Network node resource dynamic allocation method and system

Abstract

The invention relates to the technical field of network resource allocation, and discloses a network node resource dynamic allocation method and system. The method comprises the following steps: collecting multi-dimensional resource use data of network nodes and executing preprocessing to obtain a standardized resource monitoring matrix and a topological entropy weight matrix; performing nonlinear topology perception calculation to obtain a resource demand prediction matrix; based on the resource demand prediction matrix, performing sectional progressive allocation decision on the network node resources to obtain a first resource allocation decision matrix; performing chaos disturbance fine tuning on the first resource allocation decision matrix to obtain a second resource allocation decision matrix and a resource locking time table; and performing network node resource multi-level scheduling based on the second resource allocation decision matrix and the resource locking time table, and generating a resource execution result log and performance monitoring data. According to the invention, the prediction accuracy of the real resource demand of the network node is improved, and a differentiated, multi-level and smooth transition resource scheduling execution strategy is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of network resource allocation, and particularly to a method and system for dynamically allocating network node resources. Background Art

[0002] Traditional network node resource allocation methods usually only consider single-dimensional resource monitoring and static threshold control, and cannot adapt to complex and changing network environments. These methods often rely on simple linear prediction models, which separate the network topology structure from resource demand prediction, resulting in low resource allocation efficiency and prone to problems of over-allocation or resource starvation in scenarios where there is a strong resource dependence between nodes and the network topology structure changes dynamically.

[0003] Existing technologies generally adopt "all-or-nothing" or coarse-grained resource allocation strategies in the resource allocation decision-making process, lacking a refined resource segmentation mechanism and a progressive allocation method. This "one-size-fits-all" resource adjustment usually causes system oscillations and affects business continuity. At the same time, the parameters of the resource allocation models in existing technologies are mostly statically preset, lacking the ability of adaptive adjustment, and unable to automatically optimize the resource allocation strategy according to network environment changes and business demand fluctuations, resulting in poor performance in complex network environments. Summary of the Invention

[0004] The present invention provides a method and system for dynamically allocating network node resources, which improves the prediction accuracy of the real resource requirements of network nodes and realizes a differential, multi-level, and smooth-transition resource scheduling execution strategy.

[0005] In a first aspect, the present invention provides a method for dynamically allocating network node resources, and the method for dynamically allocating network node resources includes: Collect multi-dimensional resource usage data of network nodes and perform preprocessing to obtain a standardized resource monitoring matrix and a topological entropy weight matrix; Perform non-linear topology-aware calculation based on the standardized resource monitoring matrix and the topological entropy weight matrix to obtain a resource demand prediction matrix; Based on the resource demand prediction matrix, make a segmented progressive allocation decision on network node resources to obtain a first resource allocation decision matrix; Perform chaotic perturbation fine-tuning on the first resource allocation decision matrix to obtain a second resource allocation decision matrix and a resource locking schedule; Based on the second resource allocation decision matrix and the resource locking schedule, perform multi-level scheduling of network node resources to generate a resource execution result log and performance monitoring data.

[0006] Optionally, in the first implementation manner of the first aspect of the present invention, the multi-dimensional resource usage data of the acquisition network node is used and preprocessed to obtain a standardized resource monitoring matrix and a topological entropy weight matrix, including: Collect the computing resources, memory resources, storage resources, and bandwidth resources of the network node at a fixed frequency to obtain multi-dimensional resource usage data; Perform a standardized transformation on the multi-dimensional resource usage data to obtain a standardized resource monitoring matrix; Construct a network topology relationship graph based on the connection relationship and communication situation between network nodes, and extract node topology correlation data based on the network topology relationship graph; Perform weight calculation and normalization processing on the node topology correlation data to obtain a topological entropy weight matrix.

[0007] Optionally, in the second implementation manner of the first aspect of the present invention, the non-linear topology-aware calculation is performed based on the standardized resource monitoring matrix and the topological entropy weight matrix to obtain a resource demand prediction matrix, including: Construct a three-dimensional resource demand tensor including time series, node identification, and resource dimension based on the standardized resource monitoring matrix; Perform time series decomposition processing on the three-dimensional resource demand tensor to obtain multi-component time series data; Calculate the node resource fluctuation characteristics according to the residual component in the multi-component time series data to obtain a non-linear attenuation coefficient; Perform recursive convolution calculation based on the three-dimensional resource demand tensor and the non-linear attenuation coefficient to obtain an initial resource demand prediction value, and perform topology-aware adjustment on the initial resource demand prediction value in combination with the topological entropy weight matrix to obtain a resource demand prediction matrix.

[0008] Optionally, in the third implementation manner of the first aspect of the present invention, the recursive convolution calculation is performed based on the three-dimensional resource demand tensor and the non-linear attenuation coefficient to obtain an initial resource demand prediction value, and the initial resource demand prediction value is topologically aware adjusted in combination with the topological entropy weight matrix to obtain a resource demand prediction matrix, including: Extract the resource usage records of each network node at multiple historical time points from the three-dimensional resource demand tensor, and assign different weights to the data at different time points according to the non-linear attenuation coefficient to obtain time-weighted resource data; Based on the topological entropy weight matrix, screen out the node set with a topological correlation degree higher than the preset threshold with the target node to obtain the influencing node group of the target node; Accumulate and sum the product of the time-weighted resource data and the topological entropy weight of each network node in the influencing node group, and perform a recursive convolution operation to obtain an initial resource demand prediction value; Calculate the change rate of the topological entropy weight matrix within a continuous time window, and process it in combination with the logarithmic gain function to obtain a dynamically adjusted topological perception perturbation factor; Perform a multiplication operation on the initial resource demand prediction value, the dynamically adjusted topological perception perturbation factor, and the sign of the resource demand change trend to obtain a resource demand prediction matrix.

[0009] Optionally, in the fourth implementation manner of the first aspect of the present invention, the step of performing a segmented progressive allocation decision on the network node resources based on the resource demand prediction matrix to obtain a first resource allocation decision matrix includes: Define a minimum divisible unit of resources for various types of network resources, and calculate the difference between the total current available resources and the total allocated resources to obtain the available resource amount; Perform a weighted calculation on each network node based on the business criticality, topological centrality, and priority to obtain a node importance weight; Divide the resource allocation process into multiple decision segments, and calculate the resource amount to be allocated in each segment based on the number of decision segments and the available resource amount to obtain a decision segment resource allocation plan; Calculate the resource value coefficient and decision score of each network node in each decision segment according to the node importance weight and the resource demand prediction matrix, and allocate resources in descending order of the decision score based on the decision segment resource allocation plan to obtain the resource allocation results of each segment; Calculate a decision reversal threshold according to the number of network nodes, and re-evaluate the previous segment's allocation decision when the continuous decision score deviation of the node exceeds the decision reversal threshold, and finally integrate the allocation results of each decision segment to obtain a first resource allocation decision matrix.

[0010] Optionally, in the fifth implementation manner of the first aspect of the present invention, the step of calculating a decision reversal threshold according to the number of network nodes, re-evaluating the previous segment's allocation decision when the continuous decision score deviation of the node exceeds the decision reversal threshold, and finally integrating the allocation results of each decision segment to obtain a first resource allocation decision matrix includes: Calculate the decision reversal threshold of the current decision segment according to the number of network nodes and the decision segment serial number; Calculate the decision score deviation value between two consecutive decision segments of each network node, and compare the decision score deviation value with the decision reversal threshold to obtain a list of candidate nodes that need to be re-evaluated; Recalculate the resource value coefficient and decision score of the previous segment for the network nodes in the list of candidate nodes that need to be re-evaluated according to the resource demand prediction matrix to obtain a corrected previous segment allocation plan; Calculate the resource locking period of each network node based on the topology-aware perturbation factor and the decision reversal threshold, and obtain a resource allocation stability guarantee strategy; Integrate the resource allocation results of each decision segment according to the network node and resource dimension, and process the resource allocation of the core node according to the resource locking period to obtain a first resource allocation decision matrix.

[0011] Optionally, in the sixth implementation manner of the first aspect of the present invention, the chaotic perturbation fine-tuning of the first resource allocation decision matrix to obtain a second resource allocation decision matrix and a resource locking schedule includes: Import the historical performance data of the network node to construct a performance evaluation function, and calculate the deviation between the current node performance and the expected performance to obtain a node performance feedback closed-loop value; Dynamically adjust the node perturbation coefficient of the chaotic perturbation function according to the magnitude of the node performance feedback closed-loop value; Calculate a resource adjustment factor using the node performance feedback closed-loop value and the node perturbation coefficient, and compare the resource adjustment factor with a preset detachment threshold to determine whether resource adjustment is required, and obtain a resource adjustment trigger signal; Perform a resource ratio adjustment on the first resource allocation decision matrix according to the resource adjustment trigger signal, and introduce upper and lower limit constraints to ensure that the adjustment range is within the target range to obtain a second resource allocation decision matrix; Calculate the resource elastic recovery period of each network node based on the node performance feedback closed-loop value, and generate a corresponding resource locking schedule.

[0012] Optionally, in the seventh implementation manner of the first aspect of the present invention, the calculating a resource adjustment factor using the node performance feedback closed-loop value and the node perturbation coefficient, and comparing the resource adjustment factor with a preset detachment threshold to determine whether resource adjustment is required, and obtaining a resource adjustment trigger signal includes: Perform a division operation on the node performance feedback closed-loop value and the node perturbation coefficient to obtain a performance perturbation ratio, and perform a non-linear transformation on the performance perturbation ratio using the hyperbolic tangent function to obtain a basic detachment value; Multiply the basic detachment value by a preset adjustment intensity parameter to obtain a resource adjustment factor, where the adjustment intensity parameter is a fixed constant; Calculate the absolute value of the resource adjustment factor, and compare the absolute value with a preset detachment threshold. When the absolute value is greater than the preset detachment threshold, generate a positive resource adjustment trigger signal, otherwise generate a negative resource adjustment trigger signal; Determine whether to adjust the first resource allocation decision matrix based on the resource adjustment trigger signal state. When a positive resource adjustment trigger signal is received, perform resource adjustment calculations on the corresponding elements of the resource adjustment factor and the first resource allocation decision matrix. When a negative resource adjustment trigger signal is received, keep the corresponding elements of the first resource allocation decision matrix unchanged.

[0013] Optionally, in the eighth implementation manner of the first aspect of the present invention, the multi-level scheduling of network node resources based on the second resource allocation decision matrix and the resource lock schedule to generate a resource execution result log and performance monitoring data includes: Divide the network nodes into a critical node group, a standard node group, and an elastic node group according to the business importance level, and calculate the difference value between the current actual resource configuration of each network node and the target configuration in the second resource allocation decision matrix; Based on the ratio of the difference value to the current actual resource configuration, and in combination with the critical node group, the standard node group, and the elastic node group, configure a differentiated execution strategy for each network node; Construct a resource dependency graph based on the dependency relationships of different resource dimensions between network nodes, perform topological sorting on the resource dependency graph to obtain the resource scheduling execution order, and perform resource adjustment operations according to the differentiated execution strategy, the resource scheduling execution order, and the resource lock schedule; During the resource adjustment execution process, real-time monitor the performance indicators of the network nodes. When it is detected that the performance degradation exceeds the preset deterioration threshold, abort the current adjustment operation and roll back to the previous stable state, add the corresponding network nodes to the temporary protection list, and at the same time record the entire execution process to generate a resource execution result log and performance monitoring data.

[0014] In a second aspect, the present invention provides a network node resource dynamic allocation system, and the network node resource dynamic allocation system includes: An acquisition module, configured to acquire multi-dimensional resource usage data of network nodes and perform preprocessing to obtain a standardized resource monitoring matrix and a topological entropy weight matrix; A calculation module, configured to perform non-linear topological perception calculations based on the standardized resource monitoring matrix and the topological entropy weight matrix to obtain a resource demand prediction matrix; An allocation decision module, configured to perform a segmented progressive allocation decision on network node resources based on the resource demand prediction matrix to obtain a first resource allocation decision matrix; A fine-tuning module, configured to perform chaotic perturbation fine-tuning on the first resource allocation decision matrix to obtain a second resource allocation decision matrix and a resource lock schedule; A multi-level scheduling module for performing multi-level scheduling of network node resources based on the second resource allocation decision matrix and the resource locking schedule, and generating a resource execution result log and performance monitoring data.

[0015] In the technical solution provided by the present invention, by constructing a multi-dimensional network resource monitoring and acquisition matrix and performing topology-aware initialization, the limitation that resource monitoring and network topology analysis are mutually separated in the prior art is broken, and the integrated analysis of network resource usage and topology structure is realized. The non-linear influence and adaptive adjustment mechanism of the topological relationship are introduced, breaking through the linear limitation of the resource prediction model in the prior art, enabling the prediction model to sensitively perceive network topology changes and adjust the prediction strategy accordingly, and improving the prediction accuracy of the true resource requirements of network nodes. By defining the minimum divisible unit of resources and the multi-segment progressive decision mechanism, the limitations of traditional one-time global optimization or simple greedy algorithms are overcome, avoiding the local optimal trap, realizing the global optimization of resource allocation, and improving the decision reliability through the judgment reversal threshold and the allocation backtracking mechanism. By introducing controllable chaotic behavior and the resource elastic recovery period, when the resource allocation falls into local optimality but with poor performance, it can actively break the balance and explore a better allocation plan. At the same time, the adjustment range is ensured to be within a reasonable range through the upper and lower limits of resource adjustment, improving the system stability. A differential, multi-level, and smooth-transition resource scheduling execution strategy is realized, overcoming the system oscillation problem caused by the "one-size-fits-all" resource adjustment in traditional technologies. The correct order of resource scheduling is ensured through the resource dependency graph and topological sorting, guaranteeing service continuity. A complete closed-loop feedback mechanism is established, realizing the continuous self-optimization of the resource allocation system, breaking through the limitation of traditional static preset parameters, enabling the system to continuously learn and improve the resource allocation strategy, and adapting to the complex and changeable network environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0017] Figure 1 It is a schematic diagram of an embodiment of the method for dynamically allocating network node resources in an embodiment of the present invention; Figure 2 It is a schematic diagram of an embodiment of the system for dynamically allocating network node resources in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] An embodiment of the present invention provides a method and system for dynamically allocating network node resources. Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned drawings of the present invention are used to distinguish similar objects and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the term "comprising" or "having" and any variation thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0019] For ease of understanding, the specific process of the embodiment of the present invention is described below. Please refer to Figure 1 , an embodiment of the method for dynamically allocating network node resources in the embodiment of the present invention includes: Step S101, collect multi-dimensional resource usage data of network nodes and perform preprocessing to obtain a standardized resource monitoring matrix and a topological entropy weight matrix; It can be understood that the execution subject of the present invention can be a network node resource dynamic allocation system, or a terminal or a server, and specific limitations are not made here. An embodiment of the present invention is described by taking the server as the execution subject as an example.

[0020] Specifically, a resource monitoring probe module is deployed at each node in the network architecture. This module collects key resource metrics such as the node's core computing resources, memory usage, storage capacity utilization, and bandwidth utilization in real time at a set fixed frequency. Using a timestamp mechanism, each round of sampled data is periodically sequenced to form a resource sampling sequence. The collected raw multidimensional resource data is normalized. A normalization function is used to compress all sampled values to the range of 0 to 1 to eliminate dimensional differences between resource dimensions. This process compares the maximum and minimum values of each resource dimension within a historical observation window and then performs numerical conversion. A standardized resource monitoring matrix is constructed, indexed by node, resource dimension, and time, forming a three-dimensional data structure. Furthermore, to enable subsequent resource prediction to accurately perceive the potential impact of network structure on the evolution of node resource states, a topological dependency model is established between nodes. While collecting resource data, a network topology graph is constructed based on the physical connections between nodes and historical communication data. In this graph, the node set represents all devices in the network, and the edge set represents the connectivity paths between them. For each edge in the topology graph, the number of hops between two nodes and the cumulative communication volume over a certain period are recorded. Based on this, a weight function is introduced to quantify the strength of direct connections between nodes. This weight function comprehensively considers the attenuation effect of network distance and the amplification effect of communication strength. The parameter setting ensures a natural suppression of weak links over long distances and an emphasis on high-frequency channels over short distances. All edge weights are normalized, and the relative topological influence between each node and its topological neighbors is calculated. This forms a topological entropy weight matrix, which is a multiplication of the number of nodes. Each element in the matrix represents the strength of the topological influence of a node on its neighbors. By expressing the structural relationships between nodes through the normalized weight values, the resulting entropy weight matrix captures the local topological complexity and gives meaning to the network structure.

[0021] Step S102: performing nonlinear topology perception calculation based on the standardized resource monitoring matrix and the topology entropy weight matrix to obtain a resource demand prediction matrix; Specifically, a three-dimensional resource demand tensor is constructed based on the standardized resource monitoring matrix. This tensor uses time series, node identifiers, and resource dimensions as its three axes, representing the usage trends and change characteristics of each network node on various resources within a fixed time window. This tensor can reflect the dynamic evolution process of resource usage and form a spatio-temporal structured data representation. Perform multi-component decomposition on each time series in the three-dimensional resource demand tensor, parsing the resource usage data into multiple time series components including long-term change trends, periodic fluctuations, and short-term residual perturbations. Through this process, the deterministic changes and uncertain fluctuations in resource usage behavior are effectively distinguished. In particular, the resource fluctuation characteristics reflected by the residual part are of great reference value for non-linear modeling in a dynamic environment. Analyze the residual components in the multi-component time series data, extract the high-frequency perturbations and short-period abnormal changes existing in the resource usage process, evaluate the resource fluctuation intensity of each node in different time periods based on this, and calculate the corresponding non-linear attenuation coefficient accordingly. This coefficient is used to dynamically adjust the influence degree of each time period on the overall prediction, so that the weight of the time period with larger resource fluctuations is automatically reduced in subsequent predictions, enhancing the robustness of the prediction model to unstable behaviors and ensuring that the resource demand prediction is more stable and reliable. Perform recursive convolution calculations based on the three-dimensional resource demand tensor and the non-linear attenuation coefficient, spatially expand and temporally aggregate the resource change patterns through multiple convolutional operations, recursively extract the resource evolution characteristics across time steps and between nodes, and generate preliminary resource demand prediction values. This preliminary prediction result has temporal consistency and trend sensitivity and can reflect the short-term and medium-term change trends of node resource usage. To enhance the response ability of the prediction result to network structure changes, fuse the initial prediction value with the topological entropy weight matrix, and use the topological correlation strength between nodes in the topological entropy weight matrix to weighted correct the initial prediction result, realizing topology-aware calibration of resource demand prediction. By introducing the topology-aware mechanism, the potential influence of the resource status of topological neighbor nodes on the resource demand of this node is effectively captured. Especially in a network environment where the resource status has spatial propagation characteristics, this adjustment step can significantly improve the prediction accuracy and local consistency, and finally obtain the resource demand prediction matrix.

[0022] Resource usage records of each network node at multiple consecutive historical time points are extracted from the three-dimensional resource demand tensor. These records cover different resource dimensions and have a time series structure. Based on the extracted historical data, different weights are assigned to the resource data at each time point according to the pre-computed non-linear attenuation coefficient, so that in the time period with drastic changes or high-frequency fluctuations in resource usage, the corresponding data weights are naturally reduced, thereby suppressing the interference of these noise components on the judgment of the overall trend and forming a time-weighted resource data sequence. The topological entropy weight matrix is used to identify other nodes that have a strong connection with the current target node in the topological structure. Specifically, all nodes with topological weights greater than the set threshold are selected from the topological matrix to form the influence node group of the target node. These nodes form actual data flows, task interactions, or physical connection relationships with the target node in the network structure, and changes in their resource states may have direct or indirect effects on the resource requirements of the target node. For each node in the influence node group, its time-weighted resource data is multiplied by the topological entropy weight of the node to the target node, and the sum of the product results of all nodes is calculated, so as to fuse the structural information and the time evolution information, form a comprehensive resource input tensor with structural sensitivity and time series weight regulation ability, and perform multi-layer convolution processing on the comprehensive tensor through a recursive convolution mechanism to extract deep resource evolution features and generate an initial resource demand prediction value. The change of the topological entropy weight matrix within a continuous time window is detected, the change amplitude of the topological weights in adjacent time slices is calculated, and the network topology change rate is constructed based on this. This change rate represents the dynamic evolution characteristics of the network structure in the time domain. To avoid numerical instability caused by directly using the change rate, a logarithmic gain function is introduced to perform non-linear processing on the change rate to obtain a dynamic topology-aware perturbation factor. The initial resource demand prediction value is multiplied by the dynamically adjusted topology-aware perturbation factor, and combined with the direction of the change trend of the current resource demand, that is, the sign information of the trend function, to perform a joint operation to form the final resource demand prediction matrix.

[0023] Step S103: Based on the resource demand prediction matrix, perform a segmented progressive allocation decision on the network node resources to obtain the first resource allocation decision matrix; Specifically, a clear minimum divisible unit of resources is set for various resource types involved in the network system, enabling subsequent allocation operations to be carried out with a unified resource granularity, thus avoiding waste caused by overly large allocation units and preventing an increase in system complexity due to overly small allocation units. The minimum divisible unit of resources is set according to the characteristics of the resource category. On this basis, the total capacity of various resources in the current system and the total amount of occupied resources are statistically analyzed in real time, and the remaining available capacity of various resources is calculated by taking the difference between the two, providing resource boundary constraints for the entire allocation decision-making process. The actual business requirements and importance of each network node are comprehensively evaluated, and each node is scored from three dimensions: the criticality of the business, the centrality of the node in the network topology, and the priority set by the administrator, and weighted aggregation is performed according to the set weight coefficients to form the node importance weight, which reflects the priority ranking of the node in the current resource competition. The resource allocation task is divided into multiple consecutive decision segments, and reasonable division is performed between different decision segments according to the phased characteristics of the allocation and the available resource amount, forming a resource allocation plan that progresses segment by segment and gradually saturates. The total amount of resources planned to be allocated in each decision segment shows a trend of gradually increasing or dynamically adjusting back as the stage progresses, thus forming a controllable rhythm of the allocation strategy. Within each decision segment, combining the total amount of resources in this stage, the node importance weight, and the predicted resource demand data, calculate the value coefficient of the corresponding resources for each network node in this segment, and generate the corresponding decision score according to this value coefficient. The higher the decision score of a node, the stronger its current resource usage expectation and the greater the system benefit. Therefore, during the resource allocation process in this segment, resources will be preferentially tilted towards the nodes with high scores. The resource allocation is carried out step by step with the minimum divisible unit as the basic unit until the resource allocation quota for this decision segment is exhausted or the demands of all nodes are met. Calculate the judgment reversal threshold for the current stage based on the total number of current network nodes, and when it is monitored that the score change amplitude of a certain node in two adjacent decision segments exceeds this threshold, trigger a re-evaluation and correction of the allocation result of the previous segment. This mechanism effectively avoids resource misallocation caused by input prediction errors or stage disturbances, ensuring the global consistency and fairness of the allocation result. Integrate the allocation results formed in each decision segment to uniformly construct the first resource allocation decision matrix.

[0024] Dynamically calculate the judgment reversal threshold corresponding to this decision segment according to the total number of nodes in the current network and the serial number of the decision segment where it is located. This threshold is designed as a function that changes with the node scale and the progress of the stage, so as to set a relatively loose tolerance range in the initial stage to ensure the promotion efficiency of the allocation, and increase the sensitivity in the later stage to capture abnormal fluctuations in the allocation process. Calculate the decision score deviation value of each network node between two consecutive decision segments, so as to capture whether there is a sudden change in the resource value state of the node in adjacent stages. When the decision score deviation of a certain node exceeds the judgment reversal threshold of this decision segment, it is considered that there is a resource misjudgment or scheduling mutation in the allocation process of this node, and it is marked as a candidate node and added to the re-evaluation list. Subsequently, recalculate the resource value coefficient and decision score of the candidate node in the previous stage. The data used is still from the resource demand prediction matrix, and the current system state and the remaining resources are considered. Through recalculation, ensure that the correction plan is current and accurate. To enhance the stability and anti-interference ability of resource allocation, introduce a resource locking period calculation mechanism jointly determined by the topology-aware perturbation factor and the judgment reversal threshold. This mechanism establishes a resource locking window for each node to provide relatively stable resource quota guarantee for key nodes when resource fluctuations are severe or decision adjustments are frequent. The length of this period will increase with the increase of the node topology perturbation degree, ensuring that in the case of drastic changes in the topology structure or frequent adjustments of resource prediction, the system can reserve sufficient buffer time for core nodes, thus avoiding performance oscillations or resource recovery conflicts caused by frequent adjustments. After the decision results and correction adjustments in the above stages are completed, integrate the allocation results for each node and each resource dimension in all decision segments to form a resource allocation data structure. Especially when dealing with core nodes, perform final revisions according to their resource locking periods, ensure that the quotas of these nodes are relatively fixed during the resource reallocation period, and they can only participate in subsequent dynamic allocations after the lock-in period expires, so as to form a resource management strategy that takes into account both flexibility and stability, and finally obtain the first resource allocation decision matrix.

[0025] Step S104: Perform chaotic perturbation fine-tuning on the first resource allocation decision matrix to obtain a second resource allocation decision matrix and a resource locking schedule; Specifically, the historical performance monitoring data of each network node is imported from the system, and based on this, a comprehensive performance evaluation function is constructed. This function integrates multiple dimensional indicators such as response time, throughput capacity, and error rate, and reflects the current overall operating state of the node through normalization and weighted integration. On this basis, the actually observed node performance is compared with the theoretical expected performance derived from the first resource allocation decision matrix, and the difference value between the two is calculated to obtain the performance feedback closed-loop value of each node, which characterizes whether the current resource configuration meets the actual business requirements and whether there is an offset between the system load and the allocation scheme. The perturbation coefficient used for chaotic perturbation calculation is dynamically adjusted according to the size of the performance feedback closed-loop value. This perturbation coefficient is derived from the chaotic function model of the Logistic-like mapping, and its evolution sensitivity will increase significantly as the feedback deviation increases, so that stronger perturbations are introduced when the performance deviation is severe to drive the system out of the local optimum, while the perturbations tend to converge when the performance is stable or the deviation is weak, ensuring the tuning stability of the system. The perturbation coefficient, as a regulation amplitude factor, is dynamically generated according to the actual feedback of the node and participates in the subsequent resource adjustment decision-making logic in an adaptive manner. The resource adjustment factor of each node is calculated using the node performance feedback closed-loop value and the corresponding perturbation coefficient. This factor indicates whether the resource allocation ratio should be expanded or contracted, and the adjustment trend is judged in combination with the direction of the resource change trend. When this resource adjustment factor exceeds the set preset deviation threshold, it indicates that the current configuration of the node has deviated significantly from the ideal state, and the system generates a resource adjustment trigger signal. This signal is used to initiate the dynamic correction process of the first resource allocation decision matrix. The system adjusts the resource allocation ratio of the relevant nodes according to the perturbation amplitude, and a upper and lower limit constraint mechanism is introduced during the adjustment process to ensure that the adjustment result does not exceed the preset minimum or maximum resource limit, thereby avoiding new performance instability caused by excessive perturbation of the resource configuration. The adjusted allocation result constitutes the second resource allocation decision matrix. To avoid system oscillations caused by frequent resource changes, the resource elastic recovery period of each node is calculated based on the node performance feedback closed-loop value. The length of this period is non-linearly amplified according to the degree of performance deviation, forming a longer recovery buffer window for nodes with large performance fluctuations, while maintaining a short cycle adaptability for nodes with stable performance, so as to establish a flexible control mechanism for the resource adjustment rhythm. The result of the recovery period is recorded as the resource lock schedule, which specifies which nodes are in the allocation lock state in which resource dimensions within a certain period in the future, and further configuration modification is prohibited to prevent frequent resource jitters from interfering with system stability.

[0026] The numerical ratio of the node performance feedback closed-loop value to its corresponding disturbance coefficient is calculated. Through this ratio, the proportional relationship between the current performance deviation degree of the system and the disturbance elastic adjustment ability is effectively reflected. The larger the ratio, the more significant the current performance deviation of the node and the relatively insufficient existing disturbance adjustment ability. Therefore, there is a stronger demand for resource reallocation motivation for this node. To avoid the risk of instability caused by linear over-amplification of the ratio, the hyperbolic tangent function is introduced to perform non-linear transformation on the performance disturbance ratio. Through the smooth mapping effect of the hyperbolic tangent function, the maximum value is compressed to a finite interval, while maintaining sensitivity to both positive and negative directions, forming a basic detachment value that is sensitive to deviations and has the ability to clamp extreme values. A fixed preset adjustment intensity parameter is introduced. This parameter serves as a globally unified resource adjustment ratio amplification coefficient to control the influence degree of the basic detachment value on the subsequent resource adjustment amplitude. The basic detachment value is multiplied by the adjustment intensity parameter to obtain the resource adjustment factor. The absolute value of the resource adjustment factor is taken, and this absolute value is numerically compared with the resource adjustment detachment threshold set by the system to determine whether the current node meets the condition for triggering the resource adjustment operation. If the absolute value of the resource adjustment factor exceeds the preset detachment threshold, it is considered that the resource allocation of this node has deviated from the reasonable state, and a positive resource adjustment trigger signal is generated, marking that this node needs to perform resource reallocation in the current cycle; otherwise, if the absolute value of the resource adjustment factor is still lower than the threshold, the system determines that this node is still in a stable resource state and there is no need to adjust the resource allocation, thus generating a negative resource adjustment trigger signal. Whether to update the corresponding element in the first resource allocation decision matrix is determined according to the state of this trigger signal. When receiving a positive resource adjustment trigger signal, the system performs resource adjustment calculation on the resource adjustment factor of this node and its corresponding resource allocation value in the first resource allocation decision matrix, that is, adjusts the original resource allocation result according to the current ratio to form an allocation configuration closer to the current performance. When the system detects a negative resource adjustment trigger signal, it means that the resource allocation of this node is still reasonable and stable under the current cycle. Therefore, the system will keep the allocation value of this node in the first resource allocation decision matrix unchanged to avoid unnecessary resource disturbances and frequent resource recycling or expansion operations, thereby improving the stability and self-consistency of the overall system resource regulation.

[0027] Step S105: Based on the second resource allocation decision matrix and the resource locking schedule, perform multi-level scheduling of network node resources to generate a resource execution result log and performance monitoring data.

[0028] Specifically, all nodes in the network are grouped and classified according to the criticality of their business load, divided into critical node groups, standard node groups, and elastic node groups. Among them, critical nodes undertake core business loads and have extremely high requirements for the real-time and stability of resource allocation. Standard nodes undertake routine business tasks and have a certain degree of adjustment room. Elastic nodes are mostly used for auxiliary functions and have the greatest tolerance for resource changes. Calculate the difference value between the current actual resource configuration of each node and the corresponding target configuration in the second resource allocation decision matrix. This difference value reflects the deviation degree between the ideal allocation and the actual state, and calculate the relative ratio of the difference in combination with the current resource configuration of the node to judge the influence intensity of this difference in the node configuration structure. Based on the above difference value and relative ratio, combined with the grouping type to which the node belongs, allocate a set of differentiated resource scheduling execution strategies to each node. Among them, critical nodes preferentially adopt a gradual and smooth fine-tuning mechanism to gradually adjust the resource quota while ensuring service continuity; standard nodes implement a phased adjustment strategy according to the size of the difference. If the resource deviation is small, it is executed centrally, and if the deviation is large, it is promoted in batches; while elastic nodes adopt a one-time direct resource reconfiguration method to accelerate the efficiency of resource release and reallocation. To ensure the coordination between multiple resources, model the dependency relationships between various resource dimensions involved in network nodes, construct a resource dependency graph. This graph represents the sequential logic in which some resource dimensions must depend on other resource dimensions to complete configuration preferentially at the functional or performance level. By performing a topological sort on the resource dependency graph, determine the legal order of global resource adjustment, thus avoiding node configuration failures or abnormal operations caused by improper handling of resource dependency relationships. According to the differentiated execution strategy, the resource scheduling execution order, and the resource locking schedule, perform resource adjustment operations on each node one by one. Nodes within the locking period will skip the current round of adjustment to ensure the continuity of their resources and business stability. During the adjustment process, real-time monitor the key performance indicators of each node, such as response time, processing capacity, and error rate. When the system detects that the performance of a certain node shows a downward trend exceeding the preset deterioration threshold after adjustment, immediately abort the current resource adjustment process and roll back the resource configuration state of this node to the previous round of stable allocation result to avoid performance loss caused by resource adjustment. At the same time, this node is added to the temporary protection list and will not participate in subsequent resource dynamic adjustments within the set observation window to stabilize the overall operation of the system. During the entire multi-level resource scheduling execution process, record all resource change events, policy applications, adjustment paths, and the evolution trajectories of performance indicators, and generate resource execution result logs and performance monitoring data files according to nodes, time, and resource dimensions.

[0029] In the embodiments of the present invention, by constructing a multi-dimensional network resource monitoring and acquisition matrix and performing topology-aware initialization, the limitation that resource monitoring and network topology analysis are mutually separated in the prior art is broken, the integrated analysis of network resource usage and topology structure is realized, the non-linear influence and adaptive adjustment mechanism of topological relationships are introduced, the linear limitation of the resource prediction model in the prior art is broken through, so that the prediction model can sensitively perceive network topology changes and correspondingly adjust the prediction strategy, and the prediction accuracy of the real resource requirements of network nodes is improved. By defining the minimum divisible unit of resources and the multi-stage progressive decision-making mechanism, the limitations of traditional one-time global optimization or simple greedy algorithms are overcome, the local optimum trap is avoided, the global optimization of resource allocation is realized, and the decision-making reliability is improved through the decision reversal threshold and the allocation backtracking mechanism. By introducing controllable chaotic behavior and the resource elastic recovery period, when the resource allocation falls into a local optimum but the performance is poor, the balance can be actively broken to explore a better allocation scheme, and at the same time, the adjustment range is ensured to be within a reasonable range through the upper and lower limits of resource adjustment, improving the system stability. A differentiated, multi-level, and smoothly transitioning resource scheduling execution strategy is realized, the problem of system oscillation caused by the "one-size-fits-all" resource adjustment in traditional technologies is overcome, and the correct order of resource scheduling is ensured through the resource dependency graph and topological sorting, guaranteeing service continuity. A complete closed-loop feedback mechanism is established, the continuous self-optimization of the resource allocation system is realized, the limitation of traditional static preset parameters is broken through, so that the system can continuously learn and improve the resource allocation strategy to adapt to the complex and changeable network environment.

[0030] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) Collect the computing resources, memory resources, storage resources, and bandwidth resources of network nodes at a fixed frequency to obtain multi-dimensional resource usage data; (2) Perform a standardization transformation on the multi-dimensional resource usage data to obtain a standardized resource monitoring matrix; (3) Construct a network topology relationship graph based on the connection relationship and communication situation between network nodes, and extract node topology correlation data based on the network topology relationship graph; (4) Calculate the weights and perform normalization processing on the node topology correlation data to obtain a topological entropy weight matrix.

[0031] Specifically, a resource collection mechanism is deployed throughout the network infrastructure and distributed among various network nodes to collect computing resources, memory resources, storage resources, and bandwidth resources of the network nodes at a fixed frequency. The setting of the collection frequency needs to strike a balance between meeting the system's dynamic response capabilities and the data storage load. The collection of computing resources is expressed by statistics such as processor utilization rate, the number of active threads, or the floating-point instruction execution ratio. The monitoring of memory resources involves dimensions such as physical memory occupancy rate, the number of available buffers, and the memory page swap rate. The monitoring of storage resources includes the current disk write rate, the proportion of remaining storage space, and the disk I / O waiting time. The usage of bandwidth resources is characterized by parameters such as real-time network throughput, inbound and outbound rates, etc. Data for each resource dimension is collected through node-embedded probes, uniformly recorded, and marked according to the node identifier and timestamp, thus constructing a multi-dimensional resource usage data sequence. All the collected resource data is standardized. Different metrics such as the percentage of computing resources, memory in MB or GB, storage in disk blocks, and bandwidth in Mbps are uniformly mapped to the [0, 1] interval, so as to be comparable and numerically stable in subsequent algorithms. The standardization process adopts the range normalization form, that is, linear scaling is performed based on the minimum and maximum values of each resource dimension in the historical sampling data, so that the fluctuations of the same resource among different nodes are quantified as the relative strength performance within the standard scale. The resource monitoring matrix constructed through this standardization mechanism is presented in a three-dimensional structure, with the horizontal axis representing different network nodes, the vertical axis representing resource dimensions, and the depth dimension representing the time window. A network topology relationship graph is constructed based on the connection relationship and communication situation among network nodes. This topology relationship graph consists of a node set and an edge set. The node set represents all independent resource entities in the network, and the edge set represents the data interaction or physical connection existing between nodes, which is extracted through access control lists, link mapping information, MAC tables, or network routing tables. And communication statistical information between nodes is introduced as an auxiliary feature, such as the number of communications per unit time between node pairs, the total amount of transmitted bytes, the communication success rate, or the connection duration, etc. These communication features can describe the real interaction density between nodes in the network and reveal which node pairs in the network have stronger resource collaboration requirements for each other. Based on the fusion modeling of the connection relationship and communication intensity, a set of topological association indicators is established for each pair of nodes with a connection relationship. These indicators not only consider the length of the physical or logical path between nodes but also the density and stability of communication activities, thus constituting a topological association data set reflecting the dependency relationship of network nodes. Weight calculation and normalization processing are performed on the above topological association data to construct a topological entropy weight matrix. This matrix takes each node as the center, calculates the relative structural influence strength between it and all other nodes, and converts the connection weights between different nodes into normalized values with entropy significance through standardization processing.The principle of weight calculation comprehensively considers the attenuation effect of path length and the enhancement effect of communication activities. In the specific calculation process, a composite weight based on the two factors of hop count and historical traffic will be assigned to each edge. The fewer the hop count and the greater the traffic, the higher the weight, and vice versa, which will be naturally suppressed. All calculated edge weights will be normalized at the node level to form a square entropy weight matrix, where the element in the i-th row and j-th column of the matrix represents the degree of influence of the j-th node on the i-th node in the topological structure.

[0032] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) Construct a three-dimensional resource demand tensor including time series, node identification, and resource dimension based on the standardized resource monitoring matrix; (2) Perform time series decomposition processing on the three-dimensional resource demand tensor to obtain multi-component time series data; (3) Calculate the node resource fluctuation characteristics according to the residual component in the multi-component time series data to obtain the non-linear attenuation coefficient; (4) Perform recursive convolution calculation based on the three-dimensional resource demand tensor and the non-linear attenuation coefficient to obtain the initial resource demand prediction value, and perform topological perception adjustment on the initial resource demand prediction value in combination with the topological entropy weight matrix to obtain the resource demand prediction matrix.

[0033] Specifically, a three-dimensional resource demand tensor consisting of time series, node identifiers, and resource dimensions is constructed based on a standardized resource monitoring matrix. This allows a single data structure to simultaneously represent the dynamic usage status of various resources across different nodes at different times. A sliding time window mechanism ensures that each tensor generation covers a relatively continuous and representative data interval. Time series decomposition is performed on the time series data for each node-resource dimension combination in the three-dimensional resource demand tensor. Based on the decomposition principle, this process decomposes the original time series into three components: a trend term, a seasonal term, and a residual term. The trend term reflects the overall trend of resource usage, the seasonal term reveals cyclical variations, and the residual term captures random fluctuations caused by external disturbances, sudden business requests, or network fluctuations. A sliding window and minimum residual fitting method are used to improve the smoothness and stability of the trend decomposition, while a multi-scale period decomposition method is used to effectively extract periodic components of varying durations. By acquiring this multi-component time series data, a structured separation of resource usage evolution paths is achieved. The variance fluctuation characteristics of the residual term for each node in each resource dimension are statistically analyzed to identify the degree of resource usage dispersion. Based on the residual fluctuation, a nonlinear attenuation coefficient for each dimension is derived. This coefficient is designed to automatically reduce the influence of historical data from highly volatile resources on future forecasts, while maintaining a high weight for stable resource dimensions. This approach demonstrates dynamic tolerance for resource type heterogeneity in the temporal modeling process. The nonlinear attenuation coefficient is generated by normalizing the residual variance using a function mapping method. It is constructed through nonlinear processing methods such as exponential compression, logarithmic transformation, and tangent scaling, enabling the prediction model to exhibit stronger convergence capabilities under severe resource fluctuations. Recursive convolution is performed based on the three-dimensional resource demand tensor and the nonlinear attenuation coefficient to construct a preliminary resource demand forecast. This recursive convolution process not only perceives the resource state of the current node at each time step, but also incorporates the influence of resource states in historical time slices. The nonlinear attenuation coefficient is applied to the weight kernel corresponding to each historical step, automatically adjusting the influence of different historical segments on the current forecast based on volatility. This recursive convolutional structure is similar to a form of time-aware neural network. Its core advantage lies in capturing the evolutionary patterns of resource usage across time dimensions while maintaining the ability to respond to sudden changes in resource demand. During the convolution operation, the evolutionary trajectory of resource usage between nodes is systematically encoded into the prediction output, forming a preliminary estimate of resource demand at the next moment, which is the initial resource demand prediction value. In order to improve the sensitivity and structural consistency of the prediction results to changes in network structure, the initial resource demand prediction value is fused with the topological entropy weight matrix to complete the topology-aware adjustment operation. This process filters out other nodes that have a correlation with the current target node based on the topological entropy weight matrix, obtains their prediction results, and introduces the prediction information of neighboring nodes through weighted aggregation to form a prediction result with enhanced structural correlation.During the topology-aware adjustment process, the connection strength in the weight matrix will be used as a multiplier to participate in the correction calculation of each set of prediction data, so that the prediction results between nodes tend to converge in the dense area of the topology structure, while higher prediction differences are maintained in the sparse or edge areas of the structure, improving the overall prediction adaptation ability of the system. Through the above steps, a resource demand prediction matrix is obtained. This matrix is indexed by time step, node number, and resource dimension, providing high-precision prediction results of various resource types in the whole network within a specific future period.

[0034] In a specific embodiment, the process of performing recursive convolution calculation based on the three-dimensional resource demand tensor and the non-linear attenuation coefficient to obtain the initial resource demand prediction value, and performing topology-aware adjustment on the initial resource demand prediction value in combination with the topological entropy weight matrix to obtain the resource demand prediction matrix may specifically include the following steps: (1) Extract the resource usage records of each network node at multiple historical time points from the three-dimensional resource demand tensor, and assign different weights to the data at different time points according to the non-linear attenuation coefficient to obtain time-weighted resource data; (2) Based on the topological entropy weight matrix, screen out the set of nodes whose topological association degree with the target node is higher than the preset threshold to obtain the influencing node group of the target node; (3) Accumulate and sum the product of the time-weighted resource data and the topological entropy weight of each network node in the influencing node group, and perform a recursive convolution operation to obtain the initial resource demand prediction value; (4) Calculate the change rate of the topological entropy weight matrix within a continuous time window, and perform processing in combination with the logarithmic gain function to obtain a dynamically adjusted topology-aware perturbation factor; (5) Perform a product operation on the initial resource demand prediction value, the dynamically adjusted topology-aware perturbation factor, and the sign of the resource demand change trend to obtain the resource demand prediction matrix.

[0035] Specifically, resource usage records of each network node at multiple historical time points are extracted from the three-dimensional resource demand tensor, such that a resource evolution sequence is formed for each node on the time axis. These evolution sequences will serve as the basis for time modeling. Meanwhile, differential weights are assigned to these historical records in combination with a non-linear attenuation coefficient, where the non-linear attenuation coefficient is generated according to the residual perturbation degree of each time slice, and is used to reflect the effectiveness and signal-to-noise ratio level of each historical time point in resource state prediction. By applying the non-linear attenuation coefficient to the historical resource values in chronological order, weighted fusion is completed to obtain time-weighted resource data. Network structure information is introduced to enhance the local structure adaptability of the prediction. Neighborhood screening is performed on the target node based on the topological entropy weight matrix, so as to identify other node sets that have a strong topological dependence relationship with it. The specific method is to traverse all the weight values in the row corresponding to the target node in the topological entropy weight matrix, and include the nodes with weights greater than a certain set threshold in the influencing node group, and this threshold is adaptively adjusted according to the network density, average connection strength or system complexity. For the time-weighted resource data of all nodes in the influencing node group, product operations are respectively performed with their corresponding topological entropy weight values, mapping the degree of structural influence to the resource usage behavior, so that nodes with larger weights have a higher impact on the prediction result. After weighting, the product results are accumulated and summed according to the node dimension to generate a resource impact summary vector for the target node. The resource impact summary vector is input into the recursive convolution calculation process. In this process, through the recursive window sliding along the time axis, combined with the weighted historical data and the topological structure perception characteristics, deep resource usage patterns are extracted to generate the initial resource demand prediction value. To enhance the response ability of the prediction model to network dynamic changes, the evolution of the topological structure over time is analyzed. Therefore, differential operations are performed on the changes of the topological entropy weight matrix within consecutive time windows, calculating the change amplitude of the topological correlation strength between each pair of nodes between adjacent time slices. This change rate quantifies the degree of structural fluctuation and can identify potential drastic changes in the topological structure. Through non-linear mapping processing of this change rate by a logarithmic gain function, sensitive responses are ensured during large fluctuations while maintaining stable outputs during small fluctuations. This processing process generates a dynamically adjusted topological perception perturbation factor. Product operations are performed on the initial resource demand prediction value and the dynamically adjusted topological perception perturbation factor, and the direction symbol of the resource demand change trend is introduced to participate in the calculation. The trend direction symbol is obtained from the first-order change symbol of the resource demand prediction value on the time axis, and is used to identify whether the resource shows an increasing, decreasing or stable trend. By fusing the prediction value, the perturbation factor and the trend symbol, a resource demand prediction matrix with directionality, structural sensitivity and time perception ability is formed. This matrix is indexed by node identifier, resource dimension and future time step, reflecting the resource demand expectation under the triple influence of structural fluctuation, historical fluctuation and resource trend.

[0036] In a specific embodiment, the process of executing step S103 may specifically include the following steps: (1) Define the minimum divisible unit of resources for various types of network resources, and calculate the difference between the total currently available resources and the total allocated resources to obtain the allocable resource quantity; (2) Perform weighted calculation on each network node based on the business criticality, topological centrality, and priority to obtain the node importance weight; (3) Divide the resource allocation process into multiple decision segments, and calculate the resource quantity to be allocated for each segment based on the number of decision segments and the allocable resource quantity to obtain the decision segment resource allocation plan; (4) Calculate the resource value coefficient and decision score of each network node in each decision segment according to the node importance weight and the resource demand prediction matrix, and allocate resources in descending order of the decision score based on the decision segment resource allocation plan to obtain the resource allocation results for each segment; (5) Calculate the decision reversal threshold according to the number of network nodes, and re-evaluate the previous segment's allocation decision when the deviation of the continuous decision scores of the nodes exceeds the decision reversal threshold, and finally integrate the allocation results of each decision segment to obtain the first resource allocation decision matrix.

[0037] Specifically, refine the minimum operable allocation granularity for various types of resources in the system, that is, the minimum divisible unit of resources. This definition is set in combination with the hardware configurable capabilities, the operating system allocation interface, and the minimum control interval of the task scheduling mechanism. By clarifying the minimum allocation unit, iterative cumulative operations are performed in subsequent allocation executions according to the granular resource units, avoiding waste or insensitive scheduling caused by coarse granularity in resource configuration. Statistically calculate the total capacity and the allocated amount of each type of resource in the current global state, and obtain the total amount of resources that can actually be disposed of through the difference between the two. To guide the priority order of resource allocation decisions, weight evaluations are performed on all nodes in the network. This evaluation comprehensively considers three dimensions of indicators: First, the importance of the services borne by the node, that is, the criticality of the services associated with the node to the overall operation of the system or user services; second, the centrality of the node in the network topology, that is, its position as a hub in the communication path; third, the scheduling priority manually set by the operation and maintenance personnel or system policies. The data of each dimension is fused through standardization combined with the set weighting coefficients, and the comprehensive importance weight of each node is output. The higher this weight, the higher the priority of the node in resource scheduling. Divide the entire resource allocation process into multiple decision segments. The segmented structure helps to achieve progressive and traceable resource allocation, reducing resource waste or local optimal traps caused by one-time scheduling. The number of decision segments is dynamically generated based on the scale of the network nodes. For example, it is determined by adding a constant offset to the logarithmic function of the number of nodes, ensuring that there are enough allocation stages in a large network to refine the control. After dividing the decision segments, according to the total allocable amount of each type of resource and the number of decision segments, calculate the resource quota corresponding to each stage, ensuring that the resource allocation amount of each segment gradually increases or remains balanced, and record it as a segmented resource allocation plan for subsequent consumption round by round according to the node scores. In each decision segment, according to the comprehensive weight of the node and the future resource demand trend data provided by the resource demand prediction matrix, calculate the value coefficient of each node in the current resource dimension. Specifically, combine and adjust the weights of the node importance and its predicted demand intensity, and introduce a feedback adjustment factor with reference to the resource acquisition situation in the previous round to avoid concentrating resources on nodes that have obtained sufficient resources. Then, through further amplification and standardization of the value coefficient, obtain the final decision score of each node in this decision segment. The score results of all nodes are sorted separately according to the resource type. According to this sorting and the resource allocation quota of the decision segment, start allocating the minimum divisible resource units in sequence from the node with the highest score until the resources of the current segment are completely allocated or the node demand is met, forming the resource allocation result of this segment. To prevent obvious jumps in node resource configuration due to input changes, prediction errors, or competition disturbances in consecutive decision segments, a judgment reversal mechanism is introduced.This mechanism dynamically calculates the judgment reversal threshold for each decision segment based on the total number of network nodes. This threshold gradually stabilizes as the network scale grows and the allocation rounds progress, and is used as the basis for determining whether the continuous change in the node score is abnormal. When it is detected that the decision score deviation of a certain node exceeds the judgment reversal threshold in two consecutive decision segments, it indicates that there is an allocation misjudgment or policy imbalance in the previous round of allocation for this node. Therefore, the resource allocation situation of this node in the previous decision segment is re-evaluated, and its value coefficient is corrected according to the current state, and the corresponding resource units are re-allocated to ensure that the feedback of the allocation path to the actual state has the ability to respond in a timely manner. After the allocation cycle of all decision segments and the necessary reversal backtracking processing, the system integrates the resource allocation records generated in each stage to form the first resource allocation decision matrix. This matrix uses the network nodes and resource dimensions as two-dimensional indexes and records the actual resource configuration quantities obtained by each node in the current allocation cycle.

[0038] In a specific embodiment, the process of performing the steps of calculating the judgment reversal threshold according to the number of network nodes, re-evaluating the previous allocation decision when the continuous decision score deviation of the node exceeds the judgment reversal threshold, and finally integrating the allocation results of each decision segment to obtain the first resource allocation decision matrix may specifically include the following steps: (1) Calculate the judgment reversal threshold of the current decision segment according to the number of network nodes and the decision segment serial number; (2) Calculate the decision score deviation value between two consecutive decision segments for each network node, and compare the decision score deviation value with the judgment reversal threshold to obtain a list of candidate nodes that need to be re-evaluated; (3) Recalculate the resource value coefficient and decision score of the previous segment for the network nodes in the list of candidate nodes that need to be re-evaluated according to the resource demand prediction matrix to obtain the corrected previous segment allocation plan; (4) Calculate the resource locking period for each network node based on the topology-aware perturbation factor and the judgment reversal threshold to obtain the resource allocation stability guarantee strategy; (5) Integrate the resource allocation results of each decision segment according to the network nodes and resource dimensions, and process the resource allocation of the core nodes according to the resource locking period to obtain the first resource allocation decision matrix.

[0039] Specifically, the decision reversal threshold for the current decision segment is calculated based on the number of network nodes and the decision segment number. An increasing adjustment coefficient is constructed by introducing the product of the logarithmic function of node size and the decision segment number. This results in a higher threshold tolerance in the initial phase and a lower tolerance in subsequent phases, thereby improving the stability and reliability of the allocation results. After calculating the decision reversal threshold for each decision segment, the difference between the decision scores of each network node in the current segment and the previous segment is analyzed. The relative deviation is calculated by dividing the difference between the current and previous segment scores by the absolute value of the previous segment score to obtain a standardized decision score deviation ratio. This eliminates the influence of the original magnitude of the node score on the deviation judgment. Each node's score deviation ratio is then compared with the decision reversal threshold for the current decision segment to which it belongs. When the deviation ratio exceeds the threshold, the node's score is considered to be at risk of structural fluctuation or increased prediction error. The node is then included in the list of candidate nodes requiring reassessment and is considered for subsequent resource retroactive allocation. For each node in the candidate node list, the resource value coefficient and decision score calculation process from the previous decision phase is re-executed. This process uses the current resource demand forecast matrix as the demand input, incorporates the previous phase allocation results and score fluctuations as feedback correction parameters, adjusts the dynamic weight coefficient of the node importance weight, and re-evaluates the unit resource value of the node, resulting in a new round of value coefficients and scores that better reflect actual fluctuations. This recalculation process allows for local repair of resource allocation results and suppression of unstable behavior, thereby enhancing the adaptability and robustness of the segmented decision-making mechanism. The recalculated results are integrated with the current segment allocation results to form a revised resource allocation plan, ensuring that any resource imbalances caused by information lag or model errors are promptly addressed. To prevent instability caused by frequent resource switching between fluctuating nodes, a resource locking mechanism is introduced. The resource locking period for each node is calculated based on its topology-aware perturbation factor and its corresponding decision reversal threshold. The topology-aware perturbation factor reflects the degree of dynamic change in a node's topology—in other words, whether the node's association strength within the network structure fluctuates frequently. The decision reversal threshold reflects the system's tolerance for fluctuations in the node's behavior. Combining these two factors, a node resource lock-up period function is constructed using logarithmic and proportional functions. This ensures that nodes with high volatility and unstable structures have longer resource lock-up periods, during which their resource allocation results remain frozen and protected from the next round of allocation. Stable nodes, on the other hand, maintain short lock-up periods, allowing for rapid response to demand changes. After all decision segments have completed allocations and unstable nodes have undergone backtracking and lock-up adjustments, the resource allocation results for each segment are consolidated to construct a global, first-order resource allocation decision matrix.The integration process summarizes the allocation results of each segment node by node and resource dimension. At the same time, according to the resource locking cycle table, a freezing operation is imposed on the node resource dimension in the locked state, that is, no subsequent iterative overwrite is performed on its resource configuration to maintain the continuity of the allocation state. In addition, during the integration process, it is detected whether there are overlapping or gap situations in the resource allocation amount. If there is an allocation conflict for a certain type of resource between different decision segments, the non-locked state configuration of the latest segment is used as the standard for merging; if there is a vacancy in the resource configuration of a certain node, it is supplemented by the result of the final segment to ensure the integrity and consistency of the matrix.

[0040] In a specific embodiment, the process of executing step S104 may specifically include the following steps: (1) Import the historical performance data of the network node to construct a performance evaluation function, and calculate the deviation between the current node performance and the expected performance to obtain the node performance feedback closed-loop value; (2) Dynamically adjust the node perturbation coefficient of the chaotic perturbation function according to the magnitude of the node performance feedback closed-loop value; (3) Calculate the resource adjustment factor using the node performance feedback closed-loop value and the node perturbation coefficient, and compare the resource adjustment factor with a preset detachment threshold to determine whether resource adjustment is required, obtaining a resource adjustment trigger signal; (4) Perform resource ratio adjustment on the first resource allocation decision matrix according to the resource adjustment trigger signal, and at the same time introduce upper and lower limit constraints to ensure that the adjustment range is within the target range, obtaining a second resource allocation decision matrix; (5) Calculate the resource elastic recovery period of each network node based on the node performance feedback closed-loop value, and generate a corresponding resource locking schedule.

[0041] Specifically, a representative, continuous, and computable performance indicator sequence is extracted from historical performance data recorded over a long period of system operation. This includes key performance parameters such as node response time, throughput, and error rate. These are then normalized into dimensionless data, allowing different indicators to be combined and weighted within a unified numerical scale. A performance evaluation function is constructed based on preset indicator weights to score the comprehensive operational quality of each node within each time window. The weighting is adaptively configured based on system characteristics, such as increasing the weight of response time in scenarios with high real-time requirements and increasing the importance of error rate in environments with high data integrity. After the performance function is constructed, the actual observed performance value of each node within the historical time slice is compared with the expected performance value calculated based on its resource allocation status. The difference between the two is calculated to obtain the node's current performance deviation sequence. This deviation sequence is then decayed and summed using a weighted sliding process to form a node performance feedback closed-loop value, which serves as an important indicator for quantifying whether the current resource allocation matches the actual business performance. The node perturbation coefficient of the chaotic perturbation function is dynamically adjusted based on the magnitude of the node performance feedback closed-loop value. The chaotic perturbation function, based on logistic mapping or quasi-chaos distribution logic, introduces a certain amount of nonlinear random perturbation for resource fine-tuning, aiming to break resource allocation structures stuck in local optima. The system dynamically adjusts the chaotic perturbation coefficient based on the feedback loop value of the current node. When performance deviation is small or stable, the perturbation parameter is kept low to maintain resource structure stability. When performance deviation is significant, the perturbation intensity is increased to sufficiently perturb the existing allocation scheme, thereby exploring possible optimal solutions. This adjustment mechanism uses function mapping to ensure that the perturbation coefficient oscillates within a defined domain, avoiding extreme jumps while maintaining a certain level of perturbation sensitivity. After the chaotic perturbation parameter is determined, the node's performance feedback loop value is linked to its perturbation coefficient to generate the resource adjustment factor for the current node. This factor comprehensively reflects the actual adaptability of the current resource allocation to system performance and the sensitivity of future adjustments, while also reflecting the nonlinear activation effect of the perturbation in the model. The resource adjustment factor is compared with the set resource adjustment deviation threshold. When the absolute value of the factor exceeds the preset threshold, it indicates that the current node resource configuration has deviated from the reasonable range or has potential for optimization. The system triggers a positive resource adjustment signal, indicating that the node will perform resource adjustment operations in the current cycle. Conversely, if the absolute value of the resource adjustment factor is lower than the threshold, a negative resource adjustment signal is triggered, which maintains the original resource configuration and ensures that the system maintains resource configuration stability under the condition of no obvious abnormalities. Based on the result of the resource adjustment trigger signal, the resource ratio of the first resource allocation decision matrix is corrected.For the nodes that receive the positive adjustment signal, the system, based on their corresponding resource adjustment factors, proportionally amplifies or compresses the configuration values of these nodes in the resource matrix, thereby reflecting the resource response behavior driven by their actual performance feedback. To prevent excessive resource fluctuations or instability in resource allocation caused by error amplification, a set of boundary constraint conditions of upper and lower limits is introduced during the adjustment operation, such that the adjusted configuration values of the resources always remain within the allowable elastic range, neither exceeding the upper limit of the remaining available amount in the resource pool nor being lower than the lower limit of the basic guaranteed resources required for node operation. After the adjustment is completed, a new second resource allocation decision matrix is obtained. To avoid the system repeatedly performing resource adjustments due to frequent response to performance fluctuations, resulting in discontinuity and uncertainty in resource management, the resource elastic recovery period of each node is calculated based on the current closed-loop value of the node performance feedback. The recovery period reflects the resource adjustment silent interval set by the system for different nodes, that is, within the recovery period, the system will suspend further adjustment operations on this node, enabling the resources to maintain a stable allocation within a certain period of time, thereby observing the actual impact of resource adjustment on node performance and avoiding frequent changes caused by short-term fluctuations or misjudgments. The setting of the recovery period is dynamically generated based on the magnitude of the closed-loop value. By introducing an exponential or logarithmic function to map the performance deviation to the time scale, nodes with large performance deviations have longer recovery periods, while nodes with small deviations can quickly enter the next round of adjustment judgment cycle. The recovery periods of all nodes are unified and summarized to generate a resource locking schedule, which records the resource adjustment silent periods of each node in each resource dimension.

[0042] In a specific embodiment, the process of performing the step of calculating the resource adjustment factor using the closed-loop value of the node performance feedback and the node perturbation coefficient, and comparing the resource adjustment factor with a preset detachment threshold to determine whether resource adjustment is required and obtaining the resource adjustment trigger signal may specifically include the following steps: (1) Perform a division operation on the closed-loop value of the node performance feedback and the node perturbation coefficient to obtain a performance perturbation ratio, and perform a non-linear transformation on the performance perturbation ratio using the hyperbolic tangent function to obtain a basic detachment value; (2) Multiply the basic detachment value by a preset adjustment intensity parameter to obtain the resource adjustment factor, where the adjustment intensity parameter is a fixed constant; (3) Calculate the absolute value of the resource adjustment factor and compare the absolute value with the preset detachment threshold. When the absolute value is greater than the preset detachment threshold, generate a positive resource adjustment trigger signal, otherwise generate a negative resource adjustment trigger signal; (4) Determine whether to adjust the first resource allocation decision matrix based on the status of the resource adjustment trigger signal. When a positive resource adjustment trigger signal is received, perform resource adjustment calculations on the corresponding elements of the resource adjustment factor and the first resource allocation decision matrix. When a negative resource adjustment trigger signal is received, keep the corresponding elements of the first resource allocation decision matrix unchanged.

[0043] Specifically, the performance feedback closed-loop value is a deviation summary metric calculated by the system after comparing the actual operating performance of a node with the expected performance corresponding to the resource allocation, reflecting the matching degree between the current state of the node and the resource support it should have. The perturbation coefficient is a non-linear weight parameter dynamically adjusted through a chaotic perturbation function based on considering the topological perturbation intensity and performance fluctuation sensitivity, used to control the perturbation intensity of resource adjustment. Perform a division operation between the performance feedback closed-loop value and the perturbation coefficient for each node to obtain the performance perturbation ratio of the node. This ratio represents the proportion of the current resource allocation error relative to the tolerable perturbation degree of the system. The larger its value, the more significant the performance deviation is compared to the perturbation ability, and the more likely the system is in a resource imbalance state. To avoid the ratio having too wide a numerical change range, overly linear adjustment response, or lack of clamping ability, input this performance perturbation ratio into the hyperbolic tangent function to perform a non-linear transformation operation to obtain the basic detachment value. The non-linear mapping characteristic of the hyperbolic tangent function enables it to map the unbounded ratio to a finite interval, thus avoiding the system reaction getting out of control or the adjustment amplitude being too large caused by extreme ratios. At the same time, this function has central symmetry and direction preservation, that is, positive and negative deviations get equal-amplitude responses on the positive and negative semi-axes respectively, facilitating directional resource adjustment in the subsequent process. The function changes gently when the ratio is small, which is beneficial for controlling the stability of resource allocation, and responds quickly when the ratio is large, helping the system to quickly respond to large deviations, so it is suitable for resource detachment detection. Multiply the basic detachment value by a preset fixed adjustment intensity parameter to generate a resource adjustment factor. This adjustment intensity parameter is a global constant set by the system, and its role is to control the response sensitivity of the final adjustment amplitude. The larger the adjustment intensity parameter, the more radical the response amplitude of the system to performance detachment, which is suitable for quickly adapting to high-frequency change scenarios; while the smaller the adjustment intensity parameter, the more cautious the system's adjustment to changes, which is suitable for a resource allocation system that prioritizes stability. Through a linear scaling operation, the basic detachment value is transformed into a resource adjustment factor that can be directly applied to the calculation of resource ratio adjustment after one enhancement. This factor has a clear numerical direction and adjustment amplitude, and becomes the decision benchmark for whether to trigger a resource adjustment operation in the current cycle. Take the absolute value of the resource adjustment factor and compare it with a preset resource detachment threshold. This detachment threshold is used as the sensitive boundary for whether to perform actual resource fine-tuning. If the absolute value of the resource adjustment factor exceeds this detachment threshold, it indicates that the performance state of the current node has significantly deviated from the balance interval that the current configuration can bear, and the system immediately responds and generates a positive resource adjustment trigger signal to enter the adjustment path; conversely, if the resource adjustment factor does not exceed the detachment threshold, the system believes that although the current node state fluctuates, it is still within the stable boundary range and does not need to be adjusted, thus generating a negative resource adjustment trigger signal to maintain the stability of the existing configuration structure unchanged. After the state of the resource adjustment trigger signal is clear, perform a conditional operation on the first resource allocation decision matrix based on this signal state.If the resource adjustment trigger signal of a certain node is positive, the system will perform a proportional adjustment operation on the resource configuration value corresponding to the node in the first resource allocation decision matrix, multiplying the configuration value by the resource adjustment factor to dynamically amplify or reduce the original resource quota, and ensuring that the adjustment direction is always consistent with the performance deviation trend, thereby completing an adaptive resource correction operation driven by performance, guided by disturbances, and controlled by function mapping. If the trigger signal of a certain node is negative, the system will directly skip the resource configuration modification of this node and keep its original value in the first resource allocation decision matrix unchanged to reduce unnecessary resource fluctuations and maintain the overall balance of the system.

[0044] In a specific embodiment, the process of executing step S105 may specifically include the following steps: (1) Divide the network nodes into a critical node group, a standard node group, and a flexible node group according to the business importance, and calculate the difference value between the current actual resource configuration of each network node and the target configuration in the second resource allocation decision matrix; (2) Configure a differentiated execution policy for each network node according to the ratio of the difference value to the current actual resource configuration, and in combination with the critical node group, the standard node group, and the flexible node group; (3) Construct a resource dependency graph based on the dependency relationships of different resource dimensions among the network nodes, perform a topological sort on the resource dependency graph to obtain the resource scheduling execution order, and perform a resource adjustment operation according to the differentiated execution policy, the resource scheduling execution order, and the resource lock schedule; (4) During the execution of the resource adjustment, monitor the performance indicators of the network nodes in real time. When it is detected that the performance degradation exceeds the preset deterioration threshold, abort the current adjustment operation and roll back to the previous stable state, add the corresponding network nodes to the temporary protection list, and at the same time record the entire execution process to generate a resource execution result log and performance monitoring data.

[0045] Specifically, a business priority evaluation system is established. This system classifies the tasks carried by each network node with core indicators such as business impact scope, real-time requirements, disaster tolerance capabilities, and user criticality, and divides all nodes into three major business level groups, namely the critical node group, the standard node group, and the elastic node group. Among them, the critical node group corresponds to the core nodes with extremely high requirements for business continuity, such as the main control server, the main database node, or the high-concurrency interface entrance. Once the resource configuration of such nodes is insufficient, it may cause the performance of the entire network to collapse; the standard node group mainly undertakes general service processing tasks. Although there are business requirements, it has a certain buffer and tolerance capacity; the elastic node group mainly executes non-critical calculations, low-priority caching, or redundant tasks, serving as the main source and buffer for dynamic resource allocation. Based on the resource configuration status of each node at the current moment, the actually allocated resource values on each resource dimension are extracted, and compared item by item with the target configuration values formulated for it in the second resource allocation decision matrix to obtain the resource configuration difference value. Divide the resource configuration difference value by the current actual resource configuration value of the node respectively to obtain the relative ratio of resource change, which reflects the impact intensity that the resource adjustment operation may cause to the current state of the node. Combining the business level classification information of the node and the resource configuration difference ratio, a differentiated resource adjustment execution strategy is formulated for each node. For the critical node group, an incremental smoothing adjustment strategy is adopted, that is, approaching the target resource configuration step by step with the smallest divisible unit of resources, and introducing a time interval and a performance feedback monitoring mechanism between each adjustment to avoid the risk of system instability caused by a large one-time adjustment; the standard node group adopts a phased adjustment strategy, that is, it is divided into multiple adjustment stages according to the difference ratio, and advances batch by batch according to the resource change curve, and appropriately introduces performance threshold conditions for rhythm control; the elastic node group executes a direct adjustment strategy on the premise of not affecting the global resource balance, that is, completes the overall modification of the resource configuration within a single cycle to release redundant resources to the greatest extent and provide resource support for critical nodes. In order to achieve the coordinated scheduling between multiple resource types, model the internal dependency relationship between resource dimensions, construct a resource dependency graph, where each node of the graph represents a resource dimension, such as computing, memory, storage, and bandwidth, and the direction of the edge represents the resource configuration sequence relationship. For example, bandwidth expansion usually depends on the prior availability of computing and memory resources, and memory adjustment needs to be completed after the adjustment of the number of CPU cores. After construction, perform a topological sort on the resource dependency graph to determine the legal execution order between resource configurations, so as to ensure that the dependency conditions during resource adjustment are not violated, and at the same time avoid resource waste or temporary performance degradation caused by reverse configuration. In the actual execution stage of resource scheduling, link the scheduling process with the above-generated differentiated execution strategy, the resource dependency execution order, and the resource locking schedule. The node and resource dimension combinations marked in the locking schedule will skip the current round of resource adjustment to ensure the adjustment interval and system stability, and the remaining nodes will execute the adjustment operations one by one according to the resource dependency order.In each step of resource allocation adjustment, the system dynamically evaluates the real-time performance status of the current node, including indicators such as response latency, processing load, and error rate, and compares the performance changes before and after the adjustment. When it is detected that the decline in the key performance indicators of a certain node exceeds the deterioration threshold preset by the system after the resource adjustment, the resource adjustment operation of the current node is immediately aborted, and it is automatically rolled back to the previous stable configuration state. At the same time, the node is temporarily added to the resource protection list, and its readjustment qualification is postponed in the subsequent cycles to prevent the spread of performance oscillations in the system. During the entire adjustment process, all execution behaviors are recorded, including the adjustment actions of each node in each resource dimension, the configuration values before and after the adjustment, the performance status, the adjustment time consumption, whether to trigger a rollback, and whether to enter the protection list, etc., and a structured resource execution result log is formed. At the same time, the performance indicators of the nodes at each moment are sampled in real time according to the resource dimension to generate a performance monitoring data matrix.

[0046] The above describes the method for dynamically allocating network node resources in the embodiments of the present invention. Next, the system for dynamically allocating network node resources in the embodiments of the present invention will be described. Please refer to Figure 2 , an embodiment of the system for dynamically allocating network node resources in the embodiments of the present invention includes: An acquisition module 201, configured to acquire multi-dimensional resource usage data of a network node and perform preprocessing to obtain a standardized resource monitoring matrix and a topological entropy weight matrix; A calculation module 202, configured to perform non-linear topology-aware calculation according to the standardized resource monitoring matrix and the topological entropy weight matrix to obtain a resource demand prediction matrix; An allocation decision module 203, configured to perform a segmented progressive allocation decision on the network node resources based on the resource demand prediction matrix to obtain a first resource allocation decision matrix; A fine-tuning module 204, configured to perform chaotic perturbation fine-tuning on the first resource allocation decision matrix to obtain a second resource allocation decision matrix and a resource locking schedule; A multi-level scheduling module 205, configured to perform multi-level scheduling of network node resources based on the second resource allocation decision matrix and the resource locking schedule to generate a resource execution result log and performance monitoring data.

[0047] Through the collaborative cooperation of the above-mentioned various components, by constructing a multi-dimensional network resource monitoring and acquisition matrix and performing topology-aware initialization, the limitation of the mutual separation of resource monitoring and network topology analysis in the prior art is broken, and the integrated analysis of network resource usage and topology structure is realized. The non-linear influence and adaptive adjustment mechanism of the topological relationship are introduced, breaking through the linear limitation of the resource prediction model in the prior art, enabling the prediction model to sensitively perceive network topology changes and adjust the prediction strategy accordingly, and improving the prediction accuracy of the true resource requirements of network nodes. By defining the minimum divisible unit of resources and the multi-segment progressive decision-making mechanism, the limitations of traditional one-time global optimization or simple greedy algorithms are overcome, the local optimum trap is avoided, the global optimization of resource allocation is achieved, and the decision-making reliability is improved through the decision reversal threshold and the allocation backtracking mechanism. The introduction of controllable chaotic behavior and the resource elastic recovery period can actively break the balance when the resource allocation falls into a local optimum but with poor performance, explore a better allocation scheme, and at the same time ensure that the adjustment range is within a reasonable range through the upper and lower limits of resource adjustment, improving the system stability. A differentiated, multi-level, and smooth-transition resource scheduling execution strategy is realized, overcoming the system oscillation problem caused by the "one-size-fits-all" resource adjustment in traditional technologies. The correct order of resource scheduling is ensured through the resource dependency graph and topological sorting, guaranteeing business continuity. A complete closed-loop feedback mechanism is established, realizing the continuous self-optimization of the resource allocation system, breaking through the limitation of traditional static preset parameters, enabling the system to continuously learn and improve the resource allocation strategy, and adapting to the complex and changeable network environment.

[0048] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, systems, and units can refer to the corresponding processes in the foregoing method embodiments and will not be repeated here.

[0049] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a network node resource dynamic allocation device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0050] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for dynamically allocating network node resources, characterized in that Including: Collecting multi-dimensional resource usage data of network nodes and performing preprocessing to obtain a standardized resource monitoring matrix and a topological entropy weight matrix; Performing non-linear topological perception calculation based on the standardized resource monitoring matrix and the topological entropy weight matrix to obtain a resource demand prediction matrix; Making a segmented progressive allocation decision on network node resources based on the resource demand prediction matrix to obtain a first resource allocation decision matrix; Performing chaotic perturbation fine-tuning on the first resource allocation decision matrix to obtain a second resource allocation decision matrix and a resource locking schedule; Performing multi-level scheduling of network node resources based on the second resource allocation decision matrix and the resource locking schedule to generate a resource execution result log and performance monitoring data.

2. The network node resource dynamic allocation method according to claim 1, characterized in that, The collecting multi-dimensional resource usage data of network nodes and performing preprocessing to obtain a standardized resource monitoring matrix and a topological entropy weight matrix includes: Collecting computing resources, memory resources, storage resources, and bandwidth resources of network nodes at a fixed frequency to obtain multi-dimensional resource usage data; Performing standardized conversion on the multi-dimensional resource usage data to obtain a standardized resource monitoring matrix; Constructing a network topology relationship graph based on the connection relationship and communication situation between network nodes, and extracting node topology association data based on the network topology relationship graph; Performing weight calculation and normalization processing on the node topology association data to obtain a topological entropy weight matrix.

3. The network node resource dynamic allocation method according to claim 1, wherein The performing non-linear topological perception calculation based on the standardized resource monitoring matrix and the topological entropy weight matrix to obtain a resource demand prediction matrix includes: Constructing a three-dimensional resource demand tensor including time series, node identifiers, and resource dimensions based on the standardized resource monitoring matrix; Performing time series decomposition processing on the three-dimensional resource demand tensor to obtain multi-component time series data; Calculating the node resource fluctuation characteristics according to the residual component in the multi-component time series data to obtain a non-linear attenuation coefficient; Performing recursive convolution calculation based on the three-dimensional resource demand tensor and the non-linear attenuation coefficient to obtain an initial resource demand prediction value, and performing topological perception adjustment on the initial resource demand prediction value in combination with the topological entropy weight matrix to obtain a resource demand prediction matrix.

4. The network node resource dynamic allocation method according to claim 3, characterized in that The performing recursive convolution calculation based on the three-dimensional resource demand tensor and the non-linear attenuation coefficient to obtain an initial resource demand prediction value, and performing topological perception adjustment on the initial resource demand prediction value in combination with the topological entropy weight matrix to obtain a resource demand prediction matrix includes: Extracting the resource usage records of each network node at multiple historical time points from the three-dimensional resource demand tensor, and assigning different weights to the data at different time points according to the non-linear attenuation coefficient to obtain time-weighted resource data; Based on the topological entropy weight matrix, screening out a set of nodes with a topological association degree higher than a preset threshold with the target node to obtain an influencing node group of the target node; Accumulating and summing the product of the time-weighted resource data and the topological entropy weight of each network node in the influencing node group, and performing recursive convolution operation to obtain an initial resource demand prediction value; Calculate the change rate of the topological entropy weight matrix within a continuous time window, and process it in combination with the logarithmic gain function to obtain a topologically aware perturbation factor after dynamic adjustment; Perform a multiplication operation on the initial resource demand prediction value, the topologically aware perturbation factor after dynamic adjustment, and the sign of the resource demand change trend to obtain a resource demand prediction matrix.

5. The method for dynamically allocating network node resources according to claim 1, characterized in that Based on the resource demand prediction matrix, make a segmented progressive allocation decision on the network node resources to obtain a first resource allocation decision matrix, including: Define the minimum divisible unit of resources for various network resources, and calculate the difference between the total currently available resources and the total allocated resources to obtain the allocable resource amount; Perform a weighted calculation on each network node based on the business criticality, topological centrality, and priority to obtain the node importance weight; Divide the resource allocation process into multiple decision segments, and calculate the resource amount to be allocated in each segment according to the number of decision segments and the allocable resource amount to obtain the decision segment resource allocation plan; Calculate the resource value coefficient and decision score of each network node in each decision segment according to the node importance weight and the resource demand prediction matrix, and allocate resources in descending order of the decision score based on the decision segment resource allocation plan to obtain the resource allocation results of each segment; Calculate the judgment reversal threshold according to the number of network nodes, and re-evaluate the previous segment allocation decision when the continuous decision score deviation of the node exceeds the judgment reversal threshold, and finally integrate the allocation results of each decision segment to obtain the first resource allocation decision matrix.

6. The method for dynamically allocating network node resources according to claim 5, wherein The calculating the judgment reversal threshold according to the number of network nodes, and re-evaluating the previous segment allocation decision when the continuous decision score deviation of the node exceeds the judgment reversal threshold, and finally integrating the allocation results of each decision segment to obtain the first resource allocation decision matrix, including: Calculate the judgment reversal threshold of the current decision segment according to the number of network nodes and the decision segment serial number; Calculate the decision score deviation value between two consecutive decision segments of each network node, and compare the decision score deviation value with the judgment reversal threshold to obtain a list of candidate nodes that need to be re-evaluated; Recalculate the resource value coefficient and decision score of the previous segment for the network nodes in the list of candidate nodes that need to be re-evaluated according to the resource demand prediction matrix to obtain the corrected previous segment allocation plan; Calculate the resource locking period of each network node based on the topologically aware perturbation factor and the judgment reversal threshold to obtain a resource allocation stability guarantee strategy; Integrate the resource allocation results of each decision segment according to the network node and resource dimensions, and process the resource allocation of the core nodes according to the resource locking period to obtain the first resource allocation decision matrix.

7. The method for dynamically allocating network node resources according to claim 1, wherein Perform chaotic perturbation fine-tuning on the first resource allocation decision matrix to obtain a second resource allocation decision matrix and a resource locking schedule, including: Import the historical performance data of the network nodes to construct a performance evaluation function, and calculate the deviation between the current node performance and the expected performance to obtain the node performance feedback closed-loop value; Dynamically adjust the node perturbation coefficient of the chaotic perturbation function according to the magnitude of the node performance feedback closed-loop value; Calculate a resource adjustment factor using the closed-loop node performance feedback value and the node perturbation coefficient, compare the resource adjustment factor with a preset deviation threshold to determine whether resource adjustment is required, and obtain a resource adjustment trigger signal; Perform a resource ratio adjustment on the first resource allocation decision matrix according to the resource adjustment trigger signal, and introduce upper and lower limit constraints to ensure that the adjustment range is within the target range, obtaining a second resource allocation decision matrix; Calculate the resource elastic recovery period of each network node based on the closed-loop node performance feedback value, and generate a corresponding resource locking schedule.

8. The network node resource dynamic allocation method according to claim 7, characterized in that, The step of calculating a resource adjustment factor using the closed-loop node performance feedback value and the node perturbation coefficient, comparing the resource adjustment factor with a preset deviation threshold to determine whether resource adjustment is required, and obtaining a resource adjustment trigger signal includes: Perform a division operation on the closed-loop node performance feedback value and the node perturbation coefficient to obtain a performance perturbation ratio, and perform a non-linear transformation on the performance perturbation ratio using a hyperbolic tangent function to obtain a basic deviation value; Multiply the basic deviation value by a preset adjustment intensity parameter to obtain a resource adjustment factor, where the adjustment intensity parameter is a fixed constant; Calculate the absolute value of the resource adjustment factor, and compare the absolute value with the preset deviation threshold. When the absolute value is greater than the preset deviation threshold, generate a positive resource adjustment trigger signal, otherwise generate a negative resource adjustment trigger signal; Determine whether to adjust the first resource allocation decision matrix based on the status of the resource adjustment trigger signal. When a positive resource adjustment trigger signal is received, perform a resource adjustment calculation on the corresponding elements of the resource adjustment factor and the first resource allocation decision matrix. When a negative resource adjustment trigger signal is received, keep the corresponding elements of the first resource allocation decision matrix unchanged.

9. The method for dynamically allocating network node resources according to claim 1, wherein The step of performing multi-level scheduling of network node resources based on the second resource allocation decision matrix and the resource locking schedule, generating a resource execution result log and performance monitoring data includes: Divide network nodes into a critical node group, a standard node group, and an elastic node group according to the importance of services, and calculate the difference value between the current actual resource configuration of each network node and the target configuration in the second resource allocation decision matrix; Based on the ratio of the difference value to the current actual resource configuration, and in combination with the critical node group, the standard node group, and the elastic node group, configure a differentiated execution strategy for each network node; Construct a resource dependency graph based on the dependency relationships between different resource dimensions among network nodes, perform a topological sort on the resource dependency graph to obtain a resource scheduling execution order, and perform a resource adjustment operation according to the differentiated execution strategy, the resource scheduling execution order, and the resource locking schedule; During the resource adjustment execution process, monitor the performance indicators of network nodes in real time. When it is detected that the performance degradation exceeds a preset deterioration threshold, abort the current adjustment operation and roll back to the previous stable state, add the corresponding network nodes to a temporary protection list, and record the entire execution process to generate a resource execution result log and performance monitoring data.

10. A dynamic allocation system for network node resources, characterized in that For implementing the network node resource dynamic allocation method described in any one of claims 1-9, the network node resource dynamic allocation system includes: An acquisition module, configured to acquire multi-dimensional resource usage data of network nodes and perform preprocessing to obtain a standardized resource monitoring matrix and a topological entropy weight matrix; A calculation module, configured to perform non-linear topological perception calculation based on the standardized resource monitoring matrix and the topological entropy weight matrix to obtain a resource demand prediction matrix; An allocation decision module, configured to perform a segmented progressive allocation decision on network node resources based on the resource demand prediction matrix to obtain a first resource allocation decision matrix; A fine-tuning module, configured to perform chaotic perturbation fine-tuning on the first resource allocation decision matrix to obtain a second resource allocation decision matrix and a resource locking schedule; A multi-level scheduling module, configured to perform multi-level scheduling of network node resources based on the second resource allocation decision matrix and the resource locking schedule to generate a resource execution result log and performance monitoring data.

Images