Cross-platform resource matching implementation method based on multi-dimensional label dynamic weight

Abstract

The invention relates to the field of cloud computing resource management and optimization, and discloses a cross-platform resource matching implementation method based on multi-dimensional label dynamic weight, which comprises the following steps: S1, constructing a multi-dimensional label system; s2, on the basis of the multi-dimensional label system, similarity calculation is carried out on resource features and demand features; s3, dynamically adjusting the contribution weight of each dimension feature in matching calculation based on historical matching data and user feedback; s4, constructing a resource relation network, carrying out data interaction based on a standardized interface, and carrying out multi-platform resource information fusion; and S5, calculating a resource matching degree based on the optimized matching weight, and performing resource recommendation or resource allocation according to the matching degree. According to the method and the system, cross-platform computing resources, storage resources, bandwidth resources and the like are associated and managed in a structured manner by constructing the resource relationship network, so that the visualization and intelligent optimization of the topological relationship between the resources are realized.

Description

Technical Field

[0001] The present invention relates to the field of cloud computing resource management and optimization, and particularly to a method for realizing cross-platform resource matching based on dynamic weights of multi-dimensional tags. Background Art

[0002] With the rapid development of information technology, cloud computing has become an indispensable infrastructure for modern enterprises and developers, providing services such as elastic computing, storage, and networking. Cloud computing helps enterprises reduce infrastructure investment costs by providing resources on demand and supports efficient data processing, storage, and computing tasks. However, despite the widespread application of cloud computing globally, enterprises still face many challenges when using cloud computing resources. Especially for multi-platform usage and intelligent matching of resources, the existing technologies have not been able to fully meet the growing needs of enterprises.

[0003] Cloud computing service providers (such as Amazon Web Services, Microsoft Azure, Google Cloud, etc.) offer diverse resource configurations and billing models, allowing users to select suitable computing resources according to their business needs. However, this rich resource configuration and service model also brings complexity in resource selection. Among numerous platforms, how to intelligently select the best resources according to different business needs (such as computing power, storage requirements, network bandwidth, cost control, etc.) has become a core issue for enterprises in cloud computing resource scheduling.

[0004] The existing technologies have adopted various methods in the management, scheduling, and optimization of cloud computing resources, but there are still several key problems, which limit the improvement of resource utilization efficiency and the flexibility and adaptability of enterprises in resource scheduling. Most of the existing resource scheduling systems determine resource selection and scheduling based on static rules or preset weights, and these methods cannot effectively cope with dynamic user needs, resource price fluctuations, and differences in resource characteristics between platforms. In addition, in a multi-platform environment, issues such as integration, standardization of resource information, and cross-platform data interaction also pose significant challenges to the existing technologies. Summary of the Invention

[0005] To make up for the above deficiencies, the present invention provides a method for realizing cross-platform resource matching based on dynamic weights of multi-dimensional tags, aiming to solve the problem of how to achieve intelligent matching, cross-platform integration, and efficient scheduling of cloud computing resources through a multi-dimensional tag system and a dynamic weight optimization algorithm in a multi-platform environment.

[0006] In a first aspect, the present invention provides the following technical solution, a method for realizing cross-platform resource matching based on dynamic weights of multi-dimensional tags, including the steps of: S1. Construct a multi-dimensional tag system; S2. Calculate the similarity between the characteristics of the resource and the demand characteristics based on the multi-dimensional label system; S3. Dynamically adjust the contribution weights of the characteristics of each dimension in the matching calculation based on historical matching data and user feedback; S4. Construct a resource relationship network, perform data interaction based on a standardized interface, and perform multi-platform resource information fusion; S5. Calculate the resource matching degree based on the optimized matching weights, and perform resource recommendation or resource allocation according to the matching degree.

[0007] Preferably, the multi-dimensional label system in S1 includes: Functional characteristics, used to describe the uses of resources, including computing resources, storage resources, bandwidth resources, and service resources; Performance characteristics, used to describe the operating capabilities of resources, including the number of CPU cores, memory size, and network rate; Cost characteristics, used to describe the resource price and billing method, including price, billing mode, and discount information; Geographical characteristics, used to describe the physical location of resources and network access conditions, including server location and network latency; Historical usage records, used to reflect the operating stability and utilization rate of resources, including resource utilization rate, average load, and failure rate.

[0008] Preferably, the similarity calculation in S2 includes: For numerical characteristics, calculate the similarity based on the normalized distance; For categorical characteristics, calculate the similarity based on the set similarity; The number of CPU cores, memory size, network rate, price, discount information, network latency, resource utilization rate, average load, and failure rate are numerical characteristics, and the resource type, billing mode, server location, computing resources, storage resources, bandwidth resources, and service resources are categorical characteristics.

[0009] Preferably, the training process of optimizing the matching weights by reinforcement learning in S3 includes: S31. Establish an optimization environment based on reinforcement learning, use the matching weights as state variables, the weight adjustment as action variables, and the matching success rate, resource utilization rate, and matching stability as reward functions; S32. Optimize the matching weights using the deep reinforcement learning method, and improve the training convergence based on the policy improvement method; S33. Train the reinforcement learning model through historical matching data and dynamically adjust the matching weights during the resource matching process.

[0010] Preferably, in S4, cross-platform data fusion performs data interaction through RESTful API or GraphQL.

[0011] In a second aspect, the present invention provides the following technical solution, a cross-platform resource matching system based on multi-dimensional label dynamic weights, the system includes: A multi-dimensional label management module for constructing a multi-dimensional label system; A similarity calculation module for calculating the matching degree between resources and requirements based on the multi-dimensional label system in the multi-dimensional label management module; A reinforcement learning optimization module for optimizing the matching weights based on the deep reinforcement learning method; A data fusion module for constructing a cross-platform data fusion model based on a resource relationship network and performing data interaction through a standardized interface; A resource matching module for calculating the resource matching degree and performing resource recommendation or resource allocation.

[0012] Preferably, the reinforcement learning optimization module includes: A data acquisition unit for obtaining matching data from the similarity calculation module; A training unit for training a deep reinforcement learning model based on the matching data and adopting a target network to stabilize the training process; An inference unit for generating optimized matching weights based on the training results and feeding them back to the resource matching module.

[0013] Preferably, the resource matching module calculates the matching degree based on the matching weights optimized by the reinforcement learning optimization module.

[0014] In a third aspect, the invention provides the following technical solution, a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the computer program, it implements the above-mentioned cross-platform resource matching implementation method based on multi-dimensional label dynamic weights.

[0015] In a fourth aspect, the present invention provides the following technical solution, a readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the above-mentioned cross-platform resource matching implementation method based on multi-dimensional label dynamic weights.

[0016] The present invention has the following beneficial effects: 1. In the present invention, by constructing a resource relationship network, computing resources, storage resources, bandwidth resources, etc. across platforms are associated and managed in a structured manner, realizing the visualization and intelligent optimization of the topological relationship between resources. Compared with the existing independent data management method, the present invention can automatically mine the potential correlation of resources through the relationship network, making cross-platform resource scheduling and optimization more intelligent and flexible, and improving the overall efficiency of resource management.

[0017] 2. In the present invention, by establishing a standardized resource description model and combining the normalization calculation of numerical features and the set similarity calculation of categorical features, a significant improvement in resource matching accuracy is achieved. Compared with the prior art, the present invention can not only process different types of resource features, but also ensure that the resource matching results are more in line with actual needs through multi-dimensional comprehensive calculation, improving the accuracy and applicability of cross-platform resource selection.

[0018] 3. In the present invention, through standardized interfaces and multi-platform data fusion, real-time synchronization and consistency management of cross-platform resource information are realized, thereby improving the efficiency and automation of resource scheduling. Different from the existing method that relies on manual adjustment or single-platform optimization, the present invention can achieve unified resource scheduling in a multi-cloud environment and optimize it in combination with historical usage data, improving resource utilization and reducing the overall operating cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flowchart of the method for realizing cross-platform resource matching based on multi-dimensional label dynamic weights proposed by the present invention; Figure 2 is an architecture diagram of the cross-platform resource matching system based on multi-dimensional label dynamic weights proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Apparently, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] Embodiment 1: Referring to Figure 1 , in the first embodiment of the present invention, the present invention provides a method for realizing cross-platform resource matching based on multi-dimensional label dynamic weights, including the steps: S1. Construct a multi-dimensional label system; Functional features, which are used to describe the usage of resources, including computing resources, storage resources, bandwidth resources, and service resources; Performance features, which are used to describe the operating capabilities of resources, including the number of CPU cores, memory size, and network rate; Cost features, which are used to describe the resource price and billing method, including price, billing mode, and discount information; Geographic features, which are used to describe the physical location of resources and network access conditions, including server location and network latency; Historical usage records, which are used to reflect the operating stability and utilization rate of resources, including resource utilization rate, average load, and failure rate.

[0022] Specifically, in the specific implementation process, first classify the resources into five dimensions: functional features, performance features, cost features, geographic features, and historical usage records. The parameters of each dimension are standardized to ensure quantitative analysis during the matching calculation.

[0023] Functional features are used to describe the usage of resources, such as computing resources, storage resources, bandwidth resources, and service resources. Different resources have different usage scenarios, and direct comparison may lead to matching errors. Therefore, in actual applications, first screen the functional features so that computing resources are only matched with computing resources, and storage resources are only matched with storage resources, thereby reducing invalid calculations and improving matching efficiency. For example, during the cloud computing scheduling process, if a task requires computing resources, the present invention will not include storage resources in the recommendation list, reducing the matching deviation from the source.

[0024] Performance features mainly measure the operating capabilities of resources, including the number of CPU cores, memory size, and network rate. The resource specifications of different cloud platforms may use different units and representation methods. For example, the number of CPU cores may be represented by physical cores, virtual cores, or computing units (vCPU), the memory size may use units such as GB and MB, and the network rate may use Mbps or Gbps.

[0025] In the present invention, all performance features are converted into standardized values and normalized, so that resources with different specifications can be compared under the same calculation framework. For example, the number of CPU cores uses the maximum-minimum normalization method to map all data to between 0 and 1, so that different specifications of CPU cores have the same weight during the matching calculation, improving the fairness of matching. In addition, the network rate also needs to be unified in units. For example, all rates are converted to Mbps to avoid calculation errors caused by inconsistent units.

[0026] The cost characteristics involve the pricing method of resources and the calculation of fees. Different cloud platforms adopt different billing methods, such as pay-as-you-go, monthly / yearly subscription models, spot instances, etc., which makes it difficult to directly compare costs. The present invention adopts a standardized pricing model to convert different billing methods into a unified hourly cost, enabling equivalent comparison of resource costs on different platforms.

[0027] For example, for a resource with annual billing, first calculate its total cost and convert it into an equivalent cost based on hourly billing to ensure fair comparison during matching calculations.

[0028] The geographical characteristics mainly involve the physical location of resources and the network access situation, which affect the access latency of resources between different regions. Cloud computing resources are usually distributed in multiple data centers globally, and there may be significant differences in network latency between different regions, directly affecting the availability of resources. The present invention adopts a standardized geographical tag system to map the data center locations of different cloud platforms and calculates the network latency in combination with historical access data, enabling the selection of resources with lower network latency with priority during matching calculations. For example, during resource scheduling, if the user is located in the Asian region, resources in Asian data centers are preferentially matched to reduce access latency and improve computing efficiency.

[0029] Historical usage records are used to measure the stability of resources, including resource utilization rate, average load, failure rate, etc. Traditional resource matching methods usually only focus on the current performance parameters of resources and ignore the historical operating conditions, resulting in potentially unstable matching results. The present invention fully considers the stability of resources in the matching calculation by introducing historical usage records.

[0030] For example, during the matching calculation, if the number of CPU cores and price of a certain resource meet the requirements, but its historical failure rate is high, the present invention will reduce the recommended priority of this resource to reduce the risk of resource unavailability. In addition, resource utilization rate is also an important factor. Resources with high utilization rates usually mean high demand, while resources with low utilization rates may be more flexible. Therefore, the present invention introduces a resource utilization rate weighting mechanism during the matching calculation, enabling the recommended resources to not only meet the computing requirements but also be optimized in terms of load balancing.

[0031] S2. Calculate the similarity between the characteristics of the resources and the demand characteristics based on the multi-dimensional tag system; The similarity calculation includes: for numerical features, calculating the similarity based on the normalized distance; for categorical features, calculating the similarity based on the set similarity. The number of CPU cores, memory size, network rate, price, discount information, network latency, resource utilization rate, average load, and failure rate are numerical features, while resource type, billing mode, server location, computing resources, storage resources, bandwidth resources, and service resources are categorical features.

[0032] Specifically, in cross-platform resource management, resources from different sources have significant differences in computing power, price, location, and stability, and direct comparison often fails to obtain the optimal match. This solution ensures the accuracy of the match by using different similarity calculation methods for numerical features and categorical features respectively.

[0033] The similarity of numerical features is calculated using the normalized distance. Numerical features include the number of CPU cores, memory size, network rate, price, discount information, network latency, resource utilization rate, average load, and failure rate. These features are continuous, with different units and ranges, and direct comparison may lead to errors. To solve this problem, the present invention uses a normalization method to convert all numerical features to the same numerical interval to ensure comparability during calculation. The normalization uses the min-max normalization formula: ; where, is the original feature value, and are the minimum and maximum values of this feature among all resources respectively. For example, if the number of CPU cores of a computing resource is 16 cores, and assuming the range of the global number of CPU cores is from 2 to 128 cores, then the normalized value: ; Similarly, features such as network rate and price can be standardized according to this method, so that all numerical features can be converted to between 0 and 1 for easy calculation.

[0034] In the matching calculation, the similarity of numerical features is calculated using the Euclidean distance: ; where, and represent the normalized values of two resources on the i-th numerical feature respectively. For example, the feature vectors of two resources are: ; Then the similarity calculation is as follows: ; The closer the similarity is to 0, the more similar the two are. To ensure a reasonable matching result, this paper adopts a weight adjustment mechanism to adjust the weights according to the importance of the features, making the key features have a greater impact on the matching result. For example: ; Among them, is the weight of the feature, ensuring that the weight method can be flexibly adjusted in different application scenarios.

[0035] The categorical features are roughly measured by set similarity. The categorical features include resource type, billing mode, server location, computing resources, storage resources, bandwidth resources, and server resources. These features are of discrete type and cannot be directly numerically calculated. Therefore, the present invention uses the Jaccard similarity to measure the category matching degree. The Jaccard similarity calculation formula is as follows: ; Among them, represents the number of intersection elements of A and B, represents the number of union elements of A and B. For example: the categorical features of resource A are {computing resources, pay-as-you-go, us-west-1}, and the categorical features of resource B are {computing resources, prepaid, us-west-1}; The intersection of the two is {computing resources, us-west-1}, and the union is {computing resources, pay-as-you-go, prepaid, us-west-1}. Therefore, the Jaccard similarity calculation is as follows: ; The closer the Jaccard similarity is to 1, the more similar the two are. The present invention further optimizes the calculation process. For categories with strong hierarchy (such as server location), geographical distance mapping is used. For example, the actual distance between two data centers is obtained using geocoding and combined with similarity calculation to make the resource matching between different regions more accurate.

[0036] Final matching score calculation. Since the numerical features and categorical features use different methods to calculate similarity, comprehensive calculation is required to determine the final matching score. The present invention uses a weighted fusion method: ; Among them, and are adjustment coefficients used to control the influence of numerical and categorical features on the matching result. The higher the similarity of the numerical features, the smaller, so the larger, indicating a higher similarity. The similarity of the categorical features is determined by the Jaccard similarity. The higher the matching degree, the higher the score. For example, assuming that the calculated similarity of the numerical features is 0.8 and the similarity of the categorical features is 0.5, if , , then the final matching score is: ; The higher the matching score, the more suitable the resource is for the requirement. The present invention can adjust and the value of

[0037] S3. Dynamically adjust the contribution weights of each dimensional feature in the matching calculation based on historical matching data and user feedback; the training process of optimizing the matching weights by reinforcement learning includes: S31. Establish an optimization environment based on reinforcement learning, take the matching weights as state variables, take the weight adjustment as action variables, and take the matching success rate, resource utilization rate, and matching stability as reward functions; S32. Use the deep reinforcement learning method to optimize the matching weights, and improve the training convergence based on the policy improvement method; S33. Train the reinforcement learning model through historical matching data, and dynamically adjust the matching weights during the resource matching process.

[0038] Specifically, first construct an optimization environment for reinforcement learning. In this environment, the matching weights are used as state variables, the weight adjustment is used as action variables, and the matching success rate, resource utilization rate, and matching stability are used as reward functions. Specifically, the state space includes the weight values of each feature dimension, the action space is the weight adjustment strategy selected according to the current state, and the reward is given according to the performance of the matching process. In actual operation, the matching success rate reflects the effectiveness of the matching, the resource utilization rate reflects the rationality of resource allocation, and the matching stability measures the reliability of resource matching during long-term operation.

[0039] For example, assume that in a certain resource matching, the matching result calculated using the current weights leads to a low resource utilization rate and poor matching stability. Then this result will receive a low reward, and the corresponding optimization process will tend to adjust the feature weights to improve the matching effect. The specific reward function can be expressed as: ; where SuccessRate is the success rate of the matching at the t-th step, Utilization is the resource utilization rate, Stability is the stability of the matching process, is a regulation coefficient that controls the weights of different factors in the final reward.

[0040] Optimize the matching weights by deep reinforcement learning. Next, use the deep reinforcement learning method to optimize the matching weights. In the present invention, the deep Q-network (DQN) is used to train and optimize the matching weights. DQN combines Q-learning and deep neural networks, and trains a neural network to predict the value of each action in each state, and adjusts the matching weights according to these values.

[0041] During the training process, the model continuously adjusts the weights according to historical matching data to optimize the matching effect. Specifically, through the feedback rewards, the deep reinforcement learning algorithm adjusts the weight update rule for each decision. The goal of the learning process is to maximize the cumulative reward, thereby improving the success rate and stability in resource matching.

[0042] During the deep reinforcement learning process, a policy improvement method is also used to improve the convergence of training. The core idea of policy improvement is that by continuously iterating and optimizing the policy, the model can find the optimal combination of matching weights in a shorter time.

[0043] In the present invention, the policy gradient method can adjust the policy parameters by calculating the gradient of the reward function, thereby gradually approaching the optimal solution in each step of the optimization process.

[0044] Through repeated training and policy updates, the model gradually learns how to dynamically adjust the weights of features, making the role of each feature in the matching more in line with the actual needs. For example, the features of certain resource types may be more important during a specific time period, while some are less relevant, and reinforcement learning will adjust the weights according to these changes. Through this dynamic learning, the resource matching system can adapt to environmental changes and continuously optimize the matching effect.

[0045] The reinforcement learning model is trained with historical matching data. The historical data includes previous matching results, user feedback, and resource utilization situations, etc. By analyzing these data, the reinforcement learning model can identify which features have the greatest impact on the matching results and adjust the weights of each feature in the matching process accordingly.

[0046] For example, historical data shows that in certain resource scenarios, the performance of computing resources has a much greater impact on the matching results than the price, while in other scenarios, the price factor is more important. Based on these data, the reinforcement learning model can intelligently adjust the weights to make each resource matching more in line with the actual needs of users.

[0047] In addition, user feedback also plays a crucial role. By collecting user feedback after using resources, the model can learn which resource combinations are more popular with users and which feature combinations will lead to a poor experience, thereby further optimizing the matching strategy.

[0048] S4. Build a resource relationship network, perform data interaction based on a standardized interface, and conduct multi-platform resource information fusion; cross-platform data fusion performs data interaction through RESTful API or GraphQL.

[0049] Specifically, the core of the resource relationship network is to establish a graph structure that can represent the relationships between different resources. Nodes in this network represent various computing resources (such as virtual machines, storage, network bandwidth, etc.), while edges represent the association relationships between resources, such as dependency relationships, data flow relationships, or geographical distribution relationships. In this way, the resource management system can comprehensively understand the interconnectivity between resources and improve the accuracy of resource matching and scheduling.

[0050] These resources have different attributes on different platforms. For example, the computing resources on a certain platform may include vCPUs and memory, while the computing resources on another platform may be represented by the number of cores and the RAM capacity. For unified management, these resources need to be mapped to a standardized data structure, such as: ; Where: Represents the unique identifier of the resource; Represents the resource type, such as computing, storage, network; Specifies the resource source, such as AWS, Azure, Google Cloud; Records the detailed attributes of the resource, such as the number of CPUs, memory size, storage type; Records the association relationships of the resource, such as computing resources depending on storage resources, network resources connecting to computing resources; In this way, the resource relationship network can form a complete resource topology graph, which helps managers understand the overall situation of resources and consider all relevant factors during scheduling.

[0051] To achieve cross-platform data fusion, the present invention uses RESTful API or GraphQL for data interaction. RESTful API provides a standard interface based on the HTTP specification, while GraphQL allows the client to query data on demand, improving the flexibility of data acquisition.

[0052] The key to data fusion is how to uniformly transform the resource data of different platforms so that it can be managed in the same system. The present invention adopts a data standardization conversion layer to map the data structures of each platform to a unified model.

[0053] S5. Calculate the resource matching degree based on the optimized matching weight, and perform resource recommendation or resource allocation according to the matching degree.

[0054] Specifically, the resource matching degree measures the degree of fit between the resource and the demand. When calculating, based on the optimized matching weight, the features of different dimensions are comprehensively evaluated, and finally a matching score is obtained. The matching degree calculation formula is as follows: ; Wherein: is the matching degree between resource A and demand B; is the feature weight optimized by reinforcement learning; represents the similarity of the i-th feature.

[0055] Different types of features adopt different similarity calculation methods: for numerical features (such as the number of CPU cores, memory size, bandwidth), the normalized Euclidean distance is used for calculation: ; Wherein, is the maximum value of feature to normalize all feature values to between 0 and 1. For categorical features (such as resource type, billing mode), the Jaccard similarity is used: ; For example, if the storage type of resource A is {SSD, NVMe} and the storage type of resource B is {SSD, HDD}, then: ; After all features are calculated, weighted summation is performed according to the weight to obtain the final matching degree.

[0056] After calculating the matching degree, the system can recommend the optimal resources according to the matching degree ranking. When recommending, a threshold can be set to only recommend resources with a matching degree exceeding a certain value: ; Wherein, is the minimum matching degree requirement set by the system to ensure that the recommended resources meet the user's needs. In addition, the system can also be based on the Top-K recommendation method, that is, select the top resources with the highest matching degree: ; For example, assuming there are 10 candidate resources in the resource pool, after the system calculates the matching degree, it selects the top 3 with the highest ranking as the recommendation result.

[0057] In practical applications, after resource recommendation, resource allocation is still required. When allocating, the following strategies can be based on: Optimal matching allocation: directly select the resource with the highest matching degree and allocate it to the demander.

[0058] Load balancing allocation: among resources with similar matching degrees, preferentially select resources with lower loads to avoid resource overload.

[0059] Cost optimization allocation: for resources with the same matching degree, preferentially select those with lower prices to optimize the usage cost.

[0060] Fault tolerance allocation: If the failure probability of certain resources is relatively high recently, the system will reduce the matching weight of these resources and decrease the probability of their allocation.

[0061] Suppose there are three candidate resources in the resource pool , , , and their matching degrees are 0.85, 0.80, and 0.78 respectively. If the optimal matching strategy is adopted, then will be selected; if the cost optimization strategy is adopted, then among and , the resource with a lower price will be selected.

[0062] Dynamic adjustment and feedback optimization. The present invention not only calculates the resource matching degree during the initial matching, but also makes dynamic adjustments according to the actual usage situation. For example, after resource matching, the system will monitor its stability, performance, user feedback, etc. If it is found that a certain resource performs poorly after matching (such as high failure rate, low performance), the system will reduce the future matching weight of this resource, so that it will be preferentially excluded in subsequent matching. The adjustment formula is as follows: ; where feedback) is the user feedback factor. If the feedback of the resource is poor, then this factor will be less than 1, reducing the weight of this resource in subsequent matching.

[0063] For example, for a certain resource A, in the past 10 matches, it was abandoned by users due to performance problems 3 times. Then the system will automatically adjust the weight of this resource to reduce its priority in future matching.

[0064] Embodiment 2: Referring to Figure 2 , in the second embodiment of the present invention, the present invention provides a cross-platform resource matching system based on multi-dimensional label dynamic weights. The system includes: A multi-dimensional label management module for constructing a multi-dimensional label system; A similarity calculation module for calculating the matching degree between resources and requirements based on the multi-dimensional label system in the multi-dimensional label management module; A reinforcement learning optimization module for optimizing the matching weight based on the deep reinforcement learning method; A data fusion module for constructing a cross-platform data fusion model based on the resource relationship network and performing data interaction through a standardized interface; A resource matching module for calculating the resource matching degree and performing resource recommendation or resource allocation.

[0065] Specifically, we first build a multidimensional resource tagging system, defining multiple tags for each resource. These tags include numerical tags (such as the number of CPU cores, memory size, bandwidth, storage type, and price) and categorical tags (such as resource type, billing model, and data center location). This tagging system not only ensures the standardization of resource information but also provides the foundational data structure for subsequent resource matching, similarity calculation, and optimization. For example, in a cloud computing environment, each compute instance (such as AWS EC2 or Azure VM) generates a set of tags based on its performance metrics, resource attributes, and other information. These tags describe the characteristics of that resource.

[0066] The matching degree between resources and requirements is calculated based on a multi-dimensional tagging system. The similarity calculation module uses the tag differences between each resource and the requirement to calculate the matching degree. For example, when an enterprise submits a requirement (e.g., 8-core CPU, 32GB of memory, NVMe storage), the system selects matching resources from the known resource pool and calculates the matching degree based on the difference between their tags and the required tags.

[0067] The reinforcement learning optimization module, based on deep reinforcement learning, automatically learns how to optimize resource matching weights. Traditional matching algorithms often rely on fixed weights, while reinforcement learning uses an iterative learning process to continuously adjust the impact of different labels on matching. For example, the system might initially set relatively even weights (e.g., 20% each for CPU, memory, and bandwidth). Based on historical data feedback (such as user-selected resource types, resource utilization, and task execution results), the reinforcement learning model will gradually adjust these weights to optimize the accuracy and adaptability of resource recommendations.

[0068] The data fusion module is used to build a cross-platform data fusion model based on the resource relationship network and implement data exchange through standardized interfaces. This module retrieves resource information by calling the APIs of various platforms (such as the AWS SDK and Azure SDK) and converts it into a unified data structure. Through standardized interfaces (such as RESTful APIs or GraphQL), the system integrates resource data across platforms, allowing users to uniformly query and manage resource data across different cloud platforms. For example, when an enterprise needs to obtain resources from both AWS and Azure, the system queries and returns unified resource information through standardized interfaces for further processing and matching.

[0069] The resource matching module is one of the core modules of the system, responsible for recommending or allocating resources based on the calculated matching degree. When the resource matching degree is obtained through the similarity calculation module and the reinforcement learning optimization module, the resource matching module performs the following tasks: Resource Recommendation: Based on the user's needs and resource matching degree, the system will recommend the optimal resources. For example, the system will recommend suitable cloud computing instances according to the enterprise's budget constraints, computing requirements, etc. If an enterprise needs an 8-core CPU, 32GB of memory, SSD storage, and a budget of no more than 3 yuan per hour, the system will calculate the matching degree of resources on each platform and return the resource with the highest matching degree.

[0070] Automatic Resource Allocation: If an enterprise selects the automatic allocation function, the resource matching module will automatically allocate resources according to the recommended resources and system rules (such as cost, performance, load, etc.), and notify the user of the allocated resource information. This process is automated, helping enterprises reduce manual operations and improve resource allocation efficiency.

[0071] Embodiment Three: In the third embodiment of the present invention, based on the same inventive concept, a computer-readable storage medium is proposed. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps of the cross-platform resource matching implementation method based on multi-dimensional label dynamic weights in the above embodiment.

[0072] Embodiment Four: In the fourth embodiment of the present invention, based on the same inventive concept, a computer device is proposed. The computer device includes: a processor and a memory; the processor and the memory communicate with each other; the memory is used to store instructions; the processor is used to execute the instructions in the memory and implement the cross-platform resource matching implementation method based on multi-dimensional label dynamic weights in the above embodiment.

[0073] It should be understood that each part of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following well-known technologies in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits with suitable combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0074] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for implementing cross-platform resource matching based on dynamic weights of multi-dimensional tags, characterized in that, Including the steps: S1. Construct a multi-dimensional label system; S2. Based on the multi-dimensional label system, calculate the similarity between the characteristics of the resource and the demand characteristics; S3. Based on historical matching data and user feedback, dynamically adjust the contribution weights of the characteristics of each dimension in the matching calculation; S4. Construct a resource relationship network, perform data interaction based on a standardized interface, and perform multi-platform resource information fusion; S5. Calculate the resource matching degree based on the optimized matching weights, and perform resource recommendation or resource allocation according to the matching degree.

2. The cross-platform resource matching implementation method based on the dynamic weight of multi-dimensional tags according to claim 1, wherein In the S1, the multi-dimensional label system includes: Function characteristics, used to describe the resource usage, including computing resources, storage resources, bandwidth resources, and service resources; Performance characteristics, used to describe the resource operation ability, including the number of CPU cores, memory size, and network rate; Cost characteristics, used to describe the resource price and billing method, including price, billing mode, and discount information; Geographical characteristics, used to describe the physical location of the resource and the network access situation, including server location and network latency; Historical usage records, used to reflect the operation stability and utilization rate of the resource, including resource utilization rate, average load, and failure rate.

3. The cross-platform resource matching implementation method based on the dynamic weight of multi-dimensional tags according to claim 1, characterized in that In the S2, the similarity calculation includes: For numerical features, calculate the similarity based on the normalized distance; For categorical features, calculate the similarity based on the set similarity; The number of CPU cores, memory size, network rate, price, discount information, network latency, resource utilization rate, average load, and failure rate are numerical features, and the resource type, billing mode, server location, computing resources, storage resources, bandwidth resources, and service resources are categorical features.

4. The cross-platform resource matching implementation method based on the dynamic weight of multi-dimensional tags according to claim 1, characterized in that, In the S3, the training process of optimizing the matching weights by reinforcement learning includes: S31. Establish an optimization environment based on reinforcement learning, use the matching weights as state variables, use the weight adjustment as action variables, and use the matching success rate, resource utilization rate, and matching stability as reward functions; S32. Use the deep reinforcement learning method to optimize the matching weights, and improve the training convergence based on the policy improvement method; S33. Train the reinforcement learning model through historical matching data, and dynamically adjust the matching weights during the resource matching process.

5. The cross-platform resource matching implementation method based on the dynamic weight of multi-dimensional tags according to claim 1, wherein In the S4, cross-platform data fusion performs data interaction through RESTful API or GraphQL.

6. Cross-platform resource matching system based on dynamic weights of multi-dimensional tags, characterized in that For the cross-platform resource matching implementation method based on multi-dimensional label dynamic weights according to any one of claims 1-5, the system includes: A multi-dimensional label management module, used to construct a multi-dimensional label system; A similarity calculation module, calculating the matching degree between the resource and the demand based on the multi-dimensional label system in the multi-dimensional label management module; A reinforcement learning optimization module, optimizing the matching weights based on the deep reinforcement learning method; A data fusion module, used to construct a cross-platform data fusion model based on the resource relationship network, and perform data interaction through a standardized interface; A resource matching module, used to calculate the resource matching degree, and perform resource recommendation or resource allocation.

7. The cross-platform resource matching system based on the dynamic weight of multi-dimensional tags according to claim 6, characterized in that, The reinforcement learning optimization module includes: A data acquisition unit, obtaining matching data from the similarity calculation module; A training unit, training a deep reinforcement learning model based on the matching data, and using a target network to stabilize the training process; The inference unit generates optimized matching weights based on the training results and feeds them back to the resource matching module.

8. The cross-platform resource matching system based on multi-dimensional label dynamic weights according to claim 7, wherein The resource matching module calculates the matching degree based on the matching weights optimized by the reinforcement learning optimization module.

9. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the cross-platform resource matching implementation method based on multi-dimensional label dynamic weights according to any one of claims 1 to 5.

10. A readable storage medium, characterized in that, A computer program is stored on the readable storage medium, and when the computer program is executed by the processor, it implements the cross-platform resource matching implementation method based on multi-dimensional label dynamic weights according to any one of claims 1 to 5.

Images