A method and system for concurrent activation of heterogeneous network elements based on AI dynamic scheduling
By using AI-driven dynamic scheduling, the problems of high response latency, low resource utilization, and complex heterogeneous management in the OSS service activation process were solved, achieving efficient concurrent activation of network elements and improving system throughput and stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INSPUR TIANYUAN COMM INFORMATION SYST CO LTD
- Filing Date
- 2026-02-24
- Publication Date
- 2026-07-10
AI Technical Summary
The existing OSS service activation process suffers from problems such as large response latency, low resource utilization, heterogeneous complexity, and passive exception handling, which are particularly evident during peak business periods.
An AI-based dynamic scheduling method is adopted, which dynamically determines the concurrency of network elements through prediction models and reinforcement learning algorithms. Combined with task queue management and session pool optimization, intelligent scheduling and anomaly prediction are achieved, thereby improving system throughput and stability.
It significantly improves work order processing throughput, reduces average latency, enhances system stability and scalability, and can handle heterogeneous management of network elements from different vendors.
Smart Images
Figure CN122372444A_ABST
Abstract
Description
Technical Field
[0001] This invention discloses a method and system for concurrent activation of heterogeneous network elements based on AI dynamic scheduling, which relates to the field of communication and information technology. Background Technology
[0002] In a typical OSS service activation process, business orders submitted by the northbound CRM system are received and split by the OM (Operational Management) system. The IM (Instant Messaging) system then queries resources, the AM (Activation Management) system sends activation commands to each network element, and the WFM (Workforce Management Function) system arranges the implementation. Existing AM modules often use a timed polling / scanning method, such as scanning for pending instructions every 10 seconds, or using a fixed concurrent thread pool to send commands to network elements one by one. This type of solution has some problems during peak business periods: Large response latency: Fixed concurrency or polling leads to task backlog, increasing the overall work order completion latency; Low resource utilization: The inability to dynamically allocate resources based on the performance differences and protocol complexity of different network elements causes some network elements to become bottlenecks; Heterogeneous complexity: Network element devices from different vendors use multiple protocols such as TL1 / SOAP / SNMP / MML / SSH / SFTP / LDAP / SMPP, lacking a unified scheduling and management mechanism; Passive exception handling: When a task fails, it often only performs a simple retry, lacking predictive and intelligent fault tolerance strategies. Summary of the Invention
[0003] This invention addresses the problems of existing technologies by providing a method and system for concurrent activation of heterogeneous network elements based on AI dynamic scheduling. It solves the problems of low activation efficiency, rigid resource allocation, and lack of intelligent scheduling and anomaly prediction in existing technologies, thereby significantly improving work order processing throughput and reducing average latency during peak business periods, while ensuring system stability and scalability.
[0004] The specific solution proposed in this invention is as follows: This invention provides a method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling, comprising: Step 1: Receive and parse the service work order, and break it down into multiple sub-tasks according to the target network element; Step 2: Receive the subtasks using a task queue that supports priority management; Step 3: Based on the historical performance data and real-time status of each network element, dynamically determine the optimal concurrency for each network element using a predictive model: Input the characteristic data for predicting network element concurrency, including: historical response time statistics of the network element, task execution success rate and failure rate, task throughput trend within the recent time window, resource utilization index of the network element itself, protocol type characteristics, time period characteristics, and current queue depth of pending tasks. Prediction model selection: For comprehensive feature correlation analysis of network element concurrency, a regression prediction model is adopted; for data with strong temporal correlation in network element concurrency, a deep prediction model that can capture temporal dependencies is adopted; based on the requirements of prediction speed and accuracy in the actual scenario, a selection or combination between lightweight and complex models is made. The predictive model outputs a suggested value for the optimal network element concurrency for each network element, providing a reliable reference for scheduling decisions. Step 4: Based on the optimal concurrency and the priority attributes of the subtasks, dynamically decide the task dequeueing order and resource allocation scheme through scheduling strategies; Step 5: According to the resource allocation scheme, allocate session resources from the session pool, and send task instructions to the target network element in parallel through the protocol adaptation layer; Step 6: Monitor and collect execution results, monitor execution metrics in real time and predict anomalies, and feed the results data back to the prediction model and scheduling strategy model to achieve continuous model optimization.
[0005] Furthermore, in step 1 of the AI-based dynamic scheduling method for concurrent activation of heterogeneous network elements, each subtask includes information such as work order identifier, subtask identifier, target network element ID, protocol type, activation instruction set, service priority, and expected completion time, and the subtasks are standardized.
[0006] Furthermore, in step 2 of the AI-based dynamic scheduling method for concurrent activation of heterogeneous network elements, a unified enqueue interface and a controlled dequeue interface are exposed to the outside world through the task queue, and subtask management is carried out using idempotency verification, task deduplication, task expiration and delayed delivery mechanisms.
[0007] Furthermore, in step 3 of the AI-based dynamic scheduling heterogeneous network element concurrent activation method, the prediction model is trained by combining offline batch training with periodic online incremental training. The training objective is to minimize the loss function after the trade-off between the predicted concurrency and the actual system throughput / latency.
[0008] Furthermore, step 4 of the aforementioned method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling includes: Problem modeling: The scheduling task is modeled as a sequence decision problem. The state space includes the current queue state, the predicted concurrency and health score of each network element, and the task priority and SLA information. The action space consists of decisions on selecting subtasks from the task queue and allocating execution resources to the subtasks. Algorithm selection: A value-based reinforcement learning algorithm is employed to learn decision-making strategies that yield long-term optimal returns in complex states. To address strategy exploration and cold-start scenarios, classical decision-making algorithms or rule-based heuristics are integrated as supplementary methods. Optimization Objective: The optimization objective of the strategy is defined by the reward function. The reward function positively incentivizes improved SLA fulfillment rate and system throughput, while negatively penalizing average task latency, execution failures, and repeated retries. The reward function is a multi-objective weighted form, and the weight coefficients are adjusted according to business preferences. Training and Deployment: The core policy model is first trained offline in a simulation environment built on historical data, employing stabilization techniques such as experience replay and target networks during training. After deployment, the policy can be continuously evaluated and fine-tuned through small-scale canary releases and online learning mechanisms.
[0009] Furthermore, in step 5 of the AI-based dynamic scheduling method for concurrent activation of heterogeneous network elements, a dynamically scalable session pool is maintained for each network element. The concurrency of the session pool is dynamically and elastically adjusted according to the predicted concurrency results. The session pool is responsible for the reuse, establishment, and recycling of sessions, avoiding the overhead caused by frequent connection establishment.
[0010] Furthermore, step 6 of the aforementioned method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling specifically includes: Real-time collection of execution metrics, including response time for each command issued, success / failure status, number of retries, and session connection establishment time, continuously feeds this data into the real-time monitoring data stream. Unsupervised machine learning detection models or threshold detection methods based on statistical distributions are used to identify sudden anomalies or abnormal time periods that deviate from normal patterns. A time series forecasting model is used to predict the changing trends of key indicators such as network element load and task latency in the near future, in order to anticipate potential congestion risks. When an anomaly is detected or a bottleneck risk is predicted, preset rules are triggered. These preset rules include reducing the distribution rate, adjusting scheduling priority, or sending alarm information to the upper-level operation and maintenance system.
[0011] Furthermore, in step 6 of the AI-based dynamic scheduling method for concurrent activation of heterogeneous network elements, the result data is fed back to the prediction model and the scheduling strategy model to achieve continuous model optimization. This includes continuously recording the result data and feeding it back for model evaluation and retraining. The specific process is as follows: Perform offline batch updates: Perform offline retraining on a periodic basis and verify performance. Perform online incremental updates: Employ parameter incremental updates or streaming learning methods to quickly adapt to business changes; Perform model management: control model versions, monitor model performance, and establish rollback mechanisms.
[0012] This invention also provides a heterogeneous network element concurrent activation system based on AI dynamic scheduling, including a work order receiving and splitting module, a task queue management module, a dynamic session pool management module, an AI concurrency prediction module, an AI intelligent scheduling module, an anomaly prediction and monitoring module, and a feedback learning module. The work order receiving and splitting module receives and parses the business work order, and splits it into multiple sub-tasks according to the target network element; The task queue management module uses a task queue that supports priority management to receive the subtasks; The AI concurrency prediction module dynamically determines the optimal concurrency for each network element based on its historical performance data and real-time status using a prediction model. Input the characteristic data for predicting network element concurrency, including: historical response time statistics of the network element, task execution success rate and failure rate, task throughput trend within the recent time window, resource utilization index of the network element itself, protocol type characteristics, time period characteristics, and current queue depth of pending tasks. Prediction model selection: For comprehensive feature correlation analysis of network element concurrency, a regression prediction model is adopted; for data with strong temporal correlation in network element concurrency, a deep prediction model that can capture temporal dependencies is adopted; based on the requirements of prediction speed and accuracy in the actual scenario, a selection or combination between lightweight and complex models is made. The predictive model outputs a suggested value for the optimal network element concurrency for each network element, providing a reliable reference for scheduling decisions. The AI intelligent scheduling module dynamically decides the task dequeueing order and resource allocation scheme based on the optimal concurrency and the priority attributes of the subtasks through scheduling strategies. The dynamic session pool management module allocates session resources from the session pool according to the resource allocation scheme, and sends task instructions to the target network element in parallel through the protocol adaptation layer; The anomaly prediction and monitoring module monitors and collects execution results: The anomaly prediction and monitoring module monitors execution indicators in real time and predicts anomalies. The feedback learning module feeds the result data back to the prediction model and scheduling strategy model to achieve continuous optimization of the model.
[0013] The present invention also provides a computer-readable medium storing computer instructions, which, when executed by a processor, cause the processor to perform the aforementioned method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling.
[0014] The advantages of this invention are: Significantly improves concurrent processing capabilities: By predicting and dynamically adjusting the session pool size and scheduling strategy, the bottleneck caused by fixed concurrency is avoided, and the overall throughput is improved; Reduce average latency: Prioritize high-priority / high-SLA work orders to reduce overall average activation latency; Improve system stability: Anomaly prediction and fault tolerance mechanisms can detect and mitigate network element failures or congestion in advance, reducing failure rate and the risk of large-scale rollback; Excellent scalability: The protocol adaptation layer makes it easy for the system to connect to network elements from new manufacturers, and the AI models and strategies have adaptive capabilities; Engineering feasibility: The solution adopts mature machine learning methods and a distributed message queue / session pool design, which has strong engineering feasibility and can be quickly implemented in the existing OSS architecture. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 This is a schematic diagram of the application framework of the present invention.
[0017] Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation
[0018] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention. Example
[0019] This invention provides a method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling, comprising: Step 1: Receive and parse the service work order, and break it down into multiple subtasks based on the target network element. Receive work orders from the northbound system and break them down into multiple subtasks based on the target network element. Each subtask includes: work order identifier, subtask identifier, target network element ID, protocol type, activation instruction set, service priority, and expected completion time (SLA). Subtasks are standardized.
[0020] Step 2: Receive the subtasks using a task queue that supports priority management. A priority-supporting message queue or a blocking priority queue can be used as the task buffer layer. The queue supports priority sorting, idempotency verification, task deduplication, task expiration, and delayed delivery mechanisms. The task queue exposes a unified enqueue interface and a controlled dequeue interface for external calls.
[0021] Step 3: Based on the historical performance data and real-time status of each network element, dynamically determine the optimal concurrency for each network element using a predictive model: Input the characteristic data for predicting network element concurrency, including: historical response time statistics of the network element, task execution success rate and failure rate, task throughput trend within the recent time window, resource utilization index of the network element itself, protocol type characteristics, time period characteristics, and current queue depth of pending tasks. Prediction model selection: For comprehensive feature correlation analysis of network element concurrency, a regression prediction model is adopted; for data with strong temporal correlation in network element concurrency, a deep prediction model that can capture temporal dependencies is adopted; based on the requirements of prediction speed and accuracy in the actual scenario, a selection or combination between lightweight and complex models is made. Training the prediction model: The prediction model is trained by combining offline batch training with periodic online incremental training. The training objective is to minimize the loss function after the trade-off between prediction concurrency and actual system throughput / latency. The predictive model outputs a suggested value for the optimal network element concurrency for each network element, providing a reliable reference for scheduling decisions.
[0022] Step 4: Based on the optimal concurrency and subtask priority attributes, dynamically determine the task dequeueing order and resource allocation scheme through a scheduling strategy: Problem modeling: The scheduling task is modeled as a sequence decision problem. The state space includes the current queue state, the predicted concurrency and health score of each network element, and the task priority and SLA information. The action space consists of decisions on selecting subtasks from the task queue and allocating execution resources to the subtasks. Algorithm selection: A value-based reinforcement learning algorithm is employed to learn decision-making strategies that yield long-term optimal returns in complex states. To address strategy exploration and cold-start scenarios, classical decision-making algorithms or rule-based heuristics are integrated as supplementary methods. Optimization Objective: The optimization objective of the strategy is defined by the reward function. The reward function positively incentivizes improved SLA fulfillment rate and system throughput, while negatively penalizing average task latency, execution failures, and repeated retries. The reward function is a multi-objective weighted form, and the weight coefficients are adjusted according to business preferences. Training and Deployment: The core policy model is first trained offline in a simulation environment built on historical data, employing stabilization techniques such as experience replay and target networks during training. After deployment, the policy can be continuously evaluated and fine-tuned through small-scale canary releases and online learning mechanisms.
[0023] Step 5: According to the resource allocation scheme, allocate session resources from the session pool and send task instructions to the target network element in parallel through the protocol adaptation layer. Each network element maintains a dynamically scalable session pool. The concurrency of the session pool is dynamically and elastically adjusted according to the predicted concurrency results. The session pool is responsible for the reuse, establishment and recycling of sessions to avoid the overhead caused by frequent connection establishment.
[0024] The Protocol Abstraction Layer (PAL) provides a unified interface for network elements from different manufacturers and using different protocols. Each type of network element implements a corresponding adapter to perform protocol mapping, mapping the unified execution request from the upper layer to request messages under a specific protocol and collecting responses. The protocols include one or more of TL1, SOAP, SNMP, MML, SSH, SFTP, LDAP, and SMPP.
[0025] Step 6: Monitor and collect execution results, monitor execution metrics in real time and predict anomalies, and feed the results data back to the prediction model and scheduling strategy model to achieve continuous model optimization.
[0026] Specifically, it can include: Real-time collection of execution metrics, including response time for each command issued, success / failure status, number of retries, and session connection establishment time, continuously feeds this data into the real-time monitoring data stream. Unsupervised machine learning detection models or threshold detection methods based on statistical distributions are used to identify sudden anomalies or abnormal time periods that deviate from normal patterns. A time series forecasting model is used to predict the changing trends of key indicators such as network element load and task latency in the near future, in order to anticipate potential congestion risks. When an anomaly is detected or a bottleneck risk is predicted, preset rules are triggered. These preset rules include reducing the distribution rate, adjusting scheduling priority, or sending alarm information to the upper-level operation and maintenance system.
[0027] When feeding the results data back to the prediction model and scheduling strategy model to achieve continuous model optimization, this may include: continuously recording the results data and feeding it back for model evaluation and retraining. The specific process is as follows: Perform offline batch updates: Perform offline retraining on a periodic basis and verify performance. Perform online incremental updates: Employ parameter incremental updates or streaming learning methods to quickly adapt to business changes; Perform model management: control model versions, monitor model performance, and establish rollback mechanisms. Example
[0028] This invention also provides a heterogeneous network element concurrent activation system based on AI dynamic scheduling, including a work order receiving and splitting module, a task queue management module, a dynamic session pool management module, an AI concurrency prediction module, an AI intelligent scheduling module, an anomaly prediction and monitoring module, and a feedback learning module. The work order receiving and splitting module receives and parses the business work order, and splits it into multiple sub-tasks according to the target network element; The task queue management module uses a task queue that supports priority management to receive the subtasks; The AI concurrency prediction module dynamically determines the optimal concurrency for each network element based on its historical performance data and real-time status using a prediction model. Input the characteristic data for predicting network element concurrency, including: historical response time statistics of the network element, task execution success rate and failure rate, task throughput trend within the recent time window, resource utilization index of the network element itself, protocol type characteristics, time period characteristics, and current queue depth of pending tasks. Prediction model selection: For comprehensive feature correlation analysis of network element concurrency, a regression prediction model is adopted; for data with strong temporal correlation in network element concurrency, a deep prediction model that can capture temporal dependencies is adopted; based on the requirements of prediction speed and accuracy in the actual scenario, a selection or combination between lightweight and complex models is made. The predictive model outputs a suggested value for the optimal network element concurrency for each network element, providing a reliable reference for scheduling decisions. The AI intelligent scheduling module dynamically decides the task dequeueing order and resource allocation scheme based on the optimal concurrency and the priority attributes of the subtasks through scheduling strategies. The dynamic session pool management module allocates session resources from the session pool according to the resource allocation scheme, and sends task instructions to the target network element in parallel through the protocol adaptation layer; The anomaly prediction and monitoring module monitors and collects execution results: The anomaly prediction and monitoring module monitors execution indicators in real time and predicts anomalies. The feedback learning module feeds the result data back to the prediction model and scheduling strategy model to achieve continuous optimization of the model.
[0029] The information interaction and execution process between the modules in the above system are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description in the method embodiment of the present invention, and will not be repeated here.
[0030] Similarly, the advantages of the system of the present invention are: Significantly improves concurrent processing capabilities: By predicting and dynamically adjusting the session pool size and scheduling strategy, the bottleneck caused by fixed concurrency is avoided, and the overall throughput is improved; Reduce average latency: Prioritize high-priority / high-SLA work orders to reduce overall average activation latency; Improve system stability: Anomaly prediction and fault tolerance mechanisms can detect and mitigate network element failures or congestion in advance, reducing failure rate and the risk of large-scale rollback; Excellent scalability: The protocol adaptation layer makes it easy for the system to connect to network elements from new manufacturers, and the AI models and strategies have adaptive capabilities; Engineering feasibility: The solution adopts mature machine learning methods and a distributed message queue / session pool design, which has strong engineering feasibility and can be quickly implemented in the existing OSS architecture. Example
[0031] The present invention also provides a computer-readable medium storing computer instructions, which, when executed by a processor, cause the processor to perform the aforementioned method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling.
[0032] Specifically, a system or apparatus equipped with a storage medium may be provided, on which software program code implementing the functions of any of the embodiments described above is stored, and the computer (or CPU or MPU) of the system or apparatus may read and execute the program code stored in the storage medium.
[0033] In this case, the program code read from the storage medium can itself implement the function of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute part of the present invention.
[0034] Storage media embodiments for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer via a communication network.
[0035] Furthermore, it should be clear that not only can the program code read by the computer be executed, but also the operating system or other components operating on the computer can be instructed based on the program code to perform some or all of the actual operations, thereby realizing the function of any of the embodiments described above.
[0036] Furthermore, it is understood that the program code read from the storage medium is written to the memory set in the expansion board inserted into the computer or to the memory set in the expansion unit connected to the computer. Then, based on the instructions of the program code, the CPU or other components installed on the expansion board or expansion unit execute some and all of the actual operations, thereby realizing the function of any of the embodiments described above.
[0037] It should be noted that not all steps and modules in the above processes and system structures are mandatory; some steps or modules can be omitted as needed. The execution order of the steps is not fixed and can be adjusted as required. The system structures described in the above embodiments can be physical or logical structures. That is, some modules may be implemented by the same physical entity, or some modules may be implemented by multiple physical entities, or they may be implemented by certain components in multiple independent devices.
[0038] The above-described embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.
Claims
1. A method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling, characterized by: include: Step 1: Receive and parse the service work order, and break it down into multiple sub-tasks according to the target network element; Step 2: Receive the subtasks using a task queue that supports priority management; Step 3: Based on the historical performance data and real-time status of each network element, dynamically determine the optimal concurrency for each network element using a predictive model: Input the characteristic data for predicting network element concurrency, including: historical response time statistics of the network element, task execution success rate and failure rate, task throughput trend within the recent time window, resource utilization index of the network element itself, protocol type characteristics, time period characteristics, and current queue depth of pending tasks. Prediction model selection: For comprehensive feature correlation analysis of network element concurrency, a regression prediction model is adopted; for data with strong temporal correlation in network element concurrency, a deep prediction model that can capture temporal dependencies is adopted; based on the requirements of prediction speed and accuracy in the actual scenario, a selection or combination between lightweight and complex models is made. The predictive model outputs a suggested value for the optimal concurrency of each network element, providing a reliable reference for scheduling decisions. Step 4: Based on the optimal concurrency and the priority attributes of the subtasks, dynamically decide the task dequeueing order and resource allocation scheme through scheduling strategies; Step 5: According to the resource allocation scheme, allocate session resources from the session pool, and send task instructions to the target network element in parallel through the protocol adaptation layer; Step 6: Monitor and collect execution results, monitor execution metrics in real time and predict anomalies, and feed the results data back to the prediction model and scheduling strategy model to achieve continuous model optimization.
2. The method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling according to claim 1, characterized in that each subtask in step 1 includes: work order identifier, subtask identifier, target network element ID, protocol type, activation instruction set, service priority, and expected completion time information, and the subtasks are standardized.
3. The method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling according to claim 1, characterized in that: In step 2, a unified enqueue interface and a controlled dequeue interface are exposed to the outside world through the task queue, and subtask management is carried out using idempotent verification, task deduplication, task expiration and delayed delivery mechanisms.
4. The method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling according to claim 1, characterized in that: Step 3 involves training the prediction model: The prediction model is trained using a combination of offline batch training and periodic online incremental training. The training objective is to minimize the loss function after a trade-off between prediction concurrency and actual system throughput / latency.
5. The method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling according to claim 1, characterized in that: Step 4 includes: Problem modeling: The scheduling task is modeled as a sequence decision problem. The state space includes the current queue state, the predicted concurrency and health score of each network element, and the task priority and SLA information. The action space consists of decisions on selecting subtasks from the task queue and allocating execution resources to the subtasks. Algorithm selection: A value-based reinforcement learning algorithm is employed to learn decision-making strategies that yield long-term optimal returns in complex states. To address strategy exploration and cold-start scenarios, classical decision-making algorithms or rule-based heuristics are integrated as supplementary methods. Optimization Objective: The optimization objective of the strategy is defined by the reward function. The reward function positively incentivizes improved SLA fulfillment rate and system throughput, while negatively penalizing average task latency, execution failures, and repeated retries. The reward function is a multi-objective weighted form, and the weight coefficients are adjusted according to business preferences. Training and Deployment: The core policy model is first trained offline in a simulation environment built on historical data, employing stabilization techniques such as experience replay and target networks during training. After deployment, the policy can be continuously evaluated and fine-tuned through small-scale canary releases and online learning mechanisms.
6. The method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling according to claim 1, characterized in that in step 5, a dynamically scalable session pool is maintained for each network element, the concurrency of the session pool is dynamically and elastically adjusted according to the predicted concurrency result, and the session pool is used to be responsible for the reuse, establishment and recycling of sessions, avoiding the overhead caused by frequent connection establishment.
7. The method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling according to claim 1, characterized in that: Step 6 specifically includes: Real-time collection of execution metrics, including response time for each command issued, success / failure status, number of retries, and session connection establishment time, continuously feeds this data into the real-time monitoring data stream. Unsupervised machine learning detection models or threshold detection methods based on statistical distributions are used to identify sudden anomalies or abnormal time periods that deviate from normal patterns. A time series forecasting model is used to predict the changing trends of key indicators such as network element load and task latency in the near future, in order to anticipate potential congestion risks. When an anomaly is detected or a bottleneck risk is predicted, preset rules are triggered. These preset rules include reducing the distribution rate, adjusting scheduling priority, or sending alarm information to the upper-level operation and maintenance system.
8. The method for concurrent activation of heterogeneous network elements based on AI dynamic scheduling according to claim 1, characterized in that: Step 6 feeds the results data back to the prediction model and scheduling strategy model to achieve continuous model optimization. This includes continuously recording the results data and feeding it back for model evaluation and retraining. The specific process is as follows: Perform offline batch updates: Perform offline retraining on a periodic basis and verify performance. Perform online incremental updates: Employ parameter incremental updates or streaming learning methods to quickly adapt to business changes; Perform model management: control model versions, monitor model performance, and establish rollback mechanisms.
9. A heterogeneous network element concurrent activation system based on AI dynamic scheduling, characterized in that: It includes a work order receiving and splitting module, a task queue management module, a dynamic session pool management module, an AI concurrency prediction module, an AI intelligent scheduling module, an anomaly prediction and monitoring module, and a feedback learning module. The work order receiving and splitting module receives and parses the business work order, and splits it into multiple sub-tasks according to the target network element; The task queue management module uses a task queue that supports priority management to receive the subtasks; The AI concurrency prediction module dynamically determines the optimal concurrency for each network element based on its historical performance data and real-time status using a prediction model. Input the characteristic data for predicting network element concurrency, including: historical response time statistics of the network element, task execution success rate and failure rate, task throughput trend within the recent time window, resource utilization index of the network element itself, protocol type characteristics, time period characteristics, and current queue depth of pending tasks. Prediction model selection: For comprehensive feature correlation analysis of network element concurrency, a regression prediction model is adopted; for data with strong temporal correlation in network element concurrency, a deep prediction model that can capture temporal dependencies is adopted; based on the requirements of prediction speed and accuracy in the actual scenario, a selection or combination between lightweight and complex models is made. The predictive model outputs a suggested value for the optimal concurrency of each network element, providing a reliable reference for scheduling decisions. The AI intelligent scheduling module dynamically decides the task dequeueing order and resource allocation scheme based on the optimal concurrency and the priority attributes of the subtasks through scheduling strategies. The dynamic session pool management module allocates session resources from the session pool according to the resource allocation scheme, and sends task instructions to the target network element in parallel through the protocol adaptation layer; The anomaly prediction and monitoring module monitors and collects execution results: The anomaly prediction and monitoring module monitors execution indicators in real time and predicts anomalies. The feedback learning module feeds the result data back to the prediction model and scheduling strategy model to achieve continuous optimization of the model.
10. A computer-readable medium, characterized in that... The computer-readable medium stores computer instructions, which, when executed by a processor, cause the processor to perform the concurrent activation method for heterogeneous network elements based on AI dynamic scheduling as described in any one of claims 1 to 8.