CN-121984934-A - Dynamic scheduling method, device, equipment and storage medium for core network resources

CN121984934ACN 121984934 ACN121984934 ACN 121984934ACN-121984934-A

Abstract

A method, a device, equipment and a storage medium for dynamically scheduling core network resources are applied to the field of communication network optimization, and the method comprises the steps of generating a dynamic scheduling causal graph of the core network resources according to a knowledge base in the field of core network resource scheduling and acquired core network reference data, performing incremental causal distillation training according to the dynamic scheduling causal graph to generate a light-weight causal model, and performing real-time reasoning according to the light-weight causal model to generate scheduling decisions and interpretable resource scheduling reports. The application can efficiently extract the structured business cause and effect relationship, and improves the operation and maintenance efficiency and the reliability of the system. Meanwhile, the generation time of a scheduling strategy can be effectively reduced, real-time reasoning is realized, flexible and accurate resource scheduling is realized, and resources can be dynamically allocated according to actual demands. And a clearer decision basis and explanation can be provided for operation and maintenance personnel, the operation and maintenance personnel are supported to better understand the system behavior, and the problem is quickly positioned.

Inventors

WAN XIAOYANG
ZHANG SUQIN
YANG XUN
XIONG CHENGLONG
LUO YUNJIE
MEI HUI
YIN HAIQUAN

Assignees

中国移动通信集团江西有限公司
中国移动通信集团有限公司

Dates

Publication Date: 20260505
Application Date: 20260205

Claims (12)

1. The method for dynamically scheduling the core network resources is characterized by comprising the following steps: generating a dynamic scheduling causal graph of the core network resources according to the knowledge base in the core network resource scheduling field and the acquired core network reference data; performing incremental causal distillation training according to the dynamic scheduling causal graph to generate a lightweight causal model; And carrying out real-time reasoning according to the lightweight causal model to generate a scheduling decision and an interpretable resource scheduling report.
2. The method for dynamically scheduling core network resources according to claim 1, wherein the generating a dynamic scheduling causal graph of core network resources according to the knowledge base of the domain of core network resource scheduling and the obtained core network reference data comprises: generating a scheduling causal graph of the core network resources according to the core network resource scheduling domain knowledge base; And carrying out dynamic strengthening update on the scheduling causal graph according to the core network reference data to obtain the dynamic scheduling causal graph.
3. The method for dynamically scheduling core network resources according to claim 2, wherein the generating a scheduling cause and effect graph of the core network resources according to the knowledge base of the domain of scheduling core network resources comprises: performing domain fine adjustment on a preset large model according to the domain knowledge base of the core network resource scheduling to generate a large model of the core network resource scheduling, wherein a domain fine adjustment loss function of the large model of the core network resource scheduling is a weighted sum of text generation loss, graph structure loss and domain rule loss; Optimizing the core network resource scheduling large model based on causal prompt engineering to obtain an optimized core network resource scheduling large model, wherein an output result of the optimized core network resource scheduling large model is a structured triplet in a JavaScript object representation format; And generating the scheduling causal graph according to the output result, wherein the scheduling causal graph comprises a plurality of nodes and a plurality of causal edges.
4. The method for dynamically scheduling core network resources according to claim 3, wherein said dynamically enhancing updating said scheduling cause and effect graph according to said core network reference data to obtain said dynamic scheduling cause and effect graph comprises: acquiring service causal strength corresponding to each causal edge in the scheduling causal graph according to a conditional transfer entropy, text data in a knowledge base of the core network resource scheduling field and the core network reference data; If the number of continuous periods of which the service causal strength is greater than the preset strength is greater than the first preset times, updating the edge confidence of the causal edge according to the service causal strength and the large model confidence of the core network resource scheduling large model; acquiring side adjustment information according to the side confidence coefficient and a preset condition; And adjusting and updating the scheduling causal graph according to the side adjustment information to obtain the dynamic scheduling causal graph.
5. The method for dynamically scheduling core network resources according to claim 4, wherein the obtaining the edge adjustment information according to the edge confidence and the preset condition includes: If the continuous period of the edge confidence coefficient greater than the first confidence coefficient threshold value is greater than the second preset times, determining the edge adjustment information comprises adding a new edge in the causal graph; If the continuous period of the side confidence coefficient smaller than the second confidence coefficient threshold value is larger than a third preset times, determining the side adjustment information comprises deleting the side corresponding to the side confidence coefficient in the causal graph.
6. The method for dynamically scheduling core network resources according to claim 1, wherein said performing incremental causal distillation training according to the dynamic scheduling causal graph to generate a lightweight causal model comprises: constructing a teacher model according to the dynamic scheduling causal graph and a preset deep learning model; According to a preset intervention factor, the dynamic scheduling causal graph is interfered, and a counterfactual sample set is generated; and carrying out causal distillation training on the teacher model according to the anti-real sample to obtain the lightweight causal model.
7. The method for dynamically scheduling core network resources according to claim 6, wherein the performing an intervention on the dynamic scheduling causal graph according to a preset intervention factor, generating a counterfactual sample set, includes: Determining a real-time state of the dynamic scheduling causal graph; According to a preset intervention operator, each designated target node in the dynamic scheduling causal graph is interfered to obtain an intervention value of the target node, and the connection between the target node and a corresponding father node is cut off, wherein when the dynamic scheduling causal graph generates a new node, only the new node is determined to be the target node; and adjusting the values of other nodes related to the target node through the dynamic scheduling causal graph and the intervention value of the target node, and generating the anti-facts sample set.
8. The method for dynamically scheduling core network resources according to claim 1, further comprising: And releasing the scheduling decision gray level to core network equipment.
9. A control apparatus, characterized by comprising: the first processing module is used for generating a dynamic scheduling causal graph of the core network resources according to the core network resource scheduling domain knowledge base and the acquired core network reference data; The second processing module is used for performing incremental causal distillation training according to the dynamic scheduling causal graph to generate a lightweight causal model; And the third processing module is used for carrying out real-time reasoning according to the lightweight causal model and generating a scheduling decision and an interpretable resource scheduling report.
10. An electronic device comprising a processor, a memory and a program stored on the memory and executable on the processor, the program when executed by the processor implementing the steps of the core network resource dynamic scheduling method of any one of claims 1 to 8.
11. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon a computer program which, when executed by a processor, implements the steps of the core network resource dynamic scheduling method according to any of claims 1 to 8.
12. A computer program product comprising computer instructions which, when executed by a processor, implement the steps of the core network resource dynamic scheduling method of any one of claims 1 to 8.

Description

Dynamic scheduling method, device, equipment and storage medium for core network resources Technical Field The present application relates to the field of communications network optimization technologies, and in particular, to a method, an apparatus, a device, and a storage medium for dynamically scheduling resources of a core network. Background The current core network resource scheduling method is characterized by comprising the following technical scheme of static rule engine scheduling, predictive data driving scheduling and traditional knowledge distillation combination. The static rule engine is used for completing core network resource scheduling, wherein the static rule refers to static threshold setting based on data indexes, the rule is obtained through expert experience and analysis, and the rule is configured into a core network resource scheduling system in a manual mode. The system can monitor various data indexes of the core network equipment in real time, and once the indexes reach or exceed a set threshold value, corresponding resource scheduling operation is automatically executed according to a preset rule. The analysis of the core network resource scheduling rules needs to rely on the experience of field experts, and the rule configuration in the system needs to be manually operated, so that human errors are easy to occur. If the rule set is not reasonable, such as the threshold is too high or too low, or a logical conflict between rules, inaccuracy or failure of the resource scheduling may result. And the static rule engine can only schedule according to a fixed threshold value, and cannot sense and adapt to the dynamic change in real time. And predicting data-driven scheduling, namely collecting historical operation data of core network equipment, and predicting key index data of core network resource scheduling by adopting a deep learning model or other prediction algorithms. And analyzing the resource scheduling strategy according to the predicted data performance. The scheme relies on historical data to count correlation, and scheduling failure is caused by misjudgment of the correlation in sudden faults. The training process of the conventional LSTM predictive model or domain-wide model requires a significant amount of computational resources and time. The prediction process of the model also requires complex calculations to be performed in real time, which may have some impact on the performance of the system, especially in cases where fast response is required for resource scheduling. The traditional knowledge distillation is combined, namely a large model with strong performance and higher computational complexity in the professional field of the core network is trained according to the multi-mode data of the core network, and the knowledge learned in the large model is transferred into a light model through a knowledge distillation technology. In the distillation process, the output layer characteristics of the large model, namely the final prediction result or classification probability and other information of the large model on input data, are mainly migrated, so that the distilled light model can approximate to the output performance of the large model as much as possible. The rule setting fails to comprehensively consider business causality of business flow operation and data implication, and only the output layer characteristics of the large model are migrated in the knowledge distillation process, but causal logic of the large model in the decision process is not captured, so that the system interpretability and debugging are difficult. Therefore, how to solve the problems of causal interpretability, real-time performance and continuous evolution of resource scheduling in a dynamic environment becomes a research direction for those skilled in the art. Disclosure of Invention At least one embodiment of the application provides a method, a device, equipment and a storage medium for dynamically scheduling resources of a core network, which are used for solving the problems of causal interpretability, instantaneity and continuous evolution of resource scheduling in a dynamic environment in the prior art. In order to solve the technical problems, the application is realized as follows: In a first aspect, an embodiment of the present application provides a method for dynamically scheduling resources of a core network, including: generating a dynamic scheduling causal graph of the core network resources according to the knowledge base in the core network resource scheduling field and the acquired core network reference data; performing incremental causal distillation training according to the dynamic scheduling causal graph to generate a lightweight causal model; And carrying out real-time reasoning according to the lightweight causal model to generate a scheduling decision and an interpretable resource scheduling report. Specifically, the method for dynamically scheduling cor