Resource arrangement and automatic execution method and system based on hybrid cloud architecture

Abstract

The invention relates to the technical field of cloud computing, in particular to a resource arrangement and automatic execution method and system based on a hybrid cloud architecture. The resource arrangement and automatic execution method based on the hybrid cloud architecture comprises the steps of pulling or receiving monitoring data, performing streaming and batch cleaning, and generating input required by training and reasoning in a feature processing pipeline; the model is deployed, a prediction result is sent to a decision module in real time, and the decision module triggers automatic execution to carry out capacity expansion and shrinkage on resources; and finally, comparing an execution effect with a prediction result to realize continuous optimization. According to the resource orchestration and automatic execution method and system based on the hybrid cloud architecture, cross-cloud resource dependence analysis and task-driven orchestration execution are realized, strategy-driven dynamic deployment and elastic capacity expansion and contraction are also realized, intelligent execution optimization and exception handling can be performed in combination with artificial intelligence or a rule engine, and the resource orchestration and automatic execution method and system based on the hybrid cloud architecture are high in practicability. The resource utilization rate and the operation and maintenance efficiency of the hybrid cloud environment are improved, and the risk of manual operation is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of cloud computing, and particularly relates to a resource orchestration and automated execution method and system based on a hybrid cloud architecture. Background Art

[0002] With the deepening of enterprise digital transformation, more and more organizations adopt a hybrid cloud architecture to deploy their businesses on various infrastructures such as public clouds, private clouds, and edge clouds. However, the resource interfaces between different cloud platforms are inconsistent and the service models are heterogeneous, resulting in challenges in the deployment and operation and maintenance of applications across multiple clouds.

[0003] Existing technologies mainly rely on manual operations or static templates in resource orchestration and execution, lacking a unified cross-cloud scheduling logic and adaptive capabilities, and it is difficult to meet dynamic business requirements and efficient operation and maintenance requirements.

[0004] In order to solve the problems of resource heterogeneity, scheduling separation, and low execution efficiency existing in the prior art, the present invention proposes a resource orchestration and automated execution method and system based on a hybrid cloud architecture. Summary of the Invention

[0005] The present invention provides a simple and efficient resource orchestration and automated execution method and system based on a hybrid cloud architecture to make up for the deficiencies of the prior art.

[0006] The present invention is implemented through the following technical solutions: A resource orchestration and automated execution method based on a hybrid cloud architecture pulls or receives monitoring data at a custom fixed frequency through unified scheduling or a message queue, then performs streaming and batch cleaning on the data, and then generates the inputs required for training and inference in a feature processing pipeline; Deploy the model in the form of an online service, deliver the prediction results to the decision-making module in real time, and trigger the decision-making module to perform automated execution to scale the resources up or down; Finally, the monitoring module compares the execution effect with the prediction result and feeds it back to the model and decision-making components to achieve continuous optimization; The above entire process uses encrypted transmission and centralized key management, and the communication between services uses two-way authentication and fine-grained access control; all operations generate audit logs and are centrally analyzed, and security scans and compliance reviews are regularly performed according to a custom period, and a compliance report is generated.

[0007] This resource orchestration and automated execution method and system based on a hybrid cloud architecture encrypts all data and instructions end-to-end, centrally generates, distributes, and rotates keys to achieve encrypted transmission; When users access, role and policy management of users is performed based on the principle of least privilege.

[0008] Including the following steps: Step S1: Conduct resource discovery and unified modeling for multiple heterogeneous cloud platforms to form a standard resource abstraction model, realizing the automatic identification of multi-cloud heterogeneous resources; The standard resource abstraction model adopts a plug-in abstraction structure, supporting resource encapsulation and standardization processing for open-source cloud platforms such as OpenStack, VMware, Kubernetes, Alibaba Cloud, Amazon Web Services, and Huawei Cloud; In a hybrid cloud environment, the standard resource abstraction model runs in parallel based on two methods: polling and event-driven, to pull or receive monitoring data; uses a streaming computing framework to perform real-time cleaning on the monitoring data, and conducts historical verification at regular intervals according to a custom period; Based on the cleaned data, short-term and long-term statistical features are generated by combining streaming computing and batch processing, including the mean, maximum value, standard deviation, and growth rate within a moving window, and the features are normalized, dimension-reduced, and automatically screened to ensure efficient and non-redundant model input; Step S2: Build a resource orchestration model based on business requirements, parse resource dependencies, and generate an execution graph; Build a time series model, an ensemble learning model, and a deep learning model in parallel, and learn through a combination of offline large-scale training and online small-batch updates; use an automated hyperparameter tuning mechanism to optimize hyperparameters, and manage versions in a model repository, supporting gray-scale verification and traffic splitting to ensure the stability and accuracy of the resource orchestration model when it goes online; When building a time series model, an ensemble learning model, and a deep learning model, support visual service design, and provide import and export capabilities in YAML and JSON formats, compatible with mainstream template standards such as the cloud application topology and orchestration specification TOSCA and Terraform; The execution graph supports cyclic dependency detection and node priority scheduling, and has a task failure isolation and local retry mechanism; Step S3: Drive task execution through a scheduling engine, and call a cross-cloud adapter to send the task to the corresponding cloud platform, and the cloud platform delivers the prediction result to the decision-making module in real time; The cross-cloud adapter module supports encapsulating application programming interface (API) call parameters and authentication mechanisms through a unified interface, and automatically adapts to resource creation, modification, and destruction operations on each cloud platform; Step S4: Optimize and adjust task execution based on a policy rule engine or an artificial intelligence (AI) model; The decision-making module calculates the optimal scaling and cross-domain scheduling plan in real time according to the prediction results, current resource status, service level requirements, and cost model by using mathematical programming or self-learning algorithms, and builds in a policy rule engine to handle emergencies, ensuring the continuous and stable operation of the business; The policy rule engine constructs a resource scheduling model based on machine learning algorithms, dynamically predicts resource bottlenecks, and adjusts the task scheduling order or allocates cloud regions; Using declarative configuration management and automated deployment practices, all resource changes are stored in a configuration repository in a versioned configuration form, and an automated deployment tool synchronizes the resource changes to the running environment, enabling scaling, migration operations that can be audited and rolled back; Step S5: Monitor the task status in real time and execute the automatic rollback policy when an exception occurs; After each scaling operation, monitor the key metrics in the short term, compare the actual effects with the prediction results in multiple dimensions, trigger model retraining or policy adjustment through the alarm system, and display the core data in the visualization dashboard to help the operation and maintenance and algorithm teams respond in a timely manner; The automatic rollback policy supports any combination of two or three of the strategies based on snapshot recovery, resource reconstruction, and state migration to ensure the ultimate consistency of task execution and business continuity; Step S6: Report the execution result to the management platform and form an audit log.

[0009] In step S1, in a hybrid cloud environment, by deploying lightweight collection components and invoking the cloud platform monitoring interface, multi-dimensional metrics are aggregated into a message queue, written into a time series database in real time, and the data is regularly archived to a big data storage layer, supporting online real-time query and offline batch analysis; The collection components are deployed in a clustered manner and support primary and standby switching; The time series database supports high compression and drill-down queries and can meet the real-time read and write requirements; According to a custom period, schedule batch jobs regularly, write the data into the big data storage layer and convert it into a columnar format to achieve data archiving into the lake; Centralize the management of data table and field information for offline training use.

[0010] In step S1, in feature engineering, calculate the statistical metrics within several time windows in real time and at regular intervals, and the regular period is set by the user; Normalize or standardize the features to eliminate the differences in different dimensions and improve the model convergence speed; Use dimensionality reduction technology to eliminate redundant features and automatically retain the custom core features based on feature importance measurement.

[0011] In step S2, the resource orchestration model is incrementally updated in small batches and customized by using real-time collected data. The parameters of the resource orchestration model are automatically optimized based on a search strategy, and the parameters, metrics, and training records of each version of the resource orchestration model are tracked. Gray-scale verification is supported. The effects of the new and old models are compared within a customized small range, and then a full-scale deployment is carried out.

[0012] In step S4, the decision-making module uses a mathematical programming algorithm to establish and solve a resource allocation optimization model, and uses an adaptive algorithm to optimize the cross-domain scheduling strategy. The policy rule engine executes corresponding preset emergency policies for sudden increases or abnormal situations, and online updates the performance and cost weights according to the business priorities to achieve dynamic weight adjustment.

[0013] In step S4, when managing the configuration repository, all resource definitions are stored in a versioned manner. The automatic deployment tool continuously monitors the changes in the configuration repository and synchronously executes resource orchestration. Idempotent operations are supported to ensure that the results of each automatic deployment are consistent and there are no intermediate side effects. Smooth scaling and version switching are supported without service interruption to achieve seamless updates.

[0014] In step S5, the actual effects and predicted results are compared in multiple dimensions, and the prediction error, cost deviation, and performance changes are calculated. If the alarm system is triggered, relevant personnel or components are notified, and the model and decision parameters are dynamically updated through an open interface, and the operation and algorithm metrics are presented in real time through a dashboard.

[0015] A resource orchestration and automated execution system based on a hybrid cloud architecture is used to implement the above method. It adopts a microservices design and is divided into five parts: a data layer, an intelligent analysis layer, a decision-making layer, an execution layer, and a monitoring and feedback layer. The data layer and the intelligent analysis layer are decoupled through a message queue, and the other layers are decoupled through a service mesh. The data layer is responsible for implementing data collection and preprocessing. The intelligent analysis layer is responsible for implementing feature engineering services. The decision-making layer is responsible for implementing model training and evaluation services. The execution layer is responsible for implementing automatic deployment and execution services. The monitoring and feedback layer is responsible for real-time monitoring of the task status. When an anomaly occurs, an automatic rollback strategy is executed to trigger model retraining or policy adjustment. It also includes a configuration center and a service registry. The configuration center is responsible for centralized unified storage and distribution of global operation parameters, and the service registry is responsible for managing module instances, including discovering new services and service health checks, to ensure the high availability and easy extensibility of the system.

[0016] A resource orchestration and automated execution device based on a hybrid cloud architecture, comprising a memory and a processor; the memory is used to store a computer program, and the processor is used to implement the above method steps when executing the computer program.

[0017] A readable storage medium, on which a computer program is stored, and the computer program implements the above method steps when executed by a processor.

[0018] The beneficial effects of the present invention are as follows: The resource orchestration and automated execution method and system based on the hybrid cloud architecture support the automatic identification and standardized modeling of multi-cloud heterogeneous resources, realize cross-cloud resource dependency parsing and task-driven orchestration execution, and also realize policy-driven dynamic deployment and elastic scaling. It can combine artificial intelligence or rule engines for intelligent execution optimization and exception handling, improving the resource utilization rate and operation and maintenance efficiency of the hybrid cloud environment and reducing the risk of human operation. Description of the Drawings

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Attached Figure 1 is a schematic diagram of the resource orchestration and automated execution system based on the hybrid cloud architecture of the present invention.

[0021] Attached Figure 2 is a schematic diagram of the resource orchestration and automated execution method based on the hybrid cloud architecture of the present invention. Detailed Embodiments

[0022] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in combination with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] As shown in the attached Figure 1 figure, for the resource orchestration and automated execution method based on the hybrid cloud architecture, the monitoring data is pulled or received at a custom fixed frequency through unified scheduling or a message queue, then the data is cleaned in a streaming and batch manner, and then the inputs required for training and inference are generated in the feature processing pipeline. Deploy the model in the form of an online service, and deliver the prediction results to the decision-making module in real time. The decision-making module triggers automated execution to scale the resources up or down. Finally, the monitoring module compares the execution effect with the prediction results and feeds them back to the model and the decision-making components to achieve continuous optimization. The above-mentioned full process adopts encrypted transmission and centralized key management. The communication between services uses two-way authentication and fine-grained access control. All operations generate audit logs and are centrally analyzed. Security scans and compliance reviews are regularly performed according to a custom cycle, and a compliance report is generated.

[0024] This resource orchestration and automated execution method based on a hybrid cloud architecture encrypts all data and instructions end-to-end, centrally generates, distributes, and rotates keys to achieve encrypted transmission. When users access, role and policy management are performed on users based on the principle of least privilege.

[0025] It includes the following steps: Step S1: Conduct resource discovery and unified modeling on multiple heterogeneous cloud platforms to form a standard resource abstraction model, and achieve automatic identification of multi-cloud heterogeneous resources. The standard resource abstraction model adopts a plug-in abstraction structure, and supports resource encapsulation and standardization processing for open-source cloud platforms such as OpenStack, VMware, Kubernetes, Alibaba Cloud, Amazon Web Services, and Huawei Cloud. In a hybrid cloud environment, the standard resource abstraction model runs in parallel based on two methods: polling and event-driven, to pull or receive monitoring data. A streaming computing framework is used to clean the monitoring data in real time, and historical calibration is performed by timed batch processing according to a custom cycle. Based on the cleaned data, short-term and long-term statistical features are generated by combining streaming computing and batch processing, including the mean, maximum value, standard deviation, and growth rate within a moving window. The features are normalized, dimension-reduced, and automatically screened to ensure efficient and non-redundant model input. Step S2: Build a resource orchestration model based on business requirements, parse resource dependencies, and generate an execution graph. Build a time series model, an ensemble learning model, and a deep learning model in parallel, and learn through a combination of offline large-scale training and online small-batch updates. An automated hyperparameter tuning mechanism is used to optimize hyperparameters, and versions are managed in a model repository, supporting gray-scale verification and traffic splitting to ensure the stability and accuracy of the resource orchestration model when it goes online. When building time series models, ensemble learning models, and deep learning models, it supports visual service design and provides import and export capabilities in YAML and JSON formats, and is compatible with mainstream template standards such as the cloud application topology and orchestration specifications TOSCA and Terraform; The execution graph supports cyclic dependency detection and node priority scheduling, and has a task failure isolation and local retry mechanism; Step S3: Drive task execution through a scheduling engine, and call the cross-cloud adapter to send the task to the corresponding cloud platform. The cloud platform delivers the prediction results to the decision-making module in real time; The cross-cloud adapter module supports encapsulating application programming interface (API) call parameters and authentication mechanisms through a unified interface, and automatically adapts the resource creation, modification, and destruction operations of each cloud platform; Step S4: Optimize and adjust task execution based on a policy rule engine or an artificial intelligence (AI) model; Based on the prediction results, the current resource status, service level requirements, and cost model, the decision-making module uses mathematical programming or self-learning algorithms to calculate the optimal scaling and cross-domain scheduling solutions in real time, and builds a policy rule engine to handle emergencies, ensuring the continuous and stable operation of the business; The policy rule engine constructs a resource scheduling model based on machine learning algorithms, dynamically predicts resource bottlenecks, and adjusts the task scheduling order or allocates cloud regions; Using declarative configuration management and automatic deployment practices, all resource changes are stored in a configuration repository in a versioned configuration form. The automatic deployment tool synchronizes the resource changes to the running environment, realizing scaling, migration operations that can be audited and rolled back; Step S5: Monitor the task status in real time, and execute the automatic rollback policy when an exception occurs; After each scaling, monitor the key indicators in the short term, compare the actual effect with the prediction results in multiple dimensions, trigger model retraining or policy adjustment through the alarm system, and display the core data in the visualization dashboard to help the operation and maintenance and algorithm teams respond in time; The automatic rollback policy supports any combination of two or three strategies based on snapshot recovery, resource reconstruction, and state migration to ensure the final consistency and business continuity of task execution; Step S6: Report the execution results to the management platform and form an audit log.

[0026] In step S1, in a hybrid cloud environment, by deploying lightweight collection components and calling the cloud platform monitoring interface, multi-dimensional metrics are aggregated into a message queue, written into a time series database in real time, and the data is archived to the big data storage layer regularly, supporting online real-time query and offline batch analysis; The collection components are deployed in a cluster and support master-slave switching; The time series database supports high compression and drill-down queries, and can meet the real-time read and write requirements; According to a custom period, perform batch operations at regular intervals, write data into the big data storage layer and convert it into a columnar format to achieve data archiving into the lake; Centralize the management of data table and field information for offline training use.

[0027] In step S1, in feature engineering, calculate statistical indicators within several time windows in real-time and at regular intervals, and the regular period is set by the user; Normalize or standardize the features to eliminate the differences in different dimensions and improve the model convergence speed; Use dimensionality reduction technology to eliminate redundant features, and automatically retain custom core features based on feature importance measurement.

[0028] In step S2, use the real-time collected data to perform custom small-batch incremental updates on the resource orchestration model, automatically optimize the resource orchestration model parameters based on the search strategy, and track the parameters, metrics, and training records of each version of the resource orchestration model; support gray-scale verification, compare the effects of the new and old models within a custom small range, and then deploy comprehensively.

[0029] In step S4, the decision-making module uses a mathematical programming algorithm to establish and solve a resource allocation optimization model, and uses an adaptive algorithm to optimize the cross-domain scheduling strategy; Execute corresponding preset emergency strategies for sudden increases or abnormal situations through the policy rule engine, and update the performance and cost weights online according to the business priority to achieve dynamic weight adjustment.

[0030] In step S4, when managing the configuration repository, all resource definitions are stored in a versioned manner, and the automatic deployment tool continuously monitors changes in the configuration repository and synchronously executes resource orchestration; support idempotent operations to ensure that the results of each automatic deployment are consistent and there are no intermediate side effects; support smooth scaling and version switching without service interruption to achieve seamless updates.

[0031] In step S5, compare the actual effect with the prediction result in multiple dimensions, calculate the prediction error, cost deviation, and performance change. If the alarm system is triggered, notify relevant personnel or components, dynamically update the model and decision parameters through the open interface, and present the operation and algorithm metrics in real-time through the dashboard.

[0032] As shown in the appendix Figure 2 As shown, the resource orchestration and automated execution system based on the hybrid cloud architecture is used to implement the above method, and adopts a microservices design, which is divided into five parts: the data layer, the intelligent analysis layer, the decision-making layer, the execution layer, and the monitoring and feedback layer; The data layer is decoupled from the intelligent analysis layer through a message queue, and the remaining layers are decoupled through a service mesh; The data layer is responsible for implementing data collection and preprocessing; The intelligent analysis layer is responsible for implementing feature engineering services; The decision-making layer is responsible for implementing model training and evaluation services; The execution layer is responsible for implementing automatic deployment and execution services; The monitoring and feedback layer is responsible for real-time monitoring of task status. When an exception occurs, it executes an automatic rollback policy and triggers model retraining or policy adjustment; It also includes a configuration center and a service registry; The configuration center is responsible for centralized unified storage and distribution of global operation parameters, and the service registry is responsible for managing module instances, including discovering new services and service health checks, to ensure the high availability and easy scalability of the system.

[0033] The resource orchestration and automation execution device based on the hybrid cloud architecture includes a memory and a processor; the memory is used to store computer programs, and the processor is used to implement the above method steps when executing the computer programs.

[0034] A computer program is stored on the readable storage medium, and when the computer program is executed by a processor, the above method steps are implemented.

[0035] The resource orchestration and automation execution method and system based on the hybrid cloud architecture innovatively combines stream-batch integration processing, model hybrid prediction, self-learning decision-making, and versioned orchestration. During online inference, through service interface calls, it can ensure millisecond-level response, taking into account both real-time performance and a global perspective; by comparing prediction errors and execution costs, it automatically adjusts subsequent process parameters, realizing a full-automatic closed loop of prediction - optimization - execution - feedback.

[0036] In addition, through multi-model fusion and adaptive weights, it ensures high-precision prediction for various loads. By combining self-learning algorithms and mathematical programming, it realizes the dynamic balance of cross-domain costs and performance. Through versioned automated orchestration, it ensures the traceability and high reliability of operations.

[0037] The above-described embodiments are only one of the specific implementation manners of the present invention. Ordinary variations and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included in the protection scope of the present invention.

Claims

1. A resource orchestration and automated execution method based on a hybrid cloud architecture, characterized in that: Pull or receive monitoring data at a custom fixed frequency through unified scheduling or a message queue, then perform streaming and batch cleaning on the data, and then generate the inputs required for training and inference in the feature processing pipeline; Deploy the model in the form of an online service, send the prediction results to the decision-making module in real time, and trigger the decision-making module to automatically execute the scaling of resources; Finally, the monitoring module compares the execution effect with the prediction result and feeds it back to the model and decision-making components to achieve continuous optimization; The above full process adopts encrypted transmission and centralized key management, and the communication between services uses two-way authentication and fine-grained access control; all operations generate audit logs and are centrally analyzed, and security scans and compliance reviews are regularly performed according to a custom cycle, and compliance reports are generated.

2. The resource orchestration and automated execution method based on the hybrid cloud architecture according to claim 1, wherein: Perform end-to-end encryption on all data and instructions, centrally generate, distribute and rotate keys to achieve encrypted transmission; When users access, manage the roles and policies of users based on the principle of least privilege.

3. The resource orchestration and automated execution method based on the hybrid cloud architecture according to claim 2, wherein: Including the following steps: Step S1: Perform resource discovery and unified modeling on multiple heterogeneous cloud platforms to form a standard resource abstraction model, and realize the automatic identification of multi-cloud heterogeneous resources; The standard resource abstraction model adopts a plug-in abstraction structure, and supports the resource encapsulation and standardization processing of open-source cloud platforms OpenStack, VMware, container orchestration system Kubernetes, Alibaba Cloud, Amazon Web Services, and Huawei Cloud; In a hybrid cloud environment, the standard resource abstraction model pulls or receives monitoring data in parallel based on two methods: polling and event-driven; Adopt a streaming computing framework to perform real-time cleaning on the monitoring data, and perform historical verification at regular intervals according to a custom cycle; Based on the cleaned data, generate short-term and long-term statistical features by combining streaming computing and batch processing, including the mean, maximum value, standard deviation, and growth rate within the moving window, and perform normalization, dimensionality reduction, and automatic screening on the features to ensure that the model input is efficient and redundant-free; Step S2: Build a resource orchestration model based on business requirements, parse the resource dependency relationship and generate an execution graph; Build a time series model, an ensemble learning model, and a deep learning model in parallel, and learn through a combination of offline large-scale training and online small-batch updates; adopt an automatic hyperparameter tuning mechanism to optimize the hyperparameters, and manage the versions in the model repository, support gray-scale verification and traffic splitting to ensure the stability and accuracy of the resource orchestration model when it goes online; When building a time series model, an ensemble learning model, and a deep learning model, support visual service design, and provide the import and export capabilities in YAML format and JSON format, and be compatible with the TOSCA template and Terraform template standards; The execution graph supports cyclic dependency detection and node priority scheduling, and has a task failure isolation and local retry mechanism; Step S3: Drive the task execution through the scheduling engine, and call the cross-cloud adapter to send the task to the corresponding cloud platform, and the cloud platform sends the prediction result to the decision-making module in real time; The cross-cloud adapter module supports encapsulating application programming interface (API) call parameters and authentication mechanisms through a unified interface, and automatically adapts the creation, modification, and destruction operations of resources on each cloud platform; Step S4: Optimize and adjust the task execution based on a policy rule engine or an artificial intelligence (AI) model; The decision-making module uses mathematical programming or self-learning algorithms to calculate the optimal scaling and cross-domain scheduling solutions in real time according to the prediction results, current resource status, service level requirements, and cost model, and builds a policy rule engine to handle emergencies, ensuring the continuous and stable operation of the business; The policy rule engine constructs a resource scheduling model based on machine learning algorithms, dynamically predicts resource bottlenecks, and adjusts the task scheduling order or allocates cloud regions; Using declarative configuration management and automatic deployment practices, all resource changes are stored in a configuration repository in a versioned configuration form, and the automatic deployment tool synchronizes the resource changes to the running environment to achieve scaling and migration operations that can be audited and rolled back; Step S5: Monitor the task status in real time and execute the automatic rollback policy when an exception occurs; After each scaling operation, monitor the key metrics, compare the actual effects with the prediction results in multiple dimensions, trigger model retraining or policy adjustment through the alarm system, and display the core data in the visualization dashboard to help the operation and maintenance and algorithm teams respond in a timely manner; The automatic rollback policy supports any combination of two or three of the strategies based on snapshot recovery, resource reconstruction, and state migration to ensure the final consistency and business continuity of task execution; Step S6: Report the execution results to the management platform and form an audit log.

4. The resource orchestration and automated execution method based on the hybrid cloud architecture according to claim 3, characterized in that: In step S1, in a hybrid cloud environment, by deploying lightweight collection components and calling the cloud platform monitoring interface, multi-dimensional metrics are aggregated into a message queue, written into a time series database in real time, and the data is archived to the big data storage layer at regular intervals, supporting online real-time query and offline batch analysis; The collection components are deployed in a cluster and support primary and standby switching; The time series database supports high compression and drill-down queries, and can meet the real-time read and write requirements; According to a custom period, perform batch operations at regular intervals, write the data into the big data storage layer and convert it into a columnar format to achieve data archiving into the lake; Centralize the management of data table and field information for offline training use.

5. The resource orchestration and automated execution method based on a hybrid cloud architecture according to claim 3, characterized in that: In step S1, in feature engineering, calculate the statistical metrics within several time windows in real time and at regular intervals, and the regular interval is set by the user; Normalize or standardize the features to eliminate the differences in different dimensions and improve the model convergence speed; Use dimensionality reduction technology to eliminate redundant features, and automatically retain custom core features based on feature importance measurement.

6. The resource orchestration and automated execution method based on the hybrid cloud architecture according to claim 3, wherein: In step S2, use the real-time collected data to perform custom small-batch incremental updates on the resource orchestration model, automatically optimize the resource orchestration model parameters based on the search strategy, and track the parameters, metrics, and training records of each version of the resource orchestration model; Support gray-scale verification, compare the effects of the new and old models in a custom small range, and then deploy them comprehensively.

7. The resource orchestration and automated execution method based on the hybrid cloud architecture according to claim 3, characterized in that: In step S4, the decision-making module uses a mathematical programming algorithm to establish and solve a resource allocation optimization model, and uses an adaptive algorithm to optimize the cross-domain scheduling strategy; Execute corresponding preset emergency strategies for sudden increases or abnormal situations through the policy rule engine, and update the performance and cost weights online according to business priorities to achieve dynamic weight adjustment.

8. The resource orchestration and automated execution method based on the hybrid cloud architecture according to claim 3, characterized in that: In step S4, when managing the configuration repository, all resource definitions are stored in a versioned manner, and the automated deployment tool continuously monitors changes in the configuration repository and synchronously executes resource orchestration; Support idempotent operations to ensure consistent results for each automated deployment without intermediate side effects; support smooth scaling and version switching without service interruption to achieve seamless updates.

9. The resource orchestration and automated execution method based on the hybrid cloud architecture according to claim 3, wherein: In step S5, compare the actual effects with the prediction results in multiple dimensions, calculate the prediction error, cost deviation, and performance changes. If the alarm system is triggered, notify relevant personnel or components, dynamically update the model and decision parameters through the open interface, and present the operation and algorithm metrics in real time through the dashboard.

10. A resource orchestration and automated execution system based on a hybrid cloud architecture, characterized in that: Used to implement the method described in any one of claims 1 to 9, adopting a microservices architecture design, divided into five parts: the data layer, the intelligent analysis layer, the decision-making layer, the execution layer, and the monitoring and feedback layer; The data layer and the intelligent analysis layer are decoupled through a message queue, and the other layers are decoupled through a service mesh; The data layer is responsible for implementing data collection and preprocessing; The intelligent analysis layer is responsible for implementing feature engineering services; The decision-making layer is responsible for implementing model training and evaluation services; The execution layer is responsible for implementing automated deployment and execution services; The monitoring and feedback layer is responsible for real-time monitoring of task status, executing an automatic rollback strategy when an exception occurs, and triggering model retraining or policy adjustment; It also includes a configuration center and a service registry; The configuration center is responsible for centralized unified storage and distribution of global operation parameters, and the service registry is responsible for managing each module instance, including discovering new services and service health checks to ensure the high availability and scalability of the system.

Images