Intelligent management method and system of cloud assets

Abstract

This invention discloses an intelligent management method and system for cloud assets, relating to the field of data management technology. The method includes: defining cloud asset state characteristics for data collection; forming state vectors based on the state characteristics; calculating the total outer product matrix through density matrix mapping; calculating feature traces and determining the density matrix based on the quantized values ​​of the overall state characteristics; setting a random observation matrix and combining it with the density matrix to determine random observation values; analyzing the matrix degrees of freedom to determine the number of observations; constructing a total set corresponding to the number of observations; and optimizing and reconstructing the density matrix through sparse regularization. The method of this invention, through the definition of the state characteristic set and density matrix mapping, introduces a symmetric matrix calculation method, which can extend the representation from local cloud asset state to global asset state characteristics. By calculating the total outer product matrix and introducing feature traces, the energy expression of the global state is simplified.

Description

Technical Field

[0001] This invention relates to the field of data management technology, and in particular to an intelligent management method and system for cloud assets. Background Technology

[0002] With the continuous development and large-scale application of cloud computing technology, various forms of cloud assets (such as virtual machines, containers, storage resources, etc.) have become important carriers for the informatization and digital operation of modern enterprises. Through the characteristics of elastic computing, on-demand allocation and automated scheduling, cloud assets have greatly improved resource utilization efficiency and business flexibility. With the rapid growth of task volume, the increasing complexity of multi-dimensional resource interaction and the increasing demand for dynamic task allocation in the cloud environment, higher requirements have been put forward for intelligent management methods and systems of cloud assets. At present, the management methods for cloud assets are mainly based on centralized and rule-driven management frameworks.

[0003] However, current methods typically use a single resource dimension (such as computing power or storage capacity) as an optimization reference, failing to effectively integrate the state characteristics of multiple categories such as computing resources, storage resources, and network resources. This results in low global modeling accuracy and a lack of quantitative analysis of cloud asset collaboration relationships. They neglect the modeling of collaboration relationships between nodes or only use simple correlation indicators, achieving only local optimization of the state of a single node. Furthermore, they lack the ability to collect and intelligently adjust the real-time feedback on task allocation effects, which can easily lead to problems such as redundant resource scheduling and low utilization. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent management method and system for cloud assets. Current methods typically use a single resource dimension as an optimization reference, failing to effectively integrate the state characteristics of multiple categories such as computing resources, storage resources, and network resources. This results in low global modeling accuracy, a lack of quantitative analysis of cloud asset collaboration relationships, neglect of collaboration relationship modeling between nodes, or the use of only simple correlation indicators, achieving only local optimization of single node states. Furthermore, the lack of real-time feedback collection and intelligent adjustment capabilities for task allocation effects easily leads to problems such as redundant resource scheduling and low utilization.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an intelligent management method for cloud assets, comprising:

[0008] Define cloud asset state characteristics for data collection, form state vectors based on state characteristics, perform density matrix mapping to calculate the total outer product matrix, calculate feature traces and determine density matrix based on the quantized values ​​of the overall state of state characteristics, set random observation matrix and combine with density matrix to determine random observation values, analyze matrix degrees of freedom to determine the number of observations, construct the total set of corresponding observations, and optimize and reconstruct density matrix through sparse regularization.

[0009] Based on the off-diagonal elements of the optimized and reconstructed density matrix, collaborative relationship modeling is performed. A task requirement set for each cloud asset is defined. The task priority is calculated based on the task requirement data of the cloud assets, and the communication cost is calculated. A global scheduling objective function is defined based on the comprehensive task execution time. Dynamic planning of task requirements is performed. The task requirements are divided into stages based on the division objective. The incremental cost brought by the newly allocated task requirements in each stage is calculated. The optimal system state value of the node is updated, and the optimal allocation scheme is output.

[0010] Collect feedback data on task allocation schemes and execution results, dynamically adjust the priority weights of cloud assets by combining the diagonal elements of the optimized and reconstructed density matrix, update communication costs, and reconstruct the objective function for dynamic programming;

[0011] Construct a resource state subgraph, use a graph convolutional network (GCN) to update node features, define cross-subgraph association weights between nodes, perform cross-subgraph feature updates, and calculate abnormal state distances to mark graph nodes as anomalies.

[0012] As a preferred embodiment of the intelligent management method for cloud assets described in this invention, the steps include: calculating feature traces and determining a density matrix based on the quantized values ​​of the overall state characteristics; setting a random observation matrix and combining it with the density matrix to determine random observation values; analyzing the matrix degrees of freedom to determine the number of observations; constructing a total set corresponding to the number of observations; and optimizing and reconstructing the density matrix through sparse regularization.

[0013] The state feature set is defined as the state vector of each cloud asset for data recording, and the density matrix is ​​mapped for each cloud asset state vector. The symmetric matrix is ​​calculated by transposing the state vector, and the global feature correlation is superimposed to calculate the total outer product matrix.

[0014] The quantized value of the overall state, considering the state characteristics based on the total outer product matrix, is used to calculate the feature trace and the density matrix;

[0015] Set the basic dimension of the observation matrix to be consistent with the number of state features of the asset nodes, initialize a random matrix whose matrix elements satisfy the standard independent normal distribution, perform QR decomposition on the random matrix to obtain an orthogonal matrix Q and an upper triangular matrix Ru, extract the diagonal elements of the orthogonal matrix, adjust the direction uniform distribution, and use the adjusted orthogonal matrix Q as the random observation matrix that satisfies the Haar random distribution.

[0016] The random observation values ​​are obtained by projecting the density matrix ρ onto the random observation matrix.

[0017] The total number of asset node state features in the density matrix is ​​used to analyze the matrix degrees of freedom, considering that the density matrix is ​​a symmetric matrix.

[0018] The matrix degrees of freedom are used as the observation threshold, and the number of observations is set according to the observation threshold. A total set of observations corresponding to the number of observations is constructed, including the observation matrix and the corresponding observation value for the corresponding number of observations. The objective function is defined by sparse regularization to optimize and reconstruct the density matrix.

[0019] The gradient descent method is used to update the optimization reconstruction density matrix, and the constraint correction of the optimization reconstruction density matrix in each iteration is defined, including that the optimization reconstruction density matrix satisfies symmetry, normalization constraint and positive semidefiniteness. The cross-entropy loss function is selected to calculate the computational loss of two iterations. The convergence threshold is determined based on historical experience. If the computational loss is less than or equal to the convergence threshold, the iteration is stopped and the optimization reconstruction density matrix of the current iteration is output.

[0020] As a preferred embodiment of the intelligent management method for cloud assets described in this invention, the comprehensive task execution time defines a global scheduling objective function, performs dynamic planning of task requirements, defines a partitioning objective to divide task requirements into stages, calculates the incremental cost of newly allocated task requirements in each stage, updates the optimal system state value of the node, and outputs the optimal allocation scheme, including...

[0021] The collaborative relationship between the two corresponding cloud asset state vectors is modeled using off-diagonal elements, and the collaborative cost value of the two cloud asset state vectors is defined based on the underlying communication distance.

[0022] Define the set of task requirements for each cloud asset, and define the available resource capacity for each cloud asset;

[0023] The urgency value of a task is defined based on the maximum remaining time of the task requirements of cloud assets, and the resource utilization ratio of the task requirements of cloud assets is calculated. The task priority is then calculated by weighting the task quantity and resource allocation ratio.

[0024] Based on the collaborative cost value, the communication cost is defined, and the global scheduling objective function is defined by comprehensively considering the task execution time, resource waste and communication cost.

[0025] Perform dynamic planning for task requirements, define the goal of dynamic planning as minimizing resource waste, divide the task requirements into stages based on the total number of task requirements, and define the final task requirement division scheme after completing all stage divisions.

[0026] Based on the division of objectives, dynamic programming recursion is used to update the system state value by allocating task requirements at each stage until all task requirements are allocated and the optimal allocation scheme is output.

[0027] As a preferred embodiment of the intelligent management method for cloud assets described in this invention, the step of dynamically adjusting the priority weights of cloud assets by combining the diagonal elements of the optimized and reconstructed density matrix, updating communication costs, and reconstructing the objective function for dynamic programming includes:

[0028] Collect feedback data on task allocation schemes and execution results. The difference between the target time allocated according to task requirements and the actual completion time is used as the completion time deviation value, and the ratio of the actual resource type used by cloud assets to the initial resource quantity is used as the actual resource utilization rate.

[0029] Based on the optimized reconstruction density matrix, the diagonal elements are used as importance weights of the cloud asset state vector, and the priority weights of cloud assets are dynamically adjusted.

[0030] Adjust task priorities based on the delay in completion time provided by dynamic feedback;

[0031] The communication cost is updated based on dynamic priority weights, and the objective function is reconstructed by comprehensively adjusting task priorities and dynamic priority weights.

[0032] Based on the reconstructed objective function, a new optimal allocation scheme is obtained by performing dynamic programming again.

[0033] As a preferred embodiment of the intelligent management method for cloud assets described in this invention, the step of constructing a resource state subgraph, updating node features using a graph convolutional network (GCN), defining cross-subgraph association weights between nodes, and performing cross-subgraph feature updates includes:

[0034] Construct a resource state subgraph based on the updated optimal allocation scheme, map the information of cloud asset state vectors into the graph structure, clarify the initial values ​​of node characteristics and adjacency relationships, use different cloud asset state vectors as graph nodes and determine the node set, and define node edges based on the interaction of business needs between nodes.

[0035] The node characteristics are defined using the ratio of the data communication bandwidth usage between nodes to the maximum available bandwidth between nodes as the edge weight. These characteristics include the allocated CPU resources, allocated memory resources, current network resource utilization, amount of unused network resources, and execution feedback time of the task requirements of the corresponding cloud assets.

[0036] The graph convolutional network (GCN) is used to update node features and perform iterative propagation of node features. Information from different subgraphs is fused by combining an attention mechanism, and cross-subgraph association weights between nodes are defined to perform cross-subgraph feature updates.

[0037] As a preferred embodiment of the intelligent management method for cloud assets according to the present invention, the step of calculating the distance to abnormal states for graph node anomaly marking includes:

[0038] Based on the mean of historical node feature data as a reference feature, the Euclidean distance between the node features after cross-subgraph integration and the reference feature is calculated as the distance to the abnormal state.

[0039] An anomaly detection threshold is set based on historical experience. If the distance to the anomaly is greater than or equal to the anomaly detection threshold, the node is judged to be abnormal and marked as abnormal.

[0040] As a preferred embodiment of the intelligent management method for cloud assets according to the present invention, the step of defining cloud asset status characteristics and collecting data includes:

[0041] Characterize cloud assets by defining a set of state characteristics, including computing resources, storage resources, and network resources.

[0042] Secondly, the present invention provides an intelligent management system for cloud assets, comprising,

[0043] Data acquisition module: Defines cloud asset state characteristics, collects task requirements and state vectors, performs density matrix mapping and feature trace calculation, and finally completes density matrix initialization;

[0044] Random observation optimization module: Calculates random observation values ​​using the random observation matrix and density matrix, optimizes and reconstructs the density matrix based on matrix degrees of freedom and sparsity regularization, and generates an optimized density matrix;

[0045] Task requirement priority module: Calculates task priority from cloud asset task requirements, and constructs a global scheduling objective function by combining communication costs and task execution time;

[0046] Feedback Adjustment Module: Collects dynamic data through task allocation feedback, adjusts cloud asset priority weights and communication costs based on the diagonal elements of the optimized and reconstructed density matrix, and re-optimizes the objective function to perform dynamic programming;

[0047] Resource state modeling module: Constructs a resource state subgraph, updates node features using a graph convolutional network, and fuses cross-subgraph features;

[0048] Anomaly detection module: Based on the state features of nodes and cross-subgraph features, calculate the distance of abnormal states and mark the graph nodes of the resource state subgraph as abnormal.

[0049] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the intelligent management method for cloud assets as described in the first aspect of the present invention.

[0050] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent management method for cloud assets as described in the first aspect of the present invention.

[0051] The beneficial effects of this invention are as follows: By defining the state feature set and mapping the density matrix, and introducing a symmetric matrix calculation method, it is possible to extend the representation of the characteristics of the cloud asset state from the local cloud asset state to the global asset state. By calculating the total outer product matrix and introducing the feature trace, the energy expression of the global state is simplified. By defining the task requirement set and describing the available resource capacity, it not only distinguishes the requirements of computing resources, memory resources and network resources, but also defines a multi-dimensional operable data structure for tasks and nodes by combining resource occupancy rate and remaining capacity. By reconstructing the objective function, the reconstruction makes the scheduling decision highly flexible, and can take into account the diverse performance requirements in complex task distribution environments, thereby improving the overall adaptability of the scheduling system. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart illustrating the intelligent management method for cloud assets in Example 1.

[0054] Figure 2 This is a schematic diagram of the intelligent management system for cloud assets in Example 1. Detailed Implementation

[0055] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0056] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0057] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0058] Example 1, referring to Figures 1 to 2 This is the first embodiment of the present invention, which provides an intelligent management method for cloud assets, including the following steps:

[0059] S1. Define cloud asset state characteristics for data collection, form state vectors based on state characteristics, perform density matrix mapping to calculate the total outer product matrix, calculate feature traces and determine density matrix based on the quantized value of the overall state of state characteristics, set random observation matrix and combine with density matrix to determine random observation values, analyze matrix degrees of freedom to determine the number of observations, construct the total set of corresponding observations, and optimize and reconstruct density matrix through sparse regularization.

[0060] Preferably, data collection is performed by defining cloud asset status characteristics, including:

[0061] Characterize cloud assets by defining a set of state characteristics, including computing resources (CPU utilization, memory usage, average response time), storage resources (disk available space, I / O throughput, and failure rate), and network resources (bandwidth utilization, latency, and packet loss rate).

[0062] Furthermore, based on the quantized values ​​of the overall state characteristics, feature traces are calculated and the density matrix is ​​determined. A random observation matrix is ​​set and combined with the density matrix to determine random observations. The matrix degrees of freedom are analyzed to determine the number of observations. A total set corresponding to the number of observations is constructed, and the density matrix is ​​reconstructed through sparse regularization optimization.

[0063] The state feature set is defined as the state vector of each cloud asset for data recording. Each cloud asset state vector is then mapped using a density matrix. A symmetric matrix is ​​calculated using the transpose of the state vector, and global feature correlations are superimposed to calculate the total outer product matrix, represented as:

[0064]

[0065] in, Let N represent the total outer product matrix, and let N represent the total number of cloud assets in the same group. This represents the state vector of the i-th cloud asset. express transpose, The calculation yields a symmetric matrix of the i-th cloud asset state vector, which can be represented as: ;

[0066] Based on the total outer product matrix, the quantized value of the overall state is used to calculate the feature trace and the density matrix, which is expressed as:

[0067]

[0068]

[0069]

[0070] in, This represents the cumulative energy of all asset nodes at the j-th state characteristic, and is a quantitative representation of the overall state of the j-th state characteristic. This represents the j-th state feature of the i-th cloud asset state vector. The trace of the total outer product matrix is ​​the sum of the diagonal elements of the matrix, and F represents the total number of state features of the asset nodes. Let represent the density matrix, which represents the globally correlated states of cloud assets. The density matrix includes diagonal and off-diagonal elements, where the diagonal elements describe the contribution of a single cloud asset state vector, and the off-diagonal elements represent the degree of correlation between two cloud asset state vectors. It is represented as:

[0071]

[0072]

[0073]

[0074] in, Let represent the diagonal elements, and let represent the contribution of the i-th cloud asset state vector. This represents the cumulative energy value of all asset nodes at the i-th state characteristic. This represents the cumulative energy of all asset nodes at the i-th and j-th state characteristics. and These represent the k-th state features of the i-th and j-th cloud asset state vectors, respectively. Represents off-diagonal elements, indicating the degree of correlation between the i-th and j-th cloud asset state vectors;

[0075] Set the basic dimension of the observation matrix to be consistent with the number of state features of the asset nodes. Initialize a random matrix whose elements satisfy a standard independent normal distribution (mean 0, variance 1). Perform QR decomposition on the random matrix to obtain an orthogonal matrix Q and an upper triangular matrix Ru. Extract the diagonal elements of the orthogonal matrix and adjust their orientation to be uniformly distributed (so that the determinant is...). The adjusted orthogonal matrix Q is used as the random observation matrix that satisfies the Haar random distribution;

[0076] Projecting the density matrix ρ onto the random observation matrix yields the random observation values, expressed as:

[0077]

[0078] in, Let i represent the i-th random observation. Let the matrix represent the i-th random observation.

[0079] The total number of asset node state features based on the density matrix, considering that the density matrix is ​​a symmetric matrix, is expressed as:

[0080]

[0081] in, Represents the degrees of freedom of a matrix;

[0082] Using the matrix degrees of freedom as the observation threshold and setting the number of observations based on the threshold, a total set corresponding to the number of observations is constructed, including the observation matrix and corresponding observation values ​​for the corresponding number of observations. The objective function is then defined using sparse regularization to optimize and reconstruct the density matrix, expressed as:

[0083]

[0084] in, This represents the optimized reconstruction density matrix, where K represents the number of observations. Used to maximize the information integrity of the density matrix ρ in all random directions This represents a sparse regularization term, used to limit the complexity of the solution while ensuring the sparsity of the density matrix and reducing redundant information.

[0085] The gradient descent method is used to update the optimized reconstruction density matrix, and the constraint correction of the optimized reconstruction density matrix in each iteration is defined, including that the optimized reconstruction density matrix satisfies symmetry, normalization constraint and semi-positive definiteness (eigenvalue decomposition of the density matrix is ​​performed, and negative eigenvalues ​​are truncated to zero for correction). The cross-entropy loss function is selected to calculate the computational loss of two iterations. The convergence threshold is determined based on historical experience. If the computational loss is less than or equal to the convergence threshold, the iteration is stopped and the optimized reconstruction density matrix of the current iteration is output.

[0086] By defining the state feature set and mapping the density matrix, and introducing the symmetric matrix calculation method, we can extend the representation of the characteristics of the cloud asset state from the local cloud asset state to the global asset state. By calculating the total outer product matrix and introducing the feature trace, we simplify the energy expression of the global state. By combining the feature trace definition with the global state feature quantification value, we can directly compress the data of the large-scale cloud asset state vector into a compact mathematical quantity, which can significantly reduce the computational complexity. The combination of feature trace and energy accumulation provides a more comprehensive feature energy capture method and avoids the problem of unbalanced analysis of the contribution of different features.

[0087] By projecting the density matrix based on the random observation matrix, the key challenge of high-dimensional density matrices in information acquisition is effectively solved. By using the Haar random observation matrix, the density matrix is ​​compressed from high dimension to low dimension while maintaining the integrity of global information. This provides an efficient observation method for density matrices and reduces the bias in the sampling process.

[0088] By constructing an optimization objective based on random observations and matrix degrees of freedom calculation, the convergence efficiency of the optimization and reconstruction process is effectively controlled. On the basis of quantification of the globally relevant states, the dimension is reduced by random projection, and the observation threshold is estimated according to the matrix degrees of freedom to set the necessary number of observations. This optimization process ensures the theoretical completeness of the observations from the perspective of degrees of freedom and avoids the reconstruction instability caused by too many or too few observations. Furthermore, the sparse regularization design eliminates useless strong correlations in the data, further improving the interpretability of the results. The final output result, through the optimization of sparse regularization and cross-entropy, maximizes the preservation of global information and the sparse optimization efficiency of the reconstruction density matrix from a technical perspective.

[0089] By combining the effects of the above techniques, a cross-dimensional fusion from local state features to global feature descriptions was achieved, and the problem of high-dimensional information compression was solved using a random observation matrix.

[0090] S2, based on the off-diagonal elements of the optimized and reconstructed density matrix, performs collaborative relationship modeling, defines the task requirement set for each cloud asset, calculates task priority based on cloud asset task requirement data, calculates communication cost, defines a global scheduling objective function based on the comprehensive task execution time, performs dynamic planning of task requirements, defines a partitioning objective to divide task requirements into stages, calculates the incremental cost brought by the newly allocated task requirements in each stage, updates the optimal system state value of the node, and outputs the optimal allocation scheme.

[0091] Preferably, the global scheduling objective function is defined based on the comprehensive task execution time. Dynamic planning of task requirements is performed, a partitioning objective is defined to divide task requirements into stages, the incremental cost of newly allocated task requirements in each stage is calculated, the optimal system state value of each node is updated, and the optimal allocation scheme is output, including...

[0092] The collaborative relationship between the two corresponding cloud asset state vectors is modeled using off-diagonal elements. The collaborative cost value between the two cloud asset state vectors is defined based on the underlying communication distance, and is expressed as follows:

[0093]

[0094] in, This represents the collaborative cost value of the i-th and j-th cloud asset state vectors. This represents the off-diagonal elements of the i-th and j-th cloud asset state vectors in the optimized reconstructed density matrix. The underlying communication distance (such as network latency or bandwidth overhead) represents the state vectors of cloud assets i and j.

[0095] Define the set of task requirements for each cloud asset, including the required values ​​of the three main resources: computing, memory, and network. Also define the available capacity of each cloud asset, including the available capacity of the three main resources: computing, memory, and network.

[0096] The urgency value of a task is defined based on the maximum remaining time required for cloud asset tasks, and the resource utilization ratio of cloud asset task requirements is calculated. The task priority is then calculated by weighting the task quantity and resource allocation ratio, as follows:

[0097]

[0098] in, This indicates the priority of the j-th task requirement. The importance of a task as it gets closer to its deadline can be indicated by setting the reciprocal of the number of tasks. This represents the maximum remaining time required to complete the task, determined by the difference between the task deadline and the task submission time. This indicates the resource allocation ratio, specifically the proportion of total task resources to total node resources. This represents the resource utilization ratio required by the j-th task, which is the resource utilization ratio compared to the resource utilization ratio required by all cloud asset tasks.

[0099] Communication cost is defined based on the collaborative cost value and is expressed as follows:

[0100]

[0101] in, Let R represent the communication cost, where R represents all cloud assets at the current moment, and N represents the total number of cloud assets.

[0102] Taking into account task execution time, resource waste, and communication costs, a global scheduling objective function is defined as follows:

[0103]

[0104] in, This represents the scheduling objective function value, where M represents the total number of tasks. Indicates task requirements The execution time, which is the specific time required after the three main resources of computation, memory, and network required for the task are allocated to the node, is summed to obtain the total completion time. The weight representing resource waste is determined by the ratio of the total amount of unused resources across all cloud assets to the total amount of resources available. This represents the total available resources of all cloud assets, and is used as the total resource waste value. The weight representing communication costs is determined by the latency of communication splicing between cloud assets;

[0105] Dynamic programming is performed to determine task requirements, with the objective of minimizing resource waste. The task requirements are divided into stages based on the total number of requirements. The final task requirement allocation scheme, obtained after completing all stages, is represented as follows:

[0106]

[0107]

[0108] in, This represents the optimal system state value at stage k. This means that optimization is performed based on the optimal value in the (k-1)th stage, and then on the incremental cost in the current stage. This indicates the need for newly assigned tasks at the current stage. The resulting incremental costs This represents the task allocation variable, and the task requirement. Whether it has been assigned to the cloud asset state vector i, where 1 indicates that it has been assigned and 0 indicates that it has not been assigned. Indicates task Task requirements This represents the resource waste value of cloud asset state vector i. This represents the total available resources in the current cloud asset state vector i. Represents the off-diagonal elements of cloud asset state vectors i and j. Indicates task Whether it is distributed between cloud asset state vectors i and j, where it is 1 if it is distributed simultaneously, and 0 otherwise;

[0109] in This indicates the time cost of completing the task. This indicates the cost of wasting resources. This represents the communication coordination cost when tasks are allocated across nodes.

[0110] Based on the division of objectives, dynamic programming recursion is used to update the system state value by allocating task requirements at each stage until all task requirements are allocated and the optimal allocation scheme is output.

[0111] By modeling the collaborative relationship between the state vectors of two cloud assets using off-diagonal elements, the correlation between cloud assets is effectively captured. The communication cost between individual cloud assets is further quantified based on the underlying physical distance (such as network latency or bandwidth overhead). This makes the collaborative relationship not only depend on the logical characteristics of resource distribution, but also has a greater impact on the actual cross-node scheduling in combination with the communication cost. This allows the scheduling model to avoid the high overhead caused by long-distance communication, thereby prioritizing the path with lower cost and closer correlation in the collaborative relationship between cloud assets and improving task allocation efficiency.

[0112] By defining the task requirement set and describing the available resource capacity, it not only distinguishes the requirements for computing resources, memory resources and network resources, but also defines a multi-dimensional operable data structure for tasks and nodes by combining resource occupancy rate and remaining capacity. By defining task urgency value and task priority, it reasonably distinguishes the importance of tasks and execution order, and by dynamically calculating task priority, it guides the scheduling process to prioritize high-risk tasks.

[0113] The communication cost is defined by calculating the results of the collaborative relationship, which further refines the cross-node allocation cost of the task. The communication cost combines off-diagonal elements and communication distance, so that the resource collaboration between nodes not only focuses on the logical collaborative relationship, but also includes the physical communication overhead.

[0114] By dynamically dividing the system state values ​​into objectives and stages, a feasible recursive optimization strategy is provided for the allocation of complex task requirements. This strategy can be adjusted at any time according to the remaining task requirements and resource status, avoiding static planning problems in global scheduling and making the allocation scheme dynamic and flexible.

[0115] By combining the phased outputs of recursive dynamic programming with the final display of the optimal allocation scheme, the data from all parts of the entire technology chain are integrated. The task requirement allocation and resource status update are processed recursively. From the early definition of the priority of urgent tasks to the trade-off of cross-node communication costs, each stage can accurately capture changes in resource utilization and guide the optimization calculation of the next stage. The final output of the optimal allocation scheme is not only based on the time and resource balance of task requirements, but also considers the comprehensive trade-off of communication costs, achieving a combination of resource efficiency and scheduling flexibility.

[0116] S3 collects feedback data on task allocation schemes and execution effects, dynamically adjusts the priority weights of cloud assets by combining the diagonal elements of the optimized and reconstructed density matrix, updates communication costs, and reconstructs the objective function for dynamic programming.

[0117] Preferably, by dynamically adjusting the priority weights of cloud assets in conjunction with optimizing and reconstructing the diagonal elements of the density matrix, updating communication costs, and reconstructing the objective function for dynamic programming, including:

[0118] Collect feedback data on task allocation schemes and execution results. The difference between the target time allocated according to task requirements and the actual completion time is used as the completion time deviation value, and the ratio of the actual resource type used by cloud assets to the initial resource quantity is used as the actual resource utilization rate.

[0119] Based on the optimized reconstruction density matrix, the diagonal elements are used as importance weights for the cloud asset state vector, and the priority weights of cloud assets are dynamically adjusted, as follows:

[0120]

[0121] in, This represents the dynamic priority weight of the cloud asset state vector i. This represents the diagonal elements of the optimized reconstructed density matrix. This indicates the degree of impact of resource utilization on the priority weight of cloud assets, and is determined by the ratio of average utilization to peak utilization. This represents the resource utilization rate of cloud asset state vector i. This indicates the impact of the number of allocation failures on the priority weight of cloud assets, and is determined by the ratio of the total number of task allocation failures to the total number of task requests. Indicates the number of times task assignment failed;

[0122] Based on the delay in completion time provided by dynamic feedback, the task priority is adjusted, as shown below:

[0123]

[0124] in, This indicates an adjustment to the task priority. The weight representing the impact of task delays on priority is determined by the ratio of the sum of the actual completion times of all tasks to the sum of the target completion times. This indicates a delay in completion time;

[0125] Based on updating the communication cost using dynamic priority weights, and by comprehensively adjusting task priorities and dynamic priority weights, the objective function is reconstructed as follows:

[0126]

[0127]

[0128] in, This indicates the reconstruction of the objective function value. Indicates the cost of updating communication. Represents the dynamic priority weight of the cloud asset state vector j;

[0129] Based on the reconstructed objective function, a new optimal allocation scheme is obtained by performing dynamic programming again.

[0130] By collecting feedback data on task allocation schemes and execution results, the system can obtain real-time dynamic information on task allocation, including deviations in task completion time and statistics on actual resource utilization. Through deviations in completion time, execution efficiency issues in task allocation can be identified in a timely manner, while feedback on resource utilization reflects the balance of cloud asset resource usage.

[0131] By using the diagonal elements of the optimized and reconstructed density matrix as the basic weights of the cloud asset state vector, and combining this with dynamic feedback to adjust the dynamic priority weights, it is possible to prioritize cloud asset nodes with high resource utilization efficiency and low allocation failure rates for task allocation. This avoids scheduling decisions being influenced by a single factor, improving the overall flexibility and reliability of resource allocation.

[0132] By combining dynamic feedback to adjust task priority based on completion time delay, the accuracy deviation of resource scheduling caused by execution completion time delay errors is avoided. This allows task scheduling to maintain a sensitive response to the actual scenario, enabling more accurate identification and priority processing of critical tasks, and preventing high-priority tasks from being delayed for a long time.

[0133] By updating communication costs based on dynamic priority weights, the actual overhead of tasks in cross-asset node allocation is redefined, which increases the participation of high-quality cloud asset nodes with high utilization rates, while reasonably avoiding dependence on nodes that are far away and have high communication costs.

[0134] By reconstructing the objective function, task completion time delay, communication cost, and resource utilization are integrated and calculated in a dynamic weight model, significantly enhancing the adaptability of the global scheduling objective. This reconstruction makes scheduling decisions highly flexible, enabling the system to meet diverse performance requirements in complex task distribution environments and improving the overall adaptability of the scheduling system.

[0135] S4. Construct a resource state subgraph, use a graph convolutional network (GCN) to update node features, define cross-subgraph association weights between nodes, perform cross-subgraph feature updates, and calculate abnormal state distances to mark graph nodes as abnormal.

[0136] Preferably, a resource state subgraph is constructed, node features are updated using a graph convolutional network (GCN), cross-subgraph association weights between nodes are defined, and cross-subgraph feature updates are performed, including...

[0137] Construct a resource state subgraph based on the updated optimal allocation scheme (subgraphs for different resource categories can be constructed based on computing resources, storage resources, and network resources respectively). Map the information of cloud asset state vectors into the graph structure, clarify the initial values ​​of node features and adjacency relationships, use different cloud asset state vectors as graph nodes and determine the node set, and define node edges based on the interaction of business needs between nodes.

[0138] The node characteristics are defined using the ratio of the data communication bandwidth usage between nodes to the maximum available bandwidth between nodes as the edge weight. These characteristics include the allocated CPU resources, allocated memory resources, current network resource utilization, amount of unused network resources, and execution feedback time of the task requirements of the corresponding cloud assets.

[0139] Graph Convolutional Network (GCN) is used to update node features and iteratively propagate these features. An attention mechanism is then used to fuse information from different subgraphs. Cross-subgraph association weights between nodes are defined for cross-subgraph feature updates, as shown below:

[0140]

[0141]

[0142] in, This represents the cross-subgraph association weight between subgraphs k and m. and Let i represent the node characteristics of node i in subgraphs k and m, respectively. Indicates the total number of subgraphs. This represents the node characteristics of node i after cross-subgraph integration.

[0143] By constructing resource state subgraphs based on updated optimal allocation schemes, different types of resources (such as compute, storage, and network resources) can be mapped to independent subgraph structures, enabling hierarchical modeling of cloud asset states.

[0144] By utilizing Graph Convolutional Networks (GCNs) to update node features and iteratively propagate node features, it is possible to integrate the direct features of each cloud asset node and the relevant characteristics of neighboring nodes, so that the feature update results can cover higher-order global information, rather than being limited to the observation of local states.

[0145] By fusing information from different subgraphs through an attention mechanism, collaborative optimization scheduling across resource categories is achieved. The attention mechanism incorporates the influence between different resource categories into the same modeling framework by defining cross-subgraph association weights between nodes.

[0146] By combining cross-subgraph attention weights with the feature propagation process updated by GCN, the final output node features not only contain their own dynamic state information, but also embed the collaborative relationship characteristics with other resource categories and their neighboring nodes. This allows the final scheduling model to have more diverse decision-making basis and to perform more refined resource allocation and optimization.

[0147] Furthermore, the distance to abnormal states is calculated for graph node anomaly marking, including:

[0148] Based on the mean of historical node feature data as a reference feature, the Euclidean distance between the node feature of node i after cross-subgraph integration and the reference feature is calculated as the abnormal state distance.

[0149] An anomaly detection threshold is set based on historical experience. If the distance to the anomaly is greater than or equal to the anomaly detection threshold, the node is judged to be abnormal and marked as abnormal.

[0150] This embodiment also provides an intelligent management system for cloud assets, including,

[0151] Data acquisition module: Defines cloud asset state characteristics, collects task requirements and state vectors, performs density matrix mapping and feature trace calculation, and finally completes density matrix initialization;

[0152] Random observation optimization module: Calculates random observation values ​​using the random observation matrix and density matrix, optimizes and reconstructs the density matrix based on matrix degrees of freedom and sparsity regularization, and generates an optimized density matrix;

[0153] Task requirement priority module: Calculates task priority from cloud asset task requirements, and constructs a global scheduling objective function by combining communication costs and task execution time;

[0154] Feedback Adjustment Module: Collects dynamic data through task allocation feedback, adjusts cloud asset priority weights and communication costs based on the diagonal elements of the optimized and reconstructed density matrix, and re-optimizes the objective function to perform dynamic programming;

[0155] Resource state modeling module: Constructs a resource state subgraph, updates node features using a graph convolutional network, and fuses cross-subgraph features;

[0156] Anomaly detection module: Based on the state features of nodes and cross-subgraph features, calculate the distance of abnormal states and mark the nodes of the resource state subgraph as abnormal.

[0157] This embodiment also provides a computer device applicable to the intelligent management method of cloud assets, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent management method of cloud assets as proposed in the above embodiment.

[0158] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0159] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the intelligent management method for cloud assets as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0160] In summary, this invention, through the definition of state feature sets and density matrix mapping, introduces a symmetric matrix calculation method to extend the representation of cloud asset state characteristics from local to global. By calculating the total outer product matrix and introducing feature traces, the energy expression of the global state is simplified. By defining the task requirement set and describing the available resource capacity, it not only distinguishes the requirements for computing resources, memory resources, and network resources, but also defines a multi-dimensional operable data structure for tasks and nodes by combining resource occupancy rate and remaining capacity. By reconstructing the objective function, the reconstruction makes scheduling decisions highly flexible, capable of accommodating diverse performance requirements in complex task distribution environments, and improving the overall adaptability of the scheduling system.

[0161] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent management of cloud assets, characterized in that, include: Define cloud asset state characteristics for data collection, form state vectors based on state characteristics, perform density matrix mapping to calculate the total outer product matrix, calculate feature traces and determine density matrix based on the quantized values ​​of the overall state of state characteristics, set random observation matrix and combine with density matrix to determine random observation values, analyze matrix degrees of freedom to determine the number of observations, construct the total set of corresponding observations, and optimize and reconstruct density matrix through sparse regularization. Based on the off-diagonal elements of the optimized and reconstructed density matrix, collaborative relationship modeling is performed. A task requirement set for each cloud asset is defined. The task priority is calculated based on the task requirement data of the cloud assets, and the communication cost is calculated. A global scheduling objective function is defined based on the comprehensive task execution time. Dynamic planning of task requirements is performed. The task requirements are divided into stages based on the division objective. The incremental cost brought by the newly allocated task requirements in each stage is calculated. The optimal system state value of the node is updated, and the optimal allocation scheme is output. Collect feedback data on task allocation schemes and execution results, dynamically adjust the priority weights of cloud assets by combining the diagonal elements of the optimized and reconstructed density matrix, update communication costs, and reconstruct the objective function for dynamic programming; Construct a resource state subgraph, use a graph convolutional network (GCN) to update node features, define cross-subgraph association weights between nodes, perform cross-subgraph feature updates, and calculate abnormal state distances to mark graph nodes as anomalies. 2.The intelligent management method of cloud assets according to claim 1, wherein: The process involves calculating feature traces and determining the density matrix based on the quantized values ​​of the overall state characteristics, setting a random observation matrix and combining it with the density matrix to determine random observations, analyzing the matrix degrees of freedom to determine the number of observations, constructing a total set corresponding to the number of observations, and optimizing and reconstructing the density matrix through sparse regularization. The state feature set is defined as the state vector of each cloud asset for data recording, and the density matrix is ​​mapped for each cloud asset state vector. The symmetric matrix is ​​calculated by transposing the state vector, and the global feature correlation is superimposed to calculate the total outer product matrix. The quantized value of the overall state, considering the state characteristics based on the total outer product matrix, is used to calculate the feature trace and the density matrix; Set the basic dimension of the observation matrix to be consistent with the number of state features of the asset nodes, initialize a random matrix whose matrix elements satisfy the standard independent normal distribution, perform QR decomposition on the random matrix to obtain an orthogonal matrix Q and an upper triangular matrix Ru, extract the diagonal elements of the orthogonal matrix, adjust the direction uniform distribution, and use the adjusted orthogonal matrix Q as the random observation matrix that satisfies the Haar random distribution. The random observation values ​​are obtained by projecting the density matrix ρ onto the random observation matrix. The degrees of freedom of the matrix are analyzed by the total number of asset node state features in the density matrix; The matrix degrees of freedom are used as the observation threshold, and the number of observations is set according to the observation threshold. A total set of observations corresponding to the number of observations is constructed, including the observation matrix and the corresponding observation value for the corresponding number of observations. The objective function is defined by sparse regularization to optimize and reconstruct the density matrix. The optimization reconstruction density matrix is updated using a gradient descent method, and an optimization reconstruction density matrix constraint correction of each iteration is defined, including that the optimization reconstruction density matrix satisfies symmetry, normalization constraint and semi-positive definiteness, and a cross-entropy loss function is selected to calculate the calculation loss of two iterations, a convergence threshold is determined according to historical experience, if the calculation loss is less than or equal to the convergence threshold, the iteration is stopped, and the optimization reconstruction density matrix of the current iteration is output. 3.The intelligent management method of cloud assets according to claim 2, characterized in that: The comprehensive task execution time defines a global scheduling objective function, performs dynamic planning of task requirements, defines a division target to divide the task requirements into stages, and calculates the incremental cost brought by the newly allocated task requirements of the stage, updates the optimal system state value of the node, and outputs the optimal allocation scheme, including, The non-diagonal elements are used to model the synergy relationship of the corresponding two cloud asset state vectors, and the synergy cost value of the two cloud asset state vectors is defined according to the bottom communication distance; The task requirement set of each cloud asset is defined, and the resource available capacity of each cloud asset is also defined; The task urgency value is defined according to the maximum remaining time of the cloud asset task requirement, and the resource occupation ratio of the cloud asset task requirement is counted, and the task priority is calculated by weighting according to the number of tasks and the proportion of resource allocation, respectively; The communication cost is defined based on the synergy cost value, and the global scheduling objective function is defined by comprehensively considering the task execution time, resource waste and communication cost; The dynamic planning of task requirements is performed, and the division target of the dynamic planning is defined as the minimum resource waste, the stage division is performed according to the total number of task requirements, and the final task requirement division scheme is obtained after all stage divisions are completed; Based on the division target, the dynamic planning is recursively performed, and the system state value is updated at each stage during task requirement allocation until all task requirements are allocated, and the optimal allocation scheme is output. 4.The method of claim 3, wherein: The priority weight of the cloud asset is dynamically adjusted based on the diagonal element of the optimization reconstruction density matrix, the communication cost is updated, and the target function is reconstructed for dynamic planning, including, The feedback data of the task allocation scheme and the execution effect are collected, wherein the difference between the allocation target time of the task requirement and the actual completion time is taken as the completion time deviation value, and the ratio of the actual resource amount used by the cloud asset to the initial resource amount is taken as the actual resource utilization rate; Based on the optimization reconstruction density matrix, the diagonal element is taken as the importance weight of the cloud asset state vector, and the priority weight of the cloud asset is dynamically adjusted; The task priority is adjusted in combination with the delay of the completion time of the dynamic feedback; The communication cost is updated based on the dynamic priority weight, and the target function is reconstructed according to the comprehensive adjustment of the task priority and the dynamic priority weight; Based on the reconstructed target function, the dynamic planning is performed again to obtain the updated optimal allocation scheme. 5.The intelligent management method of cloud assets according to claim 4, characterized in that: The resource state subgraph is constructed, the node features are updated using a graph convolution network (GCN), the cross-subgraph association weight between nodes is defined, and the cross-subgraph feature is updated, including, According to the updated optimal allocation scheme, a resource state subgraph is constructed, information of the cloud asset state vector is mapped into the graph structure, initial values of node features and adjacency relations are determined, different cloud asset state vectors are taken as graph nodes and a node set is determined, and edges between nodes are defined according to business demand interaction between nodes; According to the ratio of the data communication bandwidth usage between nodes to the maximum available bandwidth between nodes as the edge weight, the node features are defined to include the allocated CPU resources, the allocated memory resources, the current network resource utilization, the unused network resource amount and the task demand execution feedback time of the corresponding cloud asset; The graph convolution network (GCN) is used to update the node features, and the iterative propagation of the node features is performed, the information of different subgraphs is fused by combining the attention mechanism, the cross-subgraph association weight between nodes is defined, and the cross-subgraph feature update is performed. 6.The intelligent management method of cloud assets according to claim 5, characterized in that: The abnormal state distance is calculated to perform graph node abnormal marking, including, Based on the historical period collected node feature data, the mean value of the node feature data is calculated as the reference feature, the Euclidean distance between the integrated node features and the reference features is calculated as the abnormal state distance; And based on historical experience, an abnormal judgment threshold is set, if the abnormal state distance is greater than or equal to the abnormal judgment threshold, the node is judged to be abnormal, and an abnormal mark is performed. 7.The intelligent management method of cloud assets according to claim 6, characterized in that: The cloud asset state features are defined for data collection, including, The cloud asset is characterized, and a state feature set is defined, including computing resources, storage resources and network resources.

8. An intelligent management system of cloud assets based on the intelligent management method of cloud assets according to any one of claims 1-7, characterized in that: including, The data collection module defines the cloud asset state features, collects the task demand and the state vector, performs the density matrix mapping and the trace calculation, and finally completes the density matrix initialization; The random observation optimization module calculates the random observation value by using the random observation matrix and the density matrix, optimizes the reconstructed density matrix according to the matrix degrees of freedom and the sparse regularization, and generates the optimized density matrix; The task demand priority module calculates the task priority from the cloud asset task demand, and constructs a global scheduling objective function combining the communication cost and the task execution time; The feedback adjustment module collects dynamic data through task allocation feedback, adjusts the cloud asset priority weight and the communication cost based on the diagonal elements of the optimized reconstructed density matrix, and reoptimizes the objective function to perform dynamic planning; The resource state modeling module constructs a resource state subgraph, updates the node features by using the graph convolution network, and fuses the cross-subgraph features; The abnormality detection module calculates the abnormal state distance based on the state features of the nodes and the cross-subgraph feature update, and performs abnormal marking on the graph nodes of the resource state subgraph. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the intelligent management method of the cloud asset according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the intelligent management method of the cloud asset according to any one of claims 1-7.

Images