Heterogeneous resource automatic discovery and integrated configuration method for hybrid cloud architecture

Abstract

The invention discloses a hybrid cloud architecture-oriented heterogeneous resource automatic discovery and integrated configuration method, and relates to the technical field of integrated configuration. Comprising the following steps: automatically identifying and dynamically adjusting a resource detection frequency to capture new resources through an adaptive learning and dynamic optimization mechanism; a deep learning model is utilized to predict a resource state change trend, detection is initiated in advance, and real-time monitoring and active response are realized; a dynamic API adaptation strategy and a self-learning mechanism are introduced, and cross-platform API intelligent adaptation and calling success rate optimization are achieved; a self-adaptive event-driven detection and multi-modal feature fusion method is adopted, and the detection frequency is dynamically adjusted based on anomaly detection; through a multi-modal attribute fusion and attribute mapping and conversion method, a unified resource description model is constructed, and standardization and consistency of resource attributes are carried out. According to the method, the operation is ensured to accord with laws and regulations such as GDPR, HIPAA and the like, and the validity of the operation and the consistency of the international standard are ensured while efficient collaboration and dynamic load balancing of cross-cloud resources are realized.

Description

Technical Field

[0001] The present invention relates to the field of integrated configuration technology, and in particular to a method for automatic discovery and integrated configuration of heterogeneous resources oriented to a hybrid cloud architecture. Background Art

[0002] With the rapid development of hybrid cloud technology, enterprises' expectations for multi-cloud resource management systems have far exceeded traditional boundaries, seeking higher levels of automation, efficiency, and consistency. However, existing technologies have exposed a series of key defects in dealing with heterogeneous resource management in hybrid cloud environments. These challenges constitute a significant obstacle to promoting resource optimization in the cloud-native era, including:

[0003] (1) Limitations of automatic resource discovery.

[0004] APIs provided by different cloud service providers vary in functionality, format, and availability, making it difficult for automated tools to fully and accurately identify all resource types and their characteristics. In addition, API support for emerging or customized services may be limited, further limiting resource discovery capabilities.

[0005] Complex resource structures and multi-level nesting (e.g., multi-level virtualization, deep network configuration) require more sophisticated detection mechanisms. Existing tools often cannot effectively recursively detect and associate resources, and therefore may miss important components.

[0006] Resource status changes rapidly, but existing monitoring technologies are usually passive and cannot capture immediate changes in resources (such as creation, deletion, or migration) in real time, resulting in delayed resource management decisions.

[0007] (2) The complexity of resource allocation.

[0008] The configuration process still relies on a lot of manual intervention, including parameter settings and cross-cloud operations, which not only increases the possibility of errors, but also reduces the level of automation and work efficiency.

[0009] Due to the significant technical differences between cloud platforms (such as service models, API interfaces, identifier systems, etc.), general configuration strategies are difficult to apply directly and require customized adaptation, which is not only time-consuming but also costly.

[0010] The lack of a unified resource configuration model framework makes it difficult to achieve seamless migration of resource descriptions and configuration instructions in a multi-cloud environment, increasing the complexity of cross-platform configuration.

[0011] (3) Compatibility and interoperability challenges

[0012] Some cloud service providers tend to build closed ecosystems. Although this helps the consistency of internal services, it also limits interoperability with other cloud environments and hinders the flexible scheduling and management of resources.

[0013] Although some cross-cloud standards (such as TOSCA, AWS CloudFormation, OpenStack Heat, etc.) attempt to provide a common resource description and configuration framework, due to differences in the degree of adoption and implementation methods of various cloud service providers, these standards do not completely solve compatibility issues in actual applications. Summary of the invention

[0014] In view of the problems existing in the above background technology, the present invention is proposed.

[0015] Therefore, the problem to be solved by the present invention is to propose a method for automatic discovery and integrated configuration of heterogeneous resources for a hybrid cloud architecture.

[0016] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0017] On the first aspect, an embodiment of the present invention provides a method for automatic discovery and integrated configuration of heterogeneous resources for a hybrid cloud architecture, which includes automatically identifying and dynamically adjusting the resource detection frequency to capture new resources through adaptive learning and dynamic optimization mechanisms; using a deep learning model to predict the trend of resource status changes, initiating detection in advance, and realizing real-time monitoring and active response; introducing a dynamic API adaptation strategy and a self-learning mechanism to realize intelligent adaptation and call success rate optimization of cross-platform APIs; using an adaptive event-driven detection and multimodal feature fusion method to dynamically adjust the detection frequency based on anomaly detection; and building a unified resource description model through multimodal attribute fusion and attribute mapping and conversion methods to standardize and ensure consistency of resource attributes.

[0018] As a preferred solution of the method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture described in the present invention, the prediction of resource status change trend using deep learning model includes the following contents: introducing a self-learning mechanism of dynamic API adaptation strategy, intelligent metadata parsing and mapping mechanism, and adaptive architecture of cross-platform API adaptation and prediction and optimization mechanism of API call success rate; wherein the adaptation module set is A={α1,α2,…,α n}; where each α i Tailored for the i-th class API to achieve cross-platform compatibility, the dynamic configuration of the PI adaptation layer is Among them, M i is the metadata of the ith API, and t is a time variable used to dynamically adjust the adaptation strategy.

[0019] As a preferred solution of the method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture described in the present invention, the adaptive event-driven detection and multimodal feature fusion include: introducing a multimodal attribute fusion method to comprehensively consider different attributes of resources from multiple dimensions; using a deep learning model to integrate resource attributes from different sources, in, is a multi-dimensional attribute fusion function; B i is a collection of attributes on different dimensions; set the mapping function to:

[0020] τM raw →M std

[0021] Among them, τ is the mapping function, M raw is a collection of resource descriptions of a specific cloud service provider, M std It is a universal model.

[0022] As a preferred solution of the method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture described in the present invention, the attribute mapping and conversion includes: the decision function is:

[0023] ΓΠ×C→A

[0024] Among them, Π is the set of strategies, C is the current resource configuration conditions, and A is the set of actions taken; the decision model is:

[0025]

[0026] Where U is the utility function, which measures the benefit of taking action a under condition c for policy π.

[0027] As a preferred solution of the method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture described in the present invention, the utility function U is used to measure the benefits of taking actions under given strategies and current conditions; the utility function U is as follows:

[0028] U(πc,a)=w cost ·u cost +w performance ·u performance +w reliability ·u reliability

[0029] Among them, w cost ,w performance , and w reliability are the weight coefficients of cost, performance and reliability respectively, and u cost ,u performance , and u reliability are the utility values ​​of cost, performance, and reliability, respectively.

[0030] As a preferred solution of the method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture described in the present invention, the set of strategies includes cost optimization strategy, performance optimization strategy, elasticity strategy and reliability strategy.

[0031] As a preferred solution of the method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture described in the present invention, the set of actions taken includes increasing resources, reducing resources, reallocating resources, optimizing configuration, and enabling spare resources.

[0032] In a second aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the method for automatic discovery and integrated configuration of heterogeneous resources for a hybrid cloud architecture as described in the first aspect of the present invention are implemented.

[0033] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the method for automatic discovery and integrated configuration of heterogeneous resources for a hybrid cloud architecture as described in the first aspect of the present invention are implemented.

[0034] The beneficial effects of the present invention are as follows: the method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture of the present invention proposes a number of innovative improvements to address the limitations of the existing technology. In terms of automatic resource discovery, by designing a resource detector and an API adaptation layer, combined with deep scanning and intelligent analysis technology, accurate identification and deep correlation mapping of resources are achieved, and real-time monitoring of resource status is ensured through a real-time detection mechanism. Specifically, the flexibility and accuracy of resource detection are improved through an adaptive learning mechanism, the intelligent identification and classification of resource types are achieved using deep learning technology, the dynamic identification function D_t is introduced to capture new resources, and deep feature extraction and adaptive optimization of resource data are achieved, thereby significantly improving the accuracy and timeliness of resource identification; in terms of resource configuration, a general resource description model and attribute mapping mechanism are constructed to achieve high automation and standardization of resource configuration. A policy-driven and efficient resource configuration engine is adopted, combined with an automated control process and an orchestration engine, to achieve codeless configuration and cross-platform consistency. The consistency and integrity of resource descriptions are improved by introducing multimodal attribute fusion methods, the consistency and portability of resource descriptions are ensured by using intelligent attribute mapping and conversion methods, and the resource description model can be dynamically adjusted according to changes in resource characteristics through adaptive attribute fusion and update methods. The attribute mapping method based on semantic understanding further improves the accuracy and consistency of resource descriptions. In terms of cross-cloud resource collaboration and interoperability, a unified monitoring and intelligent scheduling system is designed, and efficient cross-cloud resource collaboration and dynamic load balancing are achieved through adaptive algorithms and global resource views, and a compliance verification mechanism is introduced to ensure the legality of operations and consistency with international standards. The cross-platform compatibility is enhanced through the adaptive architecture of cross-platform API adaptation, the success rate of API calls is improved by using the prediction and optimization mechanism of API call success rate, and the efficiency and accuracy of resource status monitoring are improved by introducing an adaptive event-driven detection mechanism. The dynamic adjustment method based on anomaly detection can timely detect resource anomalies and respond quickly, and the intelligent feedback loop realizes the self-optimization and self-correction of the real-time detection mechanism. In addition, by implementing a role-based access control model (RBAC) and compliance verification functions, operations are ensured to comply with regulations such as GDPR and HIPAA, achieving efficient collaboration and dynamic load balancing across cloud resources while ensuring the legality of operations and consistency with international standards. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1 Design a flow chart for the automatic discovery system of heterogeneous resources.

[0037] Figure 2 Design a flow chart for the API adaptation layer.

[0038] Figure 3 Design a flow chart for real-time detection mechanism

[0039] Figure 4 Design a flow chart for the Uniform Resource Description Model.

[0040] Figure 5 Flowchart of engine configuration for efficient resources. DETAILED DESCRIPTION

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0042] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0043] 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 term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0044] Example 1

[0045] Reference Figure 1 to Figure 5 , which is the first embodiment of the present invention, and provides a method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture, including:

[0046] S1: Through adaptive learning and dynamic optimization mechanisms, it automatically identifies and dynamically adjusts the resource detection frequency to capture new resources.

[0047] like Figure 1 As shown, by introducing adaptive learning capabilities and adaptive optimization mechanisms, resource detectors can handle resource discovery and management more intelligently.

[0048] Set the initial resource detection frequency and identification parameters of various types of resources, including response time, bandwidth occupancy, processing speed, etc., to ensure that the initial detection process covers heterogeneous resources.

[0049] Specifically, the formula is expressed as:

[0050]

[0051] D represents the total data set of all known resources. D consists of resources provided by various cloud service providers, that is, through F i (R j ) to obtain resource data, n represents the number of different cloud service providers connected to the system, F i is a function that represents the API interface provided by the ith cloud service provider and is used to access and obtain resource data under the service provider. i , you can obtain resource R from the i-th cloud service provider j data, R j Represents the jth resource in the resource collection. Each resource R j All can be done through the corresponding F i Function to get its data.

[0052] Furthermore, in order to capture new resources that emerge over time, a dynamic identification function D is introduced. t , by combining with the existing resource dataset D and the API interface function Fi provided by the cloud service provider, new resources are discovered:

[0053]

[0054] in, It is a dynamic recognition function used to capture new resources that appear over time.

[0055] Traditional resource detectors may only rely on a predetermined time interval to check for new resources. In order to improve detection efficiency and accuracy, an adaptive learning mechanism is introduced that can automatically adjust the detection frequency and method based on historical data and current resource change trends. By using machine learning algorithms (such as ARIMA or LSTM models in time series analysis) to monitor the change trends of resources, and dynamically adjust the detection frequency according to these trends. For example, if the resource changes frequently, increase the detection frequency; otherwise, reduce the detection frequency. The formula is:

[0056] δ t =DynamicTimeIntervalAdjuster(t,ΔR,H(D))

[0057] Among them, δ t It represents the time interval after dynamic adjustment, t is the time variable, ΔR is the resource change, and H(D) is the historical data of resource change.

[0058] Define a deep feature extraction function F i , F iIt consists of three sub-functions, namely the depth scanning function D i 、Intelligent parsing function I i And the deep feature extraction function E i .

[0059]

[0060] in It is a deep scanning function, which is used to search and identify resources in depth; It is an intelligent parsing function used to parse the structure and attributes of resources; i It is a deep feature extraction function used to extract the key features of resources.

[0061] Deep feature extraction function F i The overall process includes:

[0062] F i (x) = D i (I i (E i (x)))

[0063] Input x passes through E i , get the eigenvector e; the eigenvector e passes through I i , get the parsed feature i; the parsed feature i is passed through D i , and get the result d after deep scanning.

[0064] Depth scan function D i The purpose of the ,is to deeply search and identify resources.,This involves recursively visiting different layers or components of a,resource to be able to gain a comprehensive understanding of the state of the,resource.

[0065] D i X→Y

[0066] Among them, X represents the input space, that is, the resources that need to be deeply searched; Y represents the output space, that is, the description of the resources after scanning, for example:

[0067]

[0068] Here, scan(x) is a scan operation on the leaf node x, and children(x) returns all the child nodes of node x.

[0069] Intelligent parsing function I i The task is to analyze the structure and attributes of resources. It involves using machine learning or rule engines to parse data, extract meaningful information, and define:

[0070] I i X→Z

[0071] Among them, X represents the input space, that is, from D i The resource description obtained; Z represents the output space, that is, the parsed resource attributes, example:

[0072]

[0073] Here, parse(x) is a parsing operation for simple data types, while parse(x_i) is a parsing operation for each element x in a complex data structure. i to parse.

[0074] Deep feature extraction function E i The main task of is to extract key features from resources. It involves machine learning models, especially deep learning models, to automatically discover and extract useful features, defined as:

[0075] E i X→W

[0076] Among them, X represents the input space, that is, the original resource; W represents the output space, that is, the extracted feature vector, for example:

[0077] E i (x) = model(x)

[0078] Here, model is a deep learning model, such as a convolutional neural network (CNN), a recurrent neural network (RNN), or other models suitable for a specific task. For example, in a computer vision task, the model may be a pre-trained CNN model for extracting features from an image.

[0079] For deep feature extraction function Rather than being limited to using fixed feature extraction models, an adaptive optimization mechanism is introduced to dynamically adjust feature extraction strategies based on the diversity of resource types. Transfer learning techniques are used to quickly adapt to different types of resource feature extraction. For example, when processing image resources, a pre-trained CNN model can be used; while for text resources, a pre-trained NLP model (such as BERT) can be used. The formula is:

[0080] E i =AdaptiveFeatureExtractor(E i ,T)

[0081] Among them, E i is a deep feature extraction function, T is a resource type identifier, and AdaptiveFeatureExtractor is an adaptive optimization function.

[0082] In view of the diversity and complexity of resource types provided by different cloud service providers, an intelligent identification and classification mechanism is proposed, which can automatically identify resource types and classify them effectively. The types of resources are automatically identified by using deep learning techniques (such as convolutional neural networks (CNN) or recurrent neural networks (RNN). In addition, clustering algorithms (such as K-means or DBSCAN) can also be used to classify resources according to their characteristics. The formula is expressed as:

[0083] C = AutoClassifier(R j ,M)

[0084] Among them, C is the resource category, R j is the resource data, and M is the pre-trained classification model.

[0085] In order to predict resource status changes in advance, a resource status prediction and active detection mechanism is proposed, which can detect resource status before it changes, thereby reducing the delay of resource management. A deep learning model (such as LSTM or GRU) is used to predict the state change trend of resources. Once it is predicted that the status may change, the system will actively initiate detection to capture these changes in advance. The formula is expressed as:

[0086] P = StatePredictor(R j ,t)

[0087] Where P is the predicted resource status, R j is the resource data, and t is the time variable.

[0088] S2: Use deep learning models to predict resource status change trends, initiate detection in advance, and achieve real-time monitoring and active response.

[0089] By introducing the self-learning mechanism of dynamic API adaptation strategy, the intelligent metadata parsing and mapping mechanism, the adaptive architecture of cross-platform API adaptation, and the prediction and optimization mechanism of API call success rate, the API adaptation layer can handle cross-platform API adaptation issues more intelligently.

[0090] Among them, the adaptation module set is A = {α1, α2, …, α n}; where each α i Tailored for the i-th class API to achieve cross-platform compatibility, the dynamic configuration of the PI adaptation layer is Among them, M i is the metadata of the ith API, and t is a time variable used to dynamically adjust the adaptation strategy.

[0091] Furthermore, the traditional API adaptation layer usually requires manual configuration of the adaptation strategy, which may not be flexible enough in the face of a rapidly changing cloud environment. In order to improve the adaptive ability of the adaptation layer, a self-learning mechanism based on machine learning is introduced. The API adaptation strategy is continuously optimized by using online learning algorithms (such as online gradient descent or stochastic gradient descent). By monitoring indicators such as the success rate and response time of the API call, the adaptation strategy is automatically adjusted to adapt to the changing cloud environment.

[0092] In order to improve the intelligence of the API adaptation layer, an intelligent metadata parsing and mapping mechanism is proposed, which can automatically parse the API metadata and automatically adjust the adaptation strategy according to these metadata. The metadata in the API document is parsed by using natural language processing technology (such as word embedding and semantic parsing), and then the adaptation strategy is automatically generated according to these metadata.

[0093] In addition, in order to achieve seamless adaptation of cross-platform APIs, an adaptive architecture is proposed, which can dynamically adjust the adaptation strategy according to different API interfaces to achieve the best adaptation effect, and use deep learning technology (such as the Transformer model) to analyze the structure of the API interface.

[0094] In order to improve the success rate of API calls, a prediction-based optimization mechanism is proposed, which can predict the possibility of API call failure and take measures to optimize it in advance. Use deep learning models (such as LSTM or GRU) to predict the success rate of AP calls and dynamically adjust the adaptation strategy based on the prediction results. For example, when it is predicted that an API call may fail, the system can automatically switch to an alternative API or take other optimization measures.

[0095] S3: Introduce dynamic API adaptation strategies and self-learning mechanisms to achieve intelligent adaptation of cross-platform APIs and optimize call success rates.

[0096] By introducing adaptive event-driven detection mechanism, multimodal feature fusion method, dynamic adjustment method based on anomaly detection, and intelligent feedback loop, the real-time detection mechanism can handle resource monitoring problems more intelligently.

[0097] Specifically, the event-driven model is: E(t) = δ t F(ΔR), where δ t is the time interval; F is the resource change function; ΔR is the resource change amount.

[0098] Perform dynamic adjustments to the real-time detection mechanism:

[0099] δ t =η(t,ΔR,H(D))

[0100] Among them, η is a dynamic time interval adjustment function, which is used to adjust the detection frequency according to the rate of change of resources; H is a high-dimensional feature extraction function, which is used to extract the pattern and trend of resource changes.

[0101] Description of high-dimensional feature extraction function H:

[0102] The goal of the high-dimensional feature extraction function H is to extract high-dimensional features that can reflect resource change patterns and trends from the resource dataset D. This is achieved by using a deep learning model (such as a convolutional neural network CNN or a recurrent neural network RNN). It is defined as: H:D→Z, where D represents the input space, i.e., the resource dataset; and Z represents the output space, i.e., the extracted feature vector.

[0103] The complete definition of the high-dimensional feature extraction function H: H(D) = MultiModalFeatureFusion(D time-series ,D logs ,D metrics )

[0104] Furthermore, traditional event-driven models usually rely on fixed detection intervals, which may lead to omissions or over-monitoring of resource status. To this end, an adaptive event-driven detection mechanism is introduced, which can dynamically adjust the detection interval according to the actual changes in resources. The detection interval adjustment strategy is trained using machine learning algorithms (such as reinforcement learning), so that the system can automatically adjust the detection frequency according to the speed and importance of resource changes. The formula is expressed as:

[0105] δ t =η(t,ΔR,H(D))=AdaptiveEventDriven(ΔR,H(D),t)

[0106] Among them, δ t is the time interval, ΔR is the resource change, H(D) is the high-dimensional feature extraction function, and t is the time variable.

[0107] In order to more comprehensively capture the trends and patterns of resource changes, a multimodal feature fusion method is proposed, which can extract features from data of different dimensions and sources. Deep learning models (such as convolutional neural networks (CNN) and recurrent neural networks (RNN)) are used to fuse multiple types of features, including but not limited to time series data, log data, and performance indicator data. The formula is expressed as:

[0108] H(D)=MultiModalFeatureFusion(D time-series ,D logs ,D metrics )

[0109] Among them, D time-seriesTime series data; D logs Log data; D metrics Performance indicator data.

[0110] Furthermore, in order to detect resource anomalies in time and respond quickly, a dynamic time interval adjustment method based on anomaly detection is invented. This method can adjust the detection frequency immediately when a potential anomaly is detected to ensure that abnormal events can be captured. By using statistical learning or deep learning techniques (such as isolation forests or autoencoders) to identify abnormal patterns in resource changes and dynamically adjust the detection frequency based on these patterns, the formula is expressed as:

[0111] δ t =η(t,ΔR,H(D))=AnomalyDetectionAdjustment(H(D),t)

[0112] In order to enable the real-time detection mechanism to self-optimize and self-correct, an intelligent feedback loop was invented. This loop is able to feed back to the system based on the detection results to optimize future detection strategies. By using reinforcement learning techniques (such as Q-learning or Actor-Critic methods) to build a closed-loop system, the system can evaluate the effectiveness of the detection results and adjust the detection strategy accordingly.

[0113] S4: Adopts adaptive event-driven detection and multimodal feature fusion methods to dynamically adjust the detection frequency based on anomaly detection to achieve accurate resource monitoring.

[0114] By introducing multimodal attribute fusion methods, intelligent attribute mapping and conversion methods, adaptive attribute fusion and update methods, and attribute mapping methods based on semantic understanding, the resource description model can handle resource description problems more intelligently.

[0115] Among them, the general resource description model is defined as:

[0116] M=(B,S,P)

[0117] In order to more comprehensively describe the characteristics of resources, a multimodal attribute fusion method is introduced, which can comprehensively consider the different attributes of resources from multiple dimensions. Deep learning models (such as multimodal attention mechanisms) are used to integrate resource attributes from different sources to ensure the consistency and completeness of resource descriptions. in, is a multi-dimensional attribute fusion function; B i It is a collection of attributes in different dimensions.

[0118] Set the mapping function to:

[0119] τM raw→M std

[0120] Among them, τ is the mapping function, M raw is a collection of resource descriptions of a specific cloud service provider, M std It is a universal model.

[0121] The task of the mapping function is to map the resource description set of a specific cloud service provider to a common model to ensure the consistency and portability of resource descriptions. To achieve this goal, deep learning models such as transfer learning techniques are used to accelerate the model training process, and reinforcement learning is used to optimize the mapping strategy.

[0122] The resource description set of a specific cloud service provider contains the original descriptions of various resources provided by the cloud service provider. These descriptions may contain different formats, naming conventions, and units, etc., and are compatible with resource descriptions of other cloud service providers through mapping and conversion.

[0123] Suppose that the resource descriptions include the following types: virtual machines, storage spaces, network interfaces, and security groups.

[0124] The common model is a standard resource description model that can be used across cloud service providers and provides a consistent resource description format. This model should contain all necessary attributes and configuration parameters to accurately describe the characteristics and status of resources. It contains the following types of resource descriptions: virtual machine, storage space Storagestd, network interface Networkstd, and security group SecurityGroupstd.

[0125] Use a deep learning model to implement the mapping function τ, such as using transfer learning technology to accelerate the model training process, and using reinforcement learning to optimize the mapping strategy, where the mapping function defines τ: τ(M raw )=DeepMap(T,M raw )

[0126] Among them, T is the deep learning model training set, M raw It is a collection of resource descriptions for a specific cloud service provider.

[0127] The dynamic properties are updated as:

[0128] M std =τ(M raw )+ΔM std

[0129] Among them, ΔM std It is an incremental update of the standardized model.

[0130] Furthermore, in order to enable the resource description model to dynamically adjust according to changes in resource characteristics, an adaptive attribute fusion and update method is proposed. This method can automatically adjust the fusion strategy and update frequency according to the latest status of the resource. By using online learning algorithms (such as online gradient descent or stochastic gradient descent) to continuously optimize the attribute fusion and update strategy, the resource description is always up to date. The formula is expressed as:

[0131] M std =τ(M raw )+ΔM std =AdaptiveUpdate(τ(M raw ),ΔM std )

[0132] In order to improve the accuracy and consistency of resource description models, a semantic understanding-based attribute mapping method is proposed. This method can improve the accuracy of mapping by understanding the context and semantics of resource descriptions. The pre-trained language model BERT or GPT is used to understand the meaning of resource descriptions and perform attribute mapping based on semantic similarity.

[0133] S5: Through multimodal attribute fusion and attribute mapping and conversion methods, a unified resource description model is constructed to achieve standardization and consistency of resource attributes.

[0134] By introducing adaptive policy learning methods, multimodal utility functions, knowledge graph-based policy reasoning methods, and two-stage reinforcement learning-based policy optimization methods, the policy-driven configuration system can handle resource allocation problems more intelligently.

[0135] The decision function is:

[0136] Γ∏×C→A

[0137] Among them, ∏ is a set of strategies; C is the current resource configuration conditions; A is the set of actions taken.

[0138] Specifically, the strategy set Π contains a series of optional strategies that define how to configure resources under given conditions. Each strategy is a function that decides what action A to take based on the current resource configuration conditions C.

[0139] Suppose there is a strategy set Π, which contains the following strategies:

[0140] Cost Optimization Strategy π cost : Give priority to the resource allocation plan with the lowest cost.

[0141] Performance optimization strategy π performance : Give priority to the resource configuration plan with the highest performance.

[0142] Elastic Strategy π elasticity : Give priority to resource configuration solutions with higher elasticity to cope with load changes.

[0143] Reliability Strategy reliability : Give priority to the most reliable resource allocation plan.

[0144] Furthermore, the current resource configuration conditions C describe the current state of the system and resource availability. These conditions usually include hardware resources, software environment, network conditions, etc.

[0145] Assume C contains the following conditions:

[0146] CPU utilizationc CPU : Current CPU usage percentage.

[0147] Memory utilizationc memory : The current memory usage percentage.

[0148] Network bandwidthc bandwidth : Current network bandwidth usage.

[0149] Storage spacec storage : Current storage space usage.

[0150] Load condition c load : Current server load.

[0151] The set of actions taken A contains various operations taken by the system to change the current resource configuration conditions C.

[0152] Assume A contains the following actions: adding resource a increase : Increase the allocation of specific resources, such as increasing the number of CPU cores or increasing the memory size; reduce resource a decrease : Reduce the allocation of specific resources, such as reducing the number of CPU cores or reducing the memory size; reallocate resources a reallocate : reallocate resources, such as moving resources from one server to another; optimize configuration a optimize : Optimize existing resource configuration, such as adjusting the load balancer settings; enable backup resources enable_backup : Enable backup resources, such as enabling backup servers or storage devices.

[0153] The decision model is:

[0154] Γ(πc)=argmaxU(π,c,a)

[0155] Where U is the utility function, which measures the benefit of taking action a under condition c for policy π.

[0156] Strategy π includes cost optimization strategy π cost :Prioritize the resource allocation solution with the lowest cost; performance optimization strategy π performance : Give priority to the resource allocation plan with the highest performance; elastic strategy π elasticity :Prioritize resource allocation solutions with higher elasticity to cope with load changes; reliability strategy π reliability : Give priority to the most reliable resource allocation plan.

[0157] The current resource configuration conditions c include: CPU utilization c CPU : Current CPU usage percentage; memory utilization c memory : Current memory usage percentage; network bandwidth c bandwidth : Current network bandwidth usage; storage space c storage : Current storage space usage; load situation c load : Current server load.

[0158] The actions a taken include the following actions: increasing resource a increase : Increase the allocation of specific resources, such as increasing the number of CPU cores or increasing the memory size; reduce resource a decrease : Reduce the allocation of specific resources, such as reducing the number of CPU cores or reducing the memory size; reallocate resources a reallocate : reallocate resources, such as moving resources from one server to another; optimize configuration a optimize : Optimize existing resource configuration, such as adjusting the load balancer settings; enable backup resources enable_backup : Enable backup resources, such as enabling backup servers or storage devices.

[0159] Furthermore, the utility function U is used to measure the benefit of taking action a under a given strategy π and current condition c.

[0160] Assume that U takes into account factors such as cost, performance, and reliability, and define a utility function U as follows:

[0161] U(πc,a)=w cost ·u cost +w performance ·u performance +w reliability ·u reliability

[0162] Here, w cost ,w performance , and w reliability are the weight coefficients of cost, performance and reliability respectively, and u cost ,u performance , and u reliabilityare the utility values ​​of cost, performance, and reliability, respectively.

[0163] Going a step further, policy-driven configuration reinforcement learning:

[0164] Γ=RLearn(π,C,A,U,t)

[0165] Where RLearn is a policy optimization function based on reinforcement learning; t is a time variable used to dynamically adjust the learning rate.

[0166] In order to make policy learning more flexible and efficient, an adaptive policy learning method is introduced, which can automatically adjust the learning strategy according to the current environment and resource conditions. Use online learning algorithms (such as online gradient descent or stochastic gradient descent) to continuously optimize the strategy to ensure that the strategy can quickly adapt to environmental changes:

[0167] Γ=AdaptiveStrategyLearning(π,C,A,U,t)

[0168] In order to more comprehensively measure the effectiveness of the strategy, a multimodal utility function U is proposed, which can comprehensively consider the benefits of different dimensions. Deep learning models (such as multimodal attention mechanisms) are used to integrate utility values ​​of different dimensions to ensure that strategy optimization can cover multiple aspects of needs.

[0169] U=MultiModalUtilityFunction(π,c,a)

[0170] In order to improve the intelligence and accuracy of policy-driven configuration, a policy reasoning method based on knowledge graph is proposed. This method can use the associations in the knowledge graph to infer the optimal policy.

[0171] Use Graph Neural Networks (GNNs) to build a relationship map between resources and optimize strategies based on these relationships.

[0172] Γ=KnowledgeGraphBasedStrategyInference(π,C,A,U,t)

[0173] In order to improve the efficiency and accuracy of policy optimization, a method based on two-stage reinforcement learning is proposed. This method first finds an approximate optimal policy through rapid learning, and then further optimizes the policy through refined learning. Use reinforcement learning algorithms (such as DQN or PPO) for preliminary learning, and then use reinforcement learning algorithms (such as TD3 or SAC) for subsequent optimization.

[0174] Γ=TwoStageReinforcementLearning(π,C,A,U,t)

[0175] Furthermore, the four-stage model includes:

[0176] Ω=(Ψ,Φ,V,F b )

[0177] Among them, Ψ is the generation phase; Φ is the execution phase; V is the verification phase; F b It is the feedback and correction stage.

[0178] The generation phases are:

[0179] ΨD→π *

[0180] Where D is the historical data; π * is the optimal strategy.

[0181] The introduction of deep reinforcement learning algorithms combined with meta-learning technology enables the generation phase to converge to the optimal strategy more quickly. Model-based reinforcement learning (MBRL) is used in combination with the meta-learning framework to accelerate the learning process.

[0182] Ψ=DeepMetaRL(D)→π *

[0183] Using the reinforcement learning algorithm DeepMetaRL(D) that combines deep learning and meta-learning techniques, we can eventually learn the optimal strategy π * Among them, Deep: means the use of deep learning technology; Meta: stands for meta-learning; RL: reinforcement learning; D stands for Deep.

[0184] The execution phase is:

[0185] Φπ→A t

[0186] Among them A t is the action set and π is the policy.

[0187] This embodiment also provides a computer device, which is suitable for the method of automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the method of automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture as proposed in the above embodiment.

[0188] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0189] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for automatically discovering and integrating heterogeneous resources for a hybrid cloud architecture as proposed in the above embodiment is implemented.

[0190] In summary, the present invention is a method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture. The technical solution of this patent proposes a number of innovative improvements to the limitations of the existing technology. In terms of automatic resource discovery, by designing a resource detector and an API adaptation layer, combined with deep scanning and intelligent analysis technology, accurate identification and deep correlation mapping of resources are achieved, and real-time monitoring of resource status is ensured through a real-time detection mechanism. Specifically, the flexibility and accuracy of resource detection are improved through an adaptive learning mechanism, and intelligent identification and classification of resource types are achieved using deep learning technology. The dynamic recognition function D_t is introduced to capture new resources, and deep feature extraction and adaptive optimization of resource data are achieved, thereby significantly improving the accuracy and timeliness of resource identification; in terms of resource configuration, a general resource description model and attribute mapping mechanism are constructed to achieve high automation and standardization of resource configuration. A policy-driven and efficient resource configuration engine is adopted, combined with an automated control process and an orchestration engine, to achieve codeless configuration and cross-platform consistency. The consistency and integrity of resource descriptions are improved by introducing multimodal attribute fusion methods, the consistency and portability of resource descriptions are ensured by using intelligent attribute mapping and conversion methods, and the resource description model can be dynamically adjusted according to changes in resource characteristics through adaptive attribute fusion and update methods. The attribute mapping method based on semantic understanding further improves the accuracy and consistency of resource descriptions. In terms of cross-cloud resource collaboration and interoperability, a unified monitoring and intelligent scheduling system is designed, and efficient cross-cloud resource collaboration and dynamic load balancing are achieved through adaptive algorithms and global resource views, and a compliance verification mechanism is introduced to ensure the legality of operations and consistency with international standards. The cross-platform compatibility is enhanced through the adaptive architecture of cross-platform API adaptation, the success rate of API calls is improved by using the prediction and optimization mechanism of API call success rate, and the efficiency and accuracy of resource status monitoring are improved by introducing an adaptive event-driven detection mechanism. The dynamic adjustment method based on anomaly detection can timely detect resource anomalies and respond quickly, and the intelligent feedback loop realizes the self-optimization and self-correction of the real-time detection mechanism. In addition, by implementing a role-based access control model (RBAC) and compliance verification functions, operations are ensured to comply with regulations such as GDPR and HIPAA, achieving efficient collaboration and dynamic load balancing across cloud resources while ensuring the legality of operations and consistency with international standards.

[0191] Example 2

[0192] Reference Figure 1 to Figure 5 , which is the second embodiment of the present invention, and this embodiment provides a method for automatic discovery and integrated configuration of heterogeneous resources for a hybrid cloud architecture. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0193] Define the cloud service provider API interface function F i:Each cloud service provider provides an API interface function that allows the system to call to obtain resource information and collect all i (R j ) to obtain resource data, for each cloud service provider i = 1, 2, ..., n, use F i Get Resource R j The acquired data is added to the resource data set D. The formula is:

[0194]

[0195] Define a dynamic identification function to detect new resources, define a dynamic time interval adjuster to adjust the detection frequency according to resource change trends, and perform dynamic resource discovery: In each resource discovery cycle, in addition to using it to acquire resources, it is also used to find new resources and dynamically adjust the time interval for the next detection based on resource change trends and historical data.

[0196] Define a deep feature extraction function: a function used to extract key features of resources, and define an adaptive feature extractor AdaptiveFeatureExtractor: dynamically adjust the feature extraction strategy according to the resource type.

[0197] Use to perform deep scanning; use to parse the structure and attributes of resources; use to extract key features of resources; call AdaptiveFeatureExtractor according to the resource type to optimize the feature extraction process.

[0198] Use the pre-trained model using deep learning technology to define the AutoClassifier: a classifier used to automatically identify resource types. For each resource, use AutoClassifier to classify it, use deep learning technologies such as CNN or RNN to identify the resource type, and use clustering algorithms such as K-means or DBSCAN to further classify the resources.

[0199] Create a set of adapter modules to handle different types of API interfaces, define an adapter module set: including adapter modules for different API interfaces, and create adapter modules: each adapter module is customized for the first type of API to achieve cross-platform compatibility.

[0200] Dynamically adjust the adaptation strategy to adapt to the ever-changing environment and define the adaptation module: create an adaptation module for the first type of API and define a dynamic configuration function to dynamically adjust the adaptation strategy based on the API interface function, metadata, and time variables.

[0201] Define an online learning algorithm OnlineLearningAlgorithm: such as online gradient descent or stochastic gradient descent, and define a self-learning adapter AdaptiveAPIAdapter: used to automatically adjust the adaptation strategy based on the API interface function, metadata, time variables, and online learning algorithm.

[0202] Automatically parse API metadata and automatically adjust the adaptation strategy based on these metadata, define the semantic parser SemanticParser: use natural language processing technology (such as word embedding and semantic parsing) to parse the metadata in the API document, define metadata to create metadata for the first type of API, and perform metadata parsing and mapping:

[0203] Use SemanticParser to parse the metadata in the API documentation;

[0204] Dynamically update metadata based on time variables.

[0205] Define AdaptiveArchitectureGenerator: Use deep learning technology (such as Transformer model) to analyze the structure and function of API interface, and use

[0206] AdaptiveArchitectureGenerator analyzes API interface functions and dynamically adjusts the adaptation strategy according to time variables.

[0207] Define the success rate predictor SuccessRatePredictor: Use a deep learning model (such as LSTM or GRU) to predict the success rate of API calls, use SuccessRatePredictor to predict the call success rate of the API interface function, and dynamically adjust the prediction results according to the time variable.

[0208] A resource change function is defined to calculate the amount of resource change, and the detection interval is calculated based on the amount of resource change and the resource change function.

[0209] Define a dynamic time interval adjustment function: adjust the detection interval according to the amount of resource changes and a high-dimensional feature extraction function, and define a high-dimensional feature extraction function: used to extract patterns and trends of resource changes.

[0210] Define an adaptive event-driven detection function AdaptiveEventDriven: use machine learning algorithms (such as reinforcement learning) to train the detection interval adjustment strategy, and define a dynamic time interval adjustment function: dynamically adjust the detection interval according to the resource change amount and high-dimensional feature extraction function.

[0211] Define the multimodal feature fusion function MultiModalFeaturer'usion: Use the deep learning model QanCNN and RNN) to fuse multiple types of features and extract different types of data: time series data, log data, and performance indicator data.

[0212] Use statistical learning or deep learning techniques (such as isolation forests or autoencoders) to identify abnormal patterns in resource changes and adjust the detection interval based on the anomaly detection results.

[0213] Enables real-time detection mechanisms to self-optimize and self-correct.

[0214] Define a set of base attributes, a set of status information, and a set of configuration parameters, and comprehensively consider the different attributes of resources from multiple dimensions.

[0215] Use deep learning models (such as multimodal attention mechanisms) to integrate resource attributes from different sources, including but not limited to performance metrics, logging, and time series data.

[0216] Map the resource description set of a specific cloud service provider to a common model, using the deep learning model DeepMap: optimize the mapping strategy through transfer learning technology and reinforcement learning.

[0217] Update the normalization model according to the latest changes in resource characteristics, and use online learning algorithms (such as online gradient descent or stochastic gradient descent) to continuously optimize attribute fusion and update strategies.

[0218] Use pre-trained language models such as BERT or GPT to understand the meaning of resource descriptions and perform attribute mapping based on semantic similarity.

[0219] Design a decision model that can select the best action based on the strategy and current configuration conditions. Use reinforcement learning methods to optimize the decision function and build a multimodal utility function to comprehensively consider the benefits of multiple dimensions.

[0220] A two-stage reinforcement learning method is used to improve learning efficiency, and an online learning algorithm is used to continuously optimize the strategy.

[0221] Define the knowledge graph-based strategy inference method KnowledgeGraphBasedStrategyInference: use graph neural network to build a relationship graph between resources, and define the multimodal utility function MultiModalUtilityFunction: integrate utility values ​​of different dimensions.

[0222] Deep reinforcement learning combined with meta-learning technology is used to generate optimal strategies. The intelligent generation stage is defined, including the use of model-based reinforcement learning (MBRL) and meta-learning framework to accelerate the learning process.

[0223] Use a dynamic decision tree algorithm to dynamically adjust policy execution and use adaptive decision trees (ADTs) to adjust execution strategies.

[0224] A fuzzy logic controller (FLC) is used to evaluate the effects of the actions.

[0225] Define the adaptive feedback and correction stage: Use the adaptive threshold adjustment algorithm (ATAA) to dynamically adjust the feedback frequency.

[0226] And dynamically adjust the priority of tasks according to the urgency of tasks and the availability of resources, dynamically adjust the number of concurrent tasks according to system load and resource utilization, and use genetic algorithms or particle swarm optimization algorithms to optimize workflow graphs.

[0227] Use deep learning models to predict resource requirements and pre-allocate resources.

[0228] Define deep learning-based Petri net optimization OptimizedPetriNet: Use a deep learning model to predict the strength of dependencies between tasks and adjust the structure of the Petri net accordingly.

[0229] Define dynamic topological sorting DynamicTopologicalSort:, use online learning algorithm to continuously optimize the topological sorting strategy.

[0230] Define adaptive resource allocation AdaptiveResourceAllocation: Use reinforcement learning algorithm to train resource allocation strategy.

[0231] Define MultiModalDependencyIdentification: Use a multimodal deep learning model to integrate task dependency information from different sources.

[0232] In summary, the patented technical solution solves multiple core technical challenges in hybrid cloud environments, including automatic resource discovery, resource configuration, cross-cloud resource collaboration and interoperability, and security and compliance. These technical solutions not only improve the ability of automatic resource discovery, the automation level and standardization of resource configuration, but also strengthen the collaboration and interoperability of cross-cloud resources, and ensure the security and compliance of the system. Compared with existing technologies, the patented technical solution has significant advantages and can significantly improve the efficiency of resource management and scheduling in hybrid cloud environments.

[0233] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture, characterized by: include, Through adaptive learning and dynamic optimization mechanisms, it automatically identifies and dynamically adjusts resource detection frequency to capture new resources; Use deep learning models to predict resource status change trends, initiate detection in advance, and achieve real-time monitoring and active response; Introducing dynamic API adaptation strategies and self-learning mechanisms to achieve intelligent adaptation of cross-platform APIs and optimize call success rates; Adopting adaptive event-driven detection and multimodal feature fusion methods, the detection frequency is dynamically adjusted based on anomaly detection; Through multimodal attribute fusion and attribute mapping and conversion methods, a unified resource description model is constructed to standardize and ensure consistency of resource attributes.

2. The method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture according to claim 1, characterized in that: The use of a deep learning model to predict resource status change trends includes the following: Introducing a self-learning mechanism for dynamic API adaptation strategies, an intelligent metadata parsing and mapping mechanism, an adaptive architecture for cross-platform API adaptation, and a prediction and optimization mechanism for API call success rates; Among them, the adaptation module set is A = {α1, α2, …, α n }; where each α i Tailored for the i-th class API to achieve cross-platform compatibility, the dynamic configuration of the PI adaptation layer is Among them, M i is the metadata of the ith API, and t is a time variable used to dynamically adjust the adaptation strategy.

3. The method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture according to claim 2, characterized in that: The adaptive event-driven detection and multimodal feature fusion includes: Introducing multimodal attribute fusion methods to comprehensively consider different attributes of resources from multiple dimensions; Use deep learning models to integrate resource attributes from different sources, in, is a multi-dimensional attribute fusion function; B i It is a collection of attributes in different dimensions; Set the mapping function to: τM raw →M std Among them, τ is the mapping function, M raw is a collection of resource descriptions of a specific cloud service provider, M std It is a universal model.

4. The method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture according to claim 3, characterized in that: The attribute mapping and conversion includes: The decision function is: ΓΠ×C→A Among them, Π is the set of strategies, C is the current resource configuration conditions, and A is the set of actions taken; The decision model is: Where U is the utility function, which measures the benefit of taking action a under condition c for policy π.

5. The method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture according to claim 4, characterized in that: The utility function U is used to measure the benefits of taking actions under a given strategy and current conditions; The utility function U is as follows: U(πc,a)=w cost ·u cost +w performance ·u performance +w reliability ·u reliability Among them, w cost , w performance , and w reliability are the weight coefficients of cost, performance and reliability respectively, and u cost ,u performance , and u reliability are the utility values ​​of cost, performance, and reliability, respectively.

6. The method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture according to claim 5, characterized in that: The set of strategies includes a cost optimization strategy, a performance optimization strategy, an elasticity strategy, and a reliability strategy.

7. The method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture according to claim 6, characterized in that: The resource configuration conditions include CPU utilization, memory utilization, network bandwidth, storage space and load conditions.

8. The method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture according to claim 7, characterized in that: The set of actions taken includes increasing resources, reducing resources, reallocating resources, optimizing configuration, and enabling spare resources.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for automatic discovery and integrated configuration of heterogeneous resources for hybrid cloud architecture described in any one of claims 1 to 7 are implemented.

Images