CN-122027505-A - Core network quality difference identification method and device based on artificial intelligence and electronic equipment

CN122027505ACN 122027505 ACN122027505 ACN 122027505ACN-122027505-A

Abstract

The application discloses a core network quality difference identification method and device based on artificial intelligence and electronic equipment, belongs to the technical field of communication network optimization, and aims to solve the problems of low accuracy and poor flexibility of core network quality difference identification in the related technology. The method comprises the steps of obtaining multisource data related to quality difference identification of a core network in the core network, carrying out anomaly detection on the multisource data according to a pre-generated dynamic threshold value, primarily identifying quality difference abnormal events in the core network, determining the dynamic threshold value by means of boundary optimization of an initial dynamic threshold value through reinforcement learning, predicting an LSTM (long-short-period memory) model of the pre-trained long-period memory network based on historical multisource data, associating the quality difference abnormal events with a pre-built multidimensional association graph, carrying out reasoning on the associated multidimensional association graph through a pre-trained graph neural network, and outputting core network quality difference constant information, wherein the core network quality difference constant information comprises quality difference root causes.

Inventors

MA ZHILIANG
ZHANG ZHAOXIAN
WANG NAN
ZHANG DI
BI MIN
YANG YULIANG
LI GUANGBO
XU XIAOMING
WANG SHANSHAN
GUO TONGWEN

Assignees

中国移动通信集团北京有限公司
中国移动通信集团有限公司

Dates

Publication Date: 20260512
Application Date: 20260126

Claims (10)

1. The core network quality difference identification method based on artificial intelligence is characterized by comprising the following steps of: Acquiring multi-source data related to quality difference identification of a core network, wherein the multi-source data comprises one or more of resource index data, performance index data and topological structure data; Performing anomaly detection on the multi-source data according to a pre-generated dynamic threshold value, and primarily identifying a quality difference anomaly event in the core network, wherein the dynamic threshold value is determined by a boundary of an initial dynamic threshold value optimized through reinforcement learning, and the initial dynamic threshold value is obtained by predicting a pre-trained long-short-term memory network LSTM model based on historical multi-source data; And correlating the quality difference abnormal event with a pre-constructed multidimensional correlation graph, reasoning the correlated multidimensional correlation graph through a pre-trained graph neural network, and outputting core network quality difference constant information, wherein nodes in the pre-constructed multidimensional correlation graph comprise multiple entities in indexes, equipment, slices and areas in the core network, edges between two nodes represent direct or indirect relations between the two corresponding entities, and the core network quality difference constant information comprises quality difference root causes.
2. The method of claim 1, wherein the anomaly detection is performed on the multi-source data based on a pre-generated dynamic threshold, and wherein the method further comprises, prior to initially identifying a constant event of a quality difference in the core network: acquiring the historical multi-source data, and performing data preprocessing operation on the historical multi-source data to obtain processed first multi-source data, wherein the data preprocessing operation comprises one or more of missing value filling, outlier rejection, multi-granularity time alignment, space granularity alignment and data normalization; performing scene clustering processing on the first multi-source data to divide the first multi-source data into corresponding core network service scenes; inputting a historical data sequence corresponding to each index in the first multi-source data into a pre-trained LSTM model aiming at each index of each core network service scene, and outputting a prediction fluctuation range of the index; generating an initial dynamic threshold of a single upper limit, a single lower limit or a comprehensive upper and lower limit according to the prediction fluctuation range and the historical standard deviation of the index in the service scene of the core network; And optimizing the boundary of the initial dynamic threshold by adopting a reinforcement learning algorithm to obtain the dynamic threshold, wherein the reward function of the reinforcement learning algorithm is the false alarm rate and the false alarm rate of the quality difference abnormal event.
3. The method according to claim 1, wherein the correlating the abnormal quality difference event with a pre-constructed multidimensional correlation graph, and reasoning the correlated multidimensional correlation graph through a pre-trained graph neural network, and before outputting the core network quality difference constant information, the method further comprises: Modeling a plurality of heterogeneous data in performance indexes, physical equipment, service slices and topological connection of a core network into a multidimensional association graph according to the historical multi-source data; The method comprises the steps of determining that the weight of an edge between two nodes is a first set value if a physical link relation exists between the two corresponding entities, performing Granger causal test if the physical link relation does not exist between the two corresponding entities, determining that the weight is a second set value if the Granger causal test passes, and calculating the linear correlation between the two corresponding entities by using a Pearson correlation coefficient as the weight if the Granger causal test does not pass.
4. The method of claim 1, wherein the quality difference root cause comprises a plurality of quality difference cause, and the core network quality difference common information further comprises a quality difference common contribution score for each of the quality difference root causes; The correlation of the quality difference abnormal event and the pre-constructed multidimensional correlation graph is carried out, the correlation multidimensional correlation graph is inferred through a pre-trained graph neural network, and the quality difference constant information of a core network is output, and the method comprises the following steps: Marking abnormal nodes in the pre-constructed multidimensional association graph according to the quality difference abnormal event, and determining abnormal characteristics of the abnormal nodes according to the quality difference abnormal event; and inputting the multidimensional association graph marked with the abnormal nodes and the abnormal characteristics into the pre-trained graph neural network, calculating the attention weight of the abnormal nodes to the neighbor nodes through the graph neural network, carrying out message transmission and node characteristic updating through a learnable weight matrix, and outputting the quality difference constant contribution degree score of each quality difference root factor.
5. The method according to claim 4, wherein the correlating the abnormal quality difference event with a pre-constructed multidimensional correlation graph, and reasoning the correlated multidimensional correlation graph through a pre-trained graph neural network, and outputting core network quality difference constant information, the method further comprises: sorting the quality difference common contribution degree scores of the quality difference root factors to determine quality difference abnormal contribution degree sorting results; Generating a quality difference alarm according to the quality difference constant contribution degree sequencing result; Automatically pushing the quality difference alarm to an operation and maintenance work order system, marking the quality difference alarm as a work order to be confirmed, and enabling operation and maintenance personnel to confirm and input root cause analysis conclusion to form a quality difference sample confirmed manually; Receiving the quality difference sample fed back by the operation and maintenance personnel, wherein the quality difference sample comprises quality difference confirmation information and/or root cause analysis conclusion of the operation and maintenance personnel aiming at the quality difference alarm mark, and the quality difference confirmation information comprises true substantial difference or false alarm; And adjusting model parameters of a pre-training model according to the quality difference sample, wherein the pre-training model comprises the LSTM model and the graph neural network.
6. The method of claim 5, wherein adjusting model parameters of a pre-trained model based on the quality difference samples comprises: Updating a first weight matrix and bias by combining incremental training with an elastic weight consolidation mechanism based on the quality difference sample aiming at the LSTM model; And updating the attention parameter and the second weight matrix by gradient back propagation based on the quality difference sample with the weight after the weight is configured by utilizing a loss function with the weight.
7. The utility model provides a core network matter is poor recognition device based on artificial intelligence which characterized in that includes: The acquisition module is used for acquiring multi-source data related to the quality difference identification of the core network in the core network, wherein the multi-source data comprises one or more of resource index data, performance index data and topological structure data; the system comprises an anomaly identification module, a quality difference detection module and a quality difference detection module, wherein the anomaly identification module is used for carrying out anomaly detection on the multi-source data according to a pre-generated dynamic threshold value, and primarily identifying a quality difference abnormal event in the core network; the model reasoning module is used for correlating the quality difference abnormal event with a pre-constructed multidimensional correlation graph, reasoning the correlated multidimensional correlation graph through a pre-trained graph neural network and outputting core network quality difference constant information, wherein nodes in the pre-constructed multidimensional correlation graph comprise multiple entities in indexes, equipment, slices and areas in the core network, edges between the two nodes represent direct or indirect relations between the two corresponding entities, and the core network quality difference constant information comprises quality difference root causes.
8. An electronic device, comprising: Processor, and A memory arranged to store computer executable instructions configured to be executed by the processor to implement the artificial intelligence based core network quality difference identification method of any of claims 1-6.
9. A computer readable storage medium storing computer executable instructions which when executed by a processor implement the artificial intelligence based core network quality difference identification method of any of claims 1-6.
10. A computer program product comprising a computer program which, when executed by a processor, implements an artificial intelligence based core network quality difference identification method as claimed in any of claims 1-6.

Description

Core network quality difference identification method and device based on artificial intelligence and electronic equipment Technical Field The application belongs to the technical field of communication network optimization, and particularly relates to a core network quality difference identification method and device based on artificial intelligence and electronic equipment. Background The quality difference of the core network is ‌, the performance of the core network is deteriorated, and in operation and maintenance practice, ‌ specific technical index abnormality in the core network can provide an evaluation basis for the quality difference of the core network. Specifically, the core network serves as a central of the communication network and is responsible for key functions such as call control, user data routing and the like, and performance degradation (such as equipment overload, link congestion, configuration errors or hardware faults) of the core network can directly cause a large-scale service interruption, which is manifested as call disconnection, call establishment failure, data transmission delay or packet loss and the like, so as to influence user perception. "quality difference" often refers to a set of quantifiable technical indicators that are used to evaluate network quality. When the indexes exceed a preset threshold, the quality difference state is judged, and the quality difference state is an important basis for the operators to perform network optimization. In the field of quality difference identification of a communication network core network, related technologies generally rely on fixed thresholds and static rules set based on network quality key indicators to achieve quality difference identification. The indexes, rules and thresholds are usually completed through manual configuration, and the updating of the rules is seriously dependent on the experience judgment of an expert, so that the adjustment process is complicated, the flexibility of quality difference identification is limited, and the accuracy and the credibility of the quality difference identification are also influenced. In view of the above, the improvement of the dynamic adaptability and accuracy of the quality difference identification of the core network, and the realization of the real-time detection and dynamic optimization of the quality difference abnormality become the problem to be solved in the current technical field. Disclosure of Invention The embodiment of the application provides a core network quality difference identification method and device based on artificial intelligence and electronic equipment, which can solve the problems of low accuracy and poor flexibility of core network quality difference identification in the related technology. The embodiment of the application provides an artificial intelligence-based core network quality difference identification method, which comprises the steps of obtaining multi-source data related to core network quality difference identification in a core network, wherein the multi-source data comprise one or more of resource index data, performance index data and topological structure data, carrying out anomaly detection on the multi-source data according to a pre-generated dynamic threshold value, initially identifying a quality difference abnormal event in the core network, determining the dynamic threshold value by strengthening learning and optimizing a boundary of an initial dynamic threshold value, obtaining the initial dynamic threshold value by predicting a pre-trained long-short-term memory network LSTM model based on historical multi-source data, carrying out correlation on the quality difference abnormal event and a pre-constructed multi-dimensional correlation graph, carrying out reasoning on the correlated multi-dimensional correlation graph through a pre-trained graph neural network, and outputting core network quality difference constant information, wherein nodes in the pre-constructed multi-dimensional correlation graph comprise indexes, equipment, slices and multiple entities in a region in the core network, edges between two nodes represent corresponding two entities, and the direct quality difference constant root-cause information is included in the core network. The embodiment of the application provides a core network quality difference identification device based on artificial intelligence, which comprises an acquisition module, an anomaly identification module, a model reasoning module and a model reasoning module, wherein the acquisition module is used for acquiring multi-source data related to core network quality difference identification in a core network, the multi-source data comprises one or more of resource index data, performance index data and topological structure data, the anomaly identification module is used for carrying out anomaly detection on the multi-source data according to a pre-generated dynamic threshold, the