge nodes are analyzed, and the resource ratio of core nodes and edge nodes is calculated at the same time, and the resources of core nodes and edge nodes are preliminarily allocated to obtain the resource ratio result of core nodes and edge nodes; Calculate the priority parameters of the core nodes using the resource allocation result of the core and edge nodes, and calculate the priority parameters of the multi-core nodes in a weighted accumulation manner to obtain the priority accumulation value of the core nodes; Based on the accumulated priority value of the core node, the resource ratio of the core and edge nodes is readjusted, and the load requirements of the core nodes and edge nodes are combined to adopt the formula: ； Calculate and generate dynamic resource allocation results; in, represents the dynamic resource allocation value between node i and node j, Represents the total number of nodes, represents the resource allocation value between nodes i and k, represents the load demand value of nodes k and j, represents the load correlation parameter between nodes i and j, represents the priority adjustment coefficient of node k, represents the load adjustment parameter of nodes i and j.

8. An intelligent management system for computing network resources, characterized in that: According to any one of claims 1 to 7, the computing network resource intelligent management method, the system comprising: The traffic feature extraction module samples the traffic of multiple nodes one by one based on the traffic monitoring data of computing nodes, calculates the traffic difference between nodes, analyzes the change rate in combination with the time series, extracts the spatial and temporal features, and obtains the spatiotemporal feature values ​​of the traffic; The node dependency weight calculation module selects the node dependency matrix based on the spatiotemporal characteristic value of the traffic flow, calculates the weight by multiplying and accumulating the spatial position and the traffic sequence, and establishes a multi-head dependency matrix by taking the average value between heads to obtain the node dependency weight value; The association prediction matrix establishment module is based on the node dependency weight value, integrates the spatial and temporal features, calls the module to accumulate node features, weights and fuses the traffic relationship between matching nodes, establishes the association matrix, and obtains the node traffic association prediction value; The dynamic resource scheduling module recursively calculates the traffic in multiple time periods in the future based on the node traffic association prediction value, cumulatively evaluates the load status, marks the overloaded nodes, reallocates resource weights according to priority, and obtains dynamic resource allocation results.

Images