Determination method and device for poor-quality root cause and storage medium

Abstract

The invention provides a poor quality root cause determination method and device and a storage medium, and is applied to the technical field of Internet. The method comprises the steps of obtaining original business data carrying poor-quality root cause labels, performing feature extraction on the original business data based on a preset feature extraction algorithm, and determining a corresponding feature vector set; based on an operator corresponding to the original service data, determining a spatial relationship between base stations related to the original service data, and constructing a spatial relationship graph of the base stations based on the spatial relationship; performing graph neural network training based on the feature vector set and the spatial relation graph, and determining a trained neural network model; and determining model parameters based on the neural network model, performing aggregation calculation based on the model parameters to determine federated model parameters, updating the neural network model based on the federated model parameters, and determining a graph federated model for determining the poor-quality root cause. According to the method, the technical effects of improving the data analysis accuracy and improving the model generalization ability are achieved.

Description

Technical Field

[0001] The present application relates to the field of Internet, and in particular to a method, device and storage medium for determining the root cause of poor quality. Background Art

[0002] With the development of Internet technology, network performance management has become a key area of ​​Internet development. In order to optimize the network performance management system and improve user experience, root cause analysis of poor network quality has become an important direction of Internet technology development.

[0003] In the existing root cause analysis methods of poor quality, characteristic indicators are mainly extracted from actual business experience, and classification and identification are performed using rule models or machine learning models, and root cause analysis of poor quality is performed based on the results of classification and identification.

[0004] In the existing technology, due to the use of rule models and machine learning models, it is impossible to fully consider the information loss problem existing in different processes in real business scenarios, and it is impossible to capture the dependency relationship between different base stations; and the existing technology is mainly based on partial regional data for analysis, without considering the joint learning process between regions, lacking joint learning and data fusion, resulting in data island problems; thus, it is impossible to fully utilize data from multiple locations for modeling and analysis, limiting the accuracy and development capabilities of the model; therefore, the existing root cause analysis technology of poor quality has technical problems of low data analysis accuracy and low model generalization ability. Summary of the invention

[0005] The present application provides a method, device and storage medium for determining the root cause of poor quality, so as to solve the technical problems of low data analysis accuracy and low model generalization ability existing in the existing poor quality root cause analysis technology.

[0006] In a first aspect, the present application provides a method for determining the root cause of poor quality, comprising:

[0007] Obtain the original business data carrying the root cause annotation of the poor quality, extract features from the original business data based on a preset feature extraction algorithm, and determine the corresponding feature vector set;

[0008] Based on the operator corresponding to the original service data, determine the spatial relationship between the base stations involved in the original service data, and construct a spatial relationship graph of the base stations based on the spatial relationship;

[0009] Perform graph neural network training based on feature vector sets and spatial relationship graphs to determine the trained neural network model;

[0010] Model parameters are determined based on the neural network model, and federated model parameters are determined based on aggregation calculations of the model parameters. The neural network model is updated based on the federated model parameters, and a graph federation model for determining the root cause of poor quality is determined.

[0011] Optionally, performing graph neural network training based on the feature vector set and the spatial relationship graph to determine a trained neural network model includes:

[0012] Determine graph data corresponding to the spatial relationship graph based on the spatial relationship graph and the feature vector set;

[0013] Train graph neural network based on graph data and obtain the trained neural network model;

[0014] The input data of the neural network model is graph data, and the output data is the feature matrix corresponding to the graph data. The feature matrix is ​​the root cause of the quality difference corresponding to the original business data.

[0015] Optionally, determining model parameters based on the neural network model, performing aggregation calculation based on the model parameters to determine federated model parameters, updating the neural network model based on the federated model parameters, and determining a graph federated model for determining a root cause of poor quality include:

[0016] When the number of iterations of the neural network model is less than a preset number of iterations and the neural network model has not converged, the neural network model is trained based on a preset sample size, a preset sample batch, and a preset number of iterations, and model parameters corresponding to the neural network model at a preset time point are determined based on the neural network model;

[0017] Aggregate calculations are performed based on the model parameters, and the federated model parameters after the aggregate calculations are obtained, and the neural network model is updated based on the federated model parameters.

[0018] Optionally, determining model parameters based on the neural network model, performing aggregation calculation based on the model parameters to determine federated model parameters, updating the neural network model based on the federated model parameters, and determining a graph federated model for determining the root cause of the poor quality further includes:

[0019] When the number of iterations of the neural network model is equal to a preset number of iterations, and / or the neural network model converges, the neural network model is determined as a graph federation model for determining the root cause of poor quality.

[0020] Optionally, original service data carrying the root cause of poor quality is obtained, including:

[0021] Obtain the operator information and business scenario information of the data collection, determine the data requirements for data collection based on the operator information and the business scenario information, determine the definition of the data interface and format of the data collection based on the data requirements, and obtain the corresponding original business data based on the data interface and format;

[0022] The raw data is cleaned and normalized, and the processed raw data is stored in the database.

[0023] Optionally, feature extraction is performed on the original business data based on a preset feature extraction algorithm, and a corresponding feature vector set is determined, including:

[0024] The original business data in the database is obtained, and features are extracted from the original business data based on a preset feature extraction algorithm, and the original business data after feature extraction is determined as a feature vector set.

[0025] Optionally, based on the operator corresponding to the original service data, determining the spatial relationship between base stations involved in the original service data, and constructing a spatial relationship graph of the base stations based on the spatial relationship, including:

[0026] Determine the operator information corresponding to the original service data, determine the base station information corresponding to the original service data based on the operator information, and determine the spatial relationship between the base stations based on the base station information;

[0027] Abstract processing is performed based on the spatial relationship, and the spatial relationship of the base stations is determined as a spatial topological structure, and a spatial relationship graph corresponding to the spatial relationship of the base stations is constructed.

[0028] In a second aspect, the present application provides a device for determining the root cause of poor quality, including:

[0029] An acquisition module is used to acquire the original business data carrying the root cause annotation of the poor quality, extract features from the original business data based on a preset feature extraction algorithm, and determine the corresponding feature vector set;

[0030] A first processing module is used to determine the spatial relationship between base stations involved in the original service data based on the operator corresponding to the original service data, and to construct a spatial relationship graph of the base stations based on the spatial relationship;

[0031] The second processing module is used to perform graph neural network training based on the feature vector set and the spatial relationship graph, and determine the trained neural network model;

[0032] The third processing module is used to determine model parameters based on the neural network model, perform aggregation calculation based on the model parameters to determine the federated model parameters, update the neural network model based on the federated model parameters, and determine the graph federated model used to determine the root cause of the quality difference.

[0033] Optionally, the device is also used to:

[0034] Determine graph data corresponding to the spatial relationship graph based on the spatial relationship graph and the feature vector set;

[0035] Train graph neural network based on graph data and obtain the trained neural network model;

[0036] The input data of the neural network model is graph data, and the output data is the feature matrix corresponding to the graph data. The feature matrix is ​​the root cause of the quality difference corresponding to the original business data.

[0037] Optionally, the device is also used to:

[0038] When the number of iterations of the neural network model is less than a preset number of iterations and the neural network model has not converged, the neural network model is trained based on a preset sample size, a preset sample batch, and a preset number of iterations, and model parameters corresponding to the neural network model at a preset time point are determined based on the neural network model;

[0039] Aggregate calculations are performed based on the model parameters, and the federated model parameters after the aggregate calculations are obtained, and the neural network model is updated based on the federated model parameters.

[0040] Optionally, the device is also used to:

[0041] When the number of iterations of the neural network model is equal to a preset number of iterations, and / or the neural network model converges, the neural network model is determined as a graph federation model for determining the root cause of poor quality.

[0042] Optionally, the device is also used to:

[0043] Obtain the operator information and business scenario information of the data collection, determine the data requirements for data collection based on the operator information and the business scenario information, determine the definition of the data interface and format of the data collection based on the data requirements, and obtain the corresponding original business data based on the data interface and format;

[0044] The raw data is cleaned and normalized, and the processed raw data is stored in the database.

[0045] Optionally, the device is also used to:

[0046] The original business data in the database is obtained, and features are extracted from the original business data based on a preset feature extraction algorithm, and the original business data after feature extraction is determined as a feature vector set.

[0047] Optionally, the device is also used to:

[0048] Determine the operator information corresponding to the original service data, determine the base station information corresponding to the original service data based on the operator information, and determine the spatial relationship between the base stations based on the base station information;

[0049] Abstract processing is performed based on the spatial relationship, and the spatial relationship of the base stations is determined as a spatial topological structure, and a spatial relationship graph corresponding to the spatial relationship of the base stations is constructed.

[0050] A third aspect of the present application provides a device for determining a root cause of poor quality, comprising:

[0051] Processor and memory;

[0052] Memory stores computer-executable instructions;

[0053] The processor executes the computer-executable instructions stored in the memory, so that the device for determining the root cause of poor quality executes the method for determining the root cause of poor quality according to any one of the first aspects.

[0054] In a fourth aspect, the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the method for determining the root cause of quality difference as described in any one of the first aspects.

[0055] The present application provides a method, device and storage medium for determining the root cause of poor quality. The method includes: obtaining original service data carrying the root cause of poor quality annotation, extracting features from the original service data based on a preset feature extraction algorithm, and determining a corresponding feature vector set; determining the spatial relationship between base stations involved in the original service data based on the operator corresponding to the original service data, and constructing a spatial relationship graph of the base stations based on the spatial relationship; training a graph neural network based on the feature vector set and the spatial relationship graph, and determining a trained neural network model; determining model parameters based on the neural network model, and determining the federated model parameters based on the model parameters by aggregate calculation, updating the neural network model based on the federated model parameters, and determining a graph federated model for determining the root cause of poor quality. A feature vector set of the original business data is obtained through a preset feature extraction algorithm. At the same time, a topological graph structure of the spatial relationship of the base stations involved in the original business data is constructed through a graph neural network, and the base stations are determined as nodes of the graph structure. The feature vector set of the original business data is determined as a feature vector matrix corresponding to the nodes in the graph structure. The graph data containing the feature vector matrix is ​​trained based on the graph neural network, and the trained neural network model is determined; the neural network model is further updated by using federated learning, and the update of the neural network model is achieved through the aggregation calculation of the model parameters and the distribution of the results of the aggregation calculation to the neural network models of each participant in the federated learning, so as to determine the final graph federated model for the root cause of the quality difference; the present application uses federated learning to integrate data from various regions for model training, and trains based on a graph neural network, thereby achieving the technical effect of improving the accuracy of data analysis and enhancing the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0057] Figure 1The process of determining the root cause of poor quality provided in the embodiment of the present application Figure 1 ;

[0058] Figure 2 The process of determining the root cause of poor quality provided in the embodiment of the present application Figure 2 ;

[0059] Figure 3 A schematic diagram of constructing a base station spatial relationship diagram for root cause analysis of poor quality provided in an embodiment of the present application;

[0060] Figure 4 Schematic diagram of horizontal federated learning for root cause analysis of poor quality provided in the embodiment of the present application Figure 1 ;

[0061] Figure 5 Schematic diagram of horizontal federated learning for root cause analysis of poor quality provided in the embodiment of the present application Figure 2 ;

[0062] Figure 6 A schematic diagram of the structure of a device for determining the root cause of poor quality provided in an embodiment of the present application;

[0063] Figure 7 A hardware structure diagram of a device for determining the root cause of poor quality provided in an embodiment of the present application.

[0064] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0065] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0066] In the existing root cause analysis method of poor quality, characteristic indicators are mainly extracted from actual business experience, and classified and identified using rule models or machine learning models, and the root cause analysis of poor quality is performed based on the results of classification and identification. In the existing technology, due to the use of rule models and machine learning models, the information loss problem existing in different processes in real business scenarios cannot be fully considered, and the dependency relationship between different base stations cannot be captured; and the existing technology is mainly based on partial regional data for analysis, without considering the joint learning process between regions, lacking joint learning and data fusion, resulting in data island problems; thus, it is impossible to fully utilize data from multiple locations for modeling and analysis, limiting the accuracy and development capabilities of the model; therefore, in the existing root cause analysis technology of poor quality, there are technical problems of low data analysis accuracy and low model generalization ability.

[0067] The present application provides a method, device and storage medium for determining the root cause of poor quality. The method includes: obtaining original service data carrying the root cause of poor quality annotation, extracting features from the original service data based on a preset feature extraction algorithm, and determining a corresponding feature vector set; determining the spatial relationship between base stations involved in the original service data based on the operator corresponding to the original service data, and constructing a spatial relationship graph of the base stations based on the spatial relationship; training a graph neural network based on the feature vector set and the spatial relationship graph, and determining a trained neural network model; determining model parameters based on the neural network model, and determining federated model parameters based on the model parameters through aggregate calculation, updating the neural network model based on the federated model parameters, and determining a graph federated model for determining the root cause of poor quality. A feature vector set of the original business data is obtained through a preset feature extraction algorithm. At the same time, a topological graph structure of the spatial relationship of the base stations involved in the original business data is constructed through a graph neural network, and the base stations are determined as nodes of the graph structure. The feature vector set of the original business data is determined as a feature vector matrix corresponding to the nodes in the graph structure. The graph data containing the feature vector matrix is ​​trained based on the graph neural network, and the trained neural network model is determined; the neural network model is further updated by using federated learning, and the update of the neural network model is achieved through the aggregation calculation of the model parameters and the distribution of the results of the aggregation calculation to the neural network models of each participant in the federated learning, so as to determine the final graph federated model for the root cause of the quality difference; the present application uses federated learning to integrate data from various regions for model training, and trains based on a graph neural network, which solves the technical problems of low data analysis accuracy and low model generalization ability existing in the prior art.

[0068] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0069] Figure 1 The process of determining the root cause of poor quality provided in the embodiment of the present application Figure 1 .like Figure 1 As shown, the method includes:

[0070] S101, obtaining original service data carrying poor quality root cause annotations, extracting features from the original service data based on a preset feature extraction algorithm, and determining a corresponding feature vector set;

[0071] In this embodiment, the acquisition of original business data carrying the root cause of poor quality annotation depends on the currently selected operator and the corresponding business scenario; the acquisition of the preset feature extraction algorithm can be: based on the preset business scenario obtained within a preset time range and a preset spatial range and the historical poor quality work order data corresponding to the preset operator, and the historical poor quality work order data is used to design the preset feature extraction algorithm.

[0072] S102, determining the spatial relationship between base stations involved in the original service data based on the operator corresponding to the original service data, and constructing a spatial relationship graph of the base stations based on the spatial relationship;

[0073] In this embodiment, the spatial relationship of the base stations may include: the distance relationship between the base stations, the propagation direction relationship of the base stations, and the coverage area relationship of the base stations; the connection relationship between the base stations is determined by comprehensively considering various factors; the base stations are determined as nodes of the graph, the connection relationship between the base stations is determined as the edges of the graph and the weights of the edges, and the relationship matrix of the graph is determined through the edges of the base stations, thereby constructing a spatial relationship graph of the base stations.

[0074] S103, performing graph neural network training based on the feature vector set and the spatial relationship graph, and determining a trained neural network model;

[0075] In this embodiment, the feature vector set corresponds to the feature vector of the node in the spatial relationship graph, and the feature vector is used as the feature vector matrix of the spatial relationship graph to represent the node features and perform graph neural network training.

[0076] S104. Determine model parameters based on the neural network model, perform aggregation calculation based on the model parameters to determine federated model parameters, update the neural network model based on the federated model parameters, and determine a graph federated model for determining the root cause of the quality difference.

[0077] In this embodiment, the graph federation model is implemented based on federated learning, wherein federated learning includes horizontal federated learning and vertical federated learning. Different types of federated learning are performed based on the characteristics of the data. Horizontal federated learning is suitable for the case where the data features of the participants overlap a lot but the sample identifications overlap a little, while vertical federated learning is suitable for the case where the training sample identifications of the participants overlap a lot but the data features overlap a little.

[0078] The present application provides a method for determining the root cause of poor quality. The method includes: obtaining original service data carrying the root cause of poor quality annotation, extracting features from the original service data based on a preset feature extraction algorithm, and determining a corresponding feature vector set; determining the spatial relationship between base stations involved in the original service data based on the operator corresponding to the original service data, and constructing a spatial relationship graph of the base stations based on the spatial relationship; training a graph neural network based on the feature vector set and the spatial relationship graph, and determining a trained neural network model; determining model parameters based on the neural network model, and determining federated model parameters based on the model parameters by aggregation calculation, updating the neural network model based on the federated model parameters, and determining a graph federated model for determining the root cause of poor quality. A feature vector set of the original business data is obtained through a preset feature extraction algorithm. At the same time, a topological graph structure of the spatial relationship of the base stations involved in the original business data is constructed through a graph neural network, and the base stations are determined as nodes of the graph structure. The feature vector set of the original business data is determined as a feature vector matrix corresponding to the nodes in the graph structure. The graph data containing the feature vector matrix is ​​trained based on the graph neural network, and the trained neural network model is determined; the neural network model is further updated by using federated learning, and the update of the neural network model is achieved through the aggregation calculation of the model parameters and the distribution of the results of the aggregation calculation to the neural network models of each participant in the federated learning, so as to determine the final graph federated model for the root cause of the quality difference; the present application uses federated learning to integrate data from various regions for model training, and trains based on a graph neural network, which solves the technical problems of low data analysis accuracy and low model generalization ability existing in the prior art.

[0079] Figure 2 The process of determining the root cause of poor quality provided in the embodiment of the present application Figure 2 .like Figure 2 As shown, the method includes:

[0080] S201, obtaining operator information and business scenario information of data collection, determining data requirements for data collection based on the operator information and the business scenario information, determining the definition of the data interface and format of data collection based on the data requirements, and obtaining corresponding original business data based on the data interface and format; performing data cleaning and normalization processing on the original data, and storing the processed original data in a database;

[0081] In this embodiment, the business scenario information may be: mobile communication and data communication.

[0082] In the first exemplary embodiment, a graph federation model corresponding to the root cause of poor quality is trained based on cell data: the data requirements for data collection include: daily granularity cell-level PM data, daily granularity cell-level MR data, cell basic and CM data, cell switching data, and historical work order data of poor quality cells; corresponding data interfaces and formats are defined for each type of data, which may include: field name, poor quality data type, data table, and data unit; corresponding original business data is collected through the data interface and format; data preprocessing is performed based on the original business data, and the preprocessing includes: removing duplicate data, cleaning data, and filling missing data; and the preprocessed data is stored in a database.

[0083] Among them, a cell refers to a specific area in a wireless communication system, which is covered by one or more base stations; each cell has a unique identifier, called a cell ID; cell basic data refers to data that describes the basic information and characteristics of a cell. These data are usually provided by mobile network operators or network equipment providers for management and each cell has a unique identifier to distinguish different cells; configuration management (CM) data refers to data that records and manages network equipment configuration information. CM data includes various configuration parameters and settings of network equipment to ensure the normal operation and performance optimization of network equipment; cell-level measurement report (MR) data refers to data that monitors and evaluates the quality of a mobile network in a cell. MR data is generated by a mobile device when using a mobile network, and records the network performance information of the device at a specific time and location; cell switching data refers to data related to recording and analyzing cell switching of a mobile device in a mobile network. Cell switching refers to the switching of a mobile device from one cell to another to maintain the continuity and stability of the network connection.

[0084] S202, obtaining original business data in a database, performing feature extraction on the original business data based on a preset feature extraction algorithm, and determining the original business data after feature extraction as a feature vector set;

[0085] In this embodiment, in the second demonstration example, network poor quality work order data in mobile network operation is obtained, and a preset feature extraction algorithm is designed based on the experience in processing network poor quality work order data, and the extracted features are standardized to reduce the impact of different dimensions of data; the preset feature extraction algorithm is used to extract features from the original business data in the database, and a feature vector set after feature extraction is obtained.

[0086] S203, determining the operator information corresponding to the original service data, and determining the base station information corresponding to the original service data based on the operator information, and determining the spatial relationship between the base stations based on the base station information; performing abstract processing based on the spatial relationship, and determining the spatial relationship of the base stations as a spatial topological structure, and constructing a spatial relationship graph corresponding to the spatial relationship of the base stations;

[0087] In this embodiment, the spatial relationship based on the base stations is abstracted and converted into a structure of a spatial topology graph; in the third exemplary embodiment, the distance relationship, direction relationship and coverage area relationship of the base stations need to be considered in the process of constructing the component graph, and the weight of each edge between the base stations is determined when the topology graph is constructed.

[0088] In this embodiment, the spatial relationship graph may be constructed by a method including: a K-nearest neighbor method, a minimum spanning tree method, and a shortest path method.

[0089] S204, determining graph data corresponding to the spatial relationship graph based on the spatial relationship graph and the feature vector set; performing graph neural network training based on the graph data, and obtaining a trained neural network model;

[0090] In this embodiment, the input data of the neural network model is graph data, the output data is a feature matrix corresponding to the graph data, and the feature matrix is ​​the root cause of the quality difference corresponding to the original business data;

[0091] In the fourth example, a graph convolutional neural network is used to train the model. The graph data is defined as G = (V, E, A, X), where V is the point set of the graph, corresponding to the base station; E is the edge set of the graph, corresponding to the connection relationship of the base station, that is, the connection constructed by the spatial relationship graph; A∈R n×n It is the adjacency matrix of the graph, which represents the topological relationship of the spatial relationship graph; X∈R n×d It is the eigenvector matrix of the graph nodes, that is, the eigenvector set, in which each node contains a d-dimensional eigenvector; the graph data G is used as the model input data, and the model output data is the root cause of the quality difference. The model is trained by calculating the error between the output data and the node's eigenvector.

[0092] The graph convolutional neural network training includes:

[0093] Define the symmetric normalized Laplacian matrix L = I of the graph data G N -D -1 / 2 AD -1 / 2 =UΛU T , where I N is the identity matrix of the graph, A is the adjacency matrix of the graph, D is the degree matrix of the graph, U is the eigenvector matrix of the normalized graph Laplacian operator, and Λ is the diagonal matrix composed of eigenvalues. Given a signal As input, another signal g w=diag(w), where w∈R n As a filter, the convolution in the spectral domain is defined as:

[0094] g w ★x=Ug w (Λ)U T x

[0095] Among them, w 0 ,w 1 As a parameter, further, by imposing the constraint w = w 0 =-w 1 , the formula can be simplified to:

[0096]

[0097] To solve To solve the gradient explosion and / or vanishing problem, graph convolutional neural networks introduce a renormalization operation: Finally, the concise form of the convolution calculation formula is expressed as:

[0098]

[0099] in, is the input matrix, is the parameter matrix, is the matrix after convolution, F, F ′ are the input dimension and the output dimension respectively, It is the adjacency matrix of the graph data G plus self-connection, that is, each vertex is connected to itself with an edge, I N is the identity matrix, yes The degree matrix of

[0100] S205. When the number of iterations of the neural network model is less than the preset number of iterations and the neural network model has not converged, the neural network model is trained based on a preset sample size, a preset sample batch, and a preset number of iterations, and model parameters corresponding to the neural network model at a preset time point are determined based on the neural network model; aggregate calculation is performed based on the model parameters, and federated model parameters after the aggregate calculation are obtained, and the neural network model is updated based on the federated model parameters;

[0101] In this embodiment, in order to improve the utilization rate of operator data in different regions and eliminate the data island problem between different regions and institutions, this application constructs a graph federation model based on graph neural network to address the root cause of poor operator quality, thereby achieving joint modeling and collaborative optimization of multiple regions and multiple institutions without leaking user privacy.

[0102] In the fifth example, the construction of the graph federation model is implemented based on horizontal federated learning, and the federated average algorithm is used for horizontal federated learning, specifically:

[0103] Determine the optimization goal of the algorithm:

[0104]

[0105] The total number of participants in federated learning is K, ω∈R d represents the model parameters of d dimensions, P k and n k are the sample set and sample number of the kth participant respectively, n is the total number of samples of all participants, is the error calculated by the gradient descent algorithm on the sample set of the kth participant, F k (ω) is the average error calculated by the gradient descent algorithm for the sample set on the kth participant, f(ω) is the mean error calculated by the gradient descent algorithm for all participants in federated learning, and the federated model parameter ω of the graph federation model at time t is set t .

[0106] The specific execution process of the algorithm can be divided into three steps:

[0107] (1) The federation model parameter ω t Sent to participants in this round of training;

[0108] (2) The selected participant uses the gradient descent algorithm on his own local data set, sets the batch size of the data to B, and the number of iterations to E, and calculates the gradient g k Then update the model parameters, specifically:

[0109]

[0110]

[0111] in, The model gradients computed locally for each participant, New model parameters for the kth participant, η is an adjustable learning rate;

[0112] (3) These participants send the new model parameters to the parameter server of federated learning, and the parameter server updates the federated model parameters based on the model parameters returned by each participant:

[0113]

[0114] In this formula, ω without superscript t+1 This is the new round of federated model parameters.

[0115] S206. When the number of iterations of the neural network model is equal to a preset number of iterations, and / or the neural network model converges, the neural network model is determined as a graph federation model for determining the root cause of the quality difference.

[0116] In this embodiment, in the sixth example, a new round of local model training of the participants is performed based on the obtained federated model parameters, and it is determined whether the current number of iterations of the neural network model exceeds the preset number of iterations or the mean error calculated by the gradient descent algorithm for all participants in the federated learning is less than or equal to the preset convergence value, that is, the neural network model converges. If so, the new round of training of the neural network model is stopped, and the neural network model with the latest updated federal model parameters is determined as the graph federation model.

[0117] By executing S201 to S206, federated learning and graph network construction are used to perform model training, thereby improving the expression of data characteristics while increasing data utilization, thereby achieving the technical effect of improving the generalization ability of the model and increasing the accuracy of data analysis.

[0118] Figure 3 A schematic diagram of constructing a base station spatial relationship diagram for root cause analysis of poor quality provided in an embodiment of the present application, Figure 4 Schematic diagram of horizontal federated learning for root cause analysis of poor quality provided in the embodiment of the present application Figure 1 , Figure 5 Schematic diagram of horizontal federated learning for root cause analysis of poor quality provided in the embodiment of the present application Figure 2 .like Figure 3 As shown, the base stations are taken as graph nodes, the spatial relationships between the base stations are determined as the connection relationships between the graph nodes, and a spatial relationship graph of the base stations is constructed based on the connection relationships; as shown Figure 4 and Figure 5 As shown in the figure, the data features of each participant overlap a lot and the sample identifiers overlap a little. Therefore, horizontal federated learning is selected for model training. Each participant must first perform local graph neural network training based on local data and upload the model parameters to the parameter server of federated learning. The parameter server aggregates the model parameters of each participant and forms a federated model parameter. The federated model parameters are distributed to each participant and the participant's neural network model is updated. When the entire training process reaches the preset maximum iteration value or the model training converges, the training process stops and the final graph federated model is obtained.

[0119] The present application provides a method for determining the root cause of poor quality, which includes: obtaining original business data carrying the root cause of poor quality annotation, performing feature extraction on the original business data based on a preset feature extraction algorithm, and determining a corresponding feature vector set; determining the spatial relationship between base stations involved in the original business data based on the operator corresponding to the original business data, and constructing a spatial relationship graph of the base stations based on the spatial relationship; performing graph neural network training based on the feature vector set and the spatial relationship graph, and determining a trained neural network model; determining model parameters based on the neural network model, and performing aggregation calculation based on the model parameters to determine federal model parameters, updating the neural network model based on the federal model parameters, and determining a graph federal model for determining the root cause of poor quality. Compared with the existing technology, the relationship between business data and base stations is expressed in the form of a network topology graph, which improves the accuracy of data feature extraction carrying root cause markers of poor quality. At the same time, graph neural networks are used for model training to extend deep learning to unstructured graph data, enhancing the model's graph structure data mining capabilities. While implementing graph neural network training, federated learning is performed to obtain the original business data from various regions, aggregate the model parameters of the neural network model training in each region, and update the local model, thereby achieving the purpose of joint learning by all parties. Compared with the existing technology, federated learning expands the number of samples of training data while maintaining the confidentiality of data, thereby improving the generalization ability of the model and the accuracy of data analysis. This solves the technical problems of low data analysis accuracy and low model generalization ability in the existing technology.

[0120] Figure 6 Schematic diagram of the structure of the device for determining the root cause of poor quality provided in the embodiment of the present application. The device of this embodiment can be in the form of software and / or hardware. Figure 6 As shown, a device 600 for determining a root cause of poor quality provided by an embodiment of the present application includes: an acquisition module 601, a first processing module 602, a second processing module 603 and a third processing module 604.

[0121] The acquisition module 601 is used to acquire the original service data carrying the root cause of the poor quality, extract features from the original service data based on a preset feature extraction algorithm, and determine a corresponding feature vector set;

[0122] The first processing module 602 is used to determine the spatial relationship between the base stations involved in the original service data based on the operator corresponding to the original service data, and construct a spatial relationship graph of the base stations based on the spatial relationship;

[0123] The second processing module 603 is used to perform graph neural network training based on the feature vector set and the spatial relationship graph, and determine a trained neural network model;

[0124] The third processing module 604 is used to determine model parameters based on the neural network model, perform aggregation calculation based on the model parameters to determine federated model parameters, update the neural network model based on the federated model parameters, and determine a graph federated model for determining the root cause of the quality difference.

[0125] In a possible implementation, the device is further used for:

[0126] Determine graph data corresponding to the spatial relationship graph based on the spatial relationship graph and the feature vector set;

[0127] Train graph neural network based on graph data and obtain the trained neural network model;

[0128] The input data of the neural network model is graph data, and the output data is the feature matrix corresponding to the graph data. The feature matrix is ​​the root cause of the quality difference corresponding to the original business data.

[0129] In a possible implementation, the device is further used for:

[0130] When the number of iterations of the neural network model is less than a preset number of iterations and the neural network model has not converged, the neural network model is trained based on a preset sample size, a preset sample batch, and a preset number of iterations, and model parameters corresponding to the neural network model at a preset time point are determined based on the neural network model;

[0131] Aggregate calculations are performed based on the model parameters, and the federated model parameters after the aggregate calculations are obtained, and the neural network model is updated based on the federated model parameters.

[0132] In a possible implementation, the device is further used for:

[0133] When the number of iterations of the neural network model is equal to a preset number of iterations, and / or the neural network model converges, the neural network model is determined as a graph federation model for determining the root cause of poor quality.

[0134] In a possible implementation, the device is further used for:

[0135] Obtain the operator information and business scenario information of the data collection, determine the data requirements for data collection based on the operator information and the business scenario information, determine the definition of the data interface and format of the data collection based on the data requirements, and obtain the corresponding original business data based on the data interface and format;

[0136] The raw data is cleaned and normalized, and the processed raw data is stored in the database.

[0137] In a possible implementation, the device is further used for:

[0138] The original business data in the database is obtained, and features are extracted from the original business data based on a preset feature extraction algorithm, and the original business data after feature extraction is determined as a feature vector set.

[0139] In a possible implementation, the device is further used for:

[0140] Determine the operator information corresponding to the original service data, determine the base station information corresponding to the original service data based on the operator information, and determine the spatial relationship between the base stations based on the base station information;

[0141] Abstract processing is performed based on the spatial relationship, and the spatial relationship of the base stations is determined as a spatial topological structure, and a spatial relationship graph corresponding to the spatial relationship of the base stations is constructed.

[0142] The present application provides a device for determining the root cause of poor quality. The device includes: an acquisition module, which is used to acquire original business data carrying the root cause of poor quality annotation, perform feature extraction on the original business data based on a preset feature extraction algorithm, and determine the corresponding feature vector set; a first processing module, which is used to determine the spatial relationship between the base stations involved in the original business data based on the operator corresponding to the original business data, and construct a spatial relationship graph of the base stations based on the spatial relationship; a second processing module, which is used to perform graph neural network training based on the feature vector set and the spatial relationship graph, and determine the trained neural network model; a third processing module, which is used to determine model parameters based on the neural network model, and perform aggregate calculation based on the model parameters to determine the federated model parameters, update the neural network model based on the federated model parameters, and determine the graph federated model used to determine the root cause of poor quality. Compared with the existing technology, the relationship between business data and base stations is expressed in the form of a network topology graph, which improves the accuracy of data feature extraction carrying root cause markers of poor quality. At the same time, graph neural networks are used for model training to extend deep learning to unstructured graph data, enhancing the model's graph structure data mining capabilities. While implementing graph neural network training, federated learning is performed to obtain the original business data from various regions, aggregate the model parameters of the neural network model training in each region, and update the local model, thereby achieving the purpose of joint learning by all parties. Compared with the existing technology, federated learning expands the number of samples of training data while maintaining the confidentiality of data, thereby improving the generalization ability of the model and the accuracy of data analysis. This solves the technical problems of low data analysis accuracy and low model generalization ability in the existing technology.

[0143] Figure 7 The hardware structure diagram of the device for determining the root cause of poor quality provided in the embodiment of the present application. Figure 7 As shown, the poor quality root cause determination device 700 includes:

[0144] Processor 701 and memory 702;

[0145] Memory stores computer-executable instructions;

[0146] The processor executes the computer-executable instructions stored in the memory 702, so that the device for determining the root cause of poor quality executes the method for determining the root cause of poor quality as described above.

[0147] It should be understood that the processor 701 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the invention can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor. The memory 702 may include a high-speed random access memory (RAM), and may also include a non-volatile memory (NVM), such as at least one disk storage, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a disk or an optical disk, etc.

[0148] The embodiment of the present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement a method for determining the root cause of quality difference.

[0149] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present application.

[0150] It should be further noted that, although the various steps in the flowchart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowchart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0151] It should be understood that the above-mentioned device embodiments are only illustrative, and the device of the present application can also be implemented in other ways. For example, the division of units / modules in the above-mentioned embodiments is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units, modules or components can be combined, or can be integrated into another system, or some features can be ignored or not executed.

[0152] In addition, unless otherwise specified, each functional unit / module in each embodiment of the present application may be integrated into one unit / module, each unit / module may exist physically separately, or two or more units / modules may be integrated together. The above-mentioned integrated unit / module may be implemented in the form of hardware or in the form of a software program module.

[0153] If the integrated unit / module is implemented in the form of hardware, the hardware may be a digital circuit, an analog circuit, etc. The physical implementation of the hardware structure includes but is not limited to transistors, memristors, etc. If not specifically stated, the processor may be any appropriate hardware processor, such as a CPU, a GPU, an FPGA, a DSP, an ASIC, etc. If not specifically stated, the storage unit may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc.

[0154] If the integrated unit / module is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0155] In the above embodiments, the description of each embodiment has its own emphasis. For the part not described in detail in a certain embodiment, please refer to the relevant description of other embodiments. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0156] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0157] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for determining the root cause of poor quality, characterized in that: include: Acquire original service data carrying the root cause annotation of poor quality, perform feature extraction on the original service data based on a preset feature extraction algorithm, and determine a corresponding feature vector set; Determine, based on the operator corresponding to the original service data, a spatial relationship between base stations involved in the original service data, and construct a spatial relationship graph of the base stations based on the spatial relationship; Performing graph neural network training based on the feature vector set and the spatial relationship graph to determine a trained neural network model; Model parameters are determined based on the neural network model, and federated model parameters are determined based on the model parameters through aggregation calculation. The neural network model is updated based on the federated model parameters to determine a graph federated model for determining the root cause of the quality difference.

2. The method according to claim 1, characterized in that The performing graph neural network training based on the feature vector set and the spatial relationship graph to determine a trained neural network model includes: Determine graph data corresponding to the spatial relationship graph based on the spatial relationship graph and the feature vector set; Performing graph neural network training based on the graph data, and obtaining the trained neural network model; The input data of the neural network model is the graph data, the output data is the feature matrix corresponding to the graph data, and the feature matrix is ​​the root cause of the quality difference corresponding to the original business data.

3. The method according to claim 2, characterized in that The determining of model parameters based on the neural network model, performing aggregation calculation based on the model parameters to determine federated model parameters, updating the neural network model based on the federated model parameters, and determining a graph federated model for determining the root cause of the poor quality include: When the number of iterations of the neural network model is less than a preset number of iterations and the neural network model has not converged, training the neural network model based on a preset sample size, a preset sample batch, and a preset number of iterations, and determining the model parameters corresponding to the neural network model at a preset time point based on the neural network model; Aggregate calculation is performed based on the model parameters, and the federated model parameters after the aggregate calculation are obtained, and the neural network model is updated based on the federated model parameters.

4. The method according to claim 3, characterized in that The method of determining model parameters based on the neural network model, performing aggregation calculation based on the model parameters to determine federated model parameters, updating the neural network model based on the federated model parameters, and determining a graph federated model for determining the root cause of the poor quality further includes: When the number of iterations of the neural network model is equal to the preset number of iterations, and / or when the neural network model converges, the neural network model is determined as the graph federation model for determining the root cause of the quality difference.

5. The method according to claim 1, characterized in that The obtaining of the original service data carrying the root cause of the poor quality includes: Acquire operator information and business scenario information of data collection, determine data requirements for data collection based on the operator information and the business scenario information, determine the definition of the data interface and format of data collection based on the data requirements, and acquire corresponding original business data based on the data interface and format; The raw data is cleaned and normalized, and the processed raw data is stored in a database.

6. The method according to claim 5, characterized in that The extracting features from the original business data based on a preset feature extraction algorithm and determining a corresponding feature vector set includes: The original business data in the database is acquired, and features are extracted from the original business data based on the preset feature extraction algorithm, and the original business data after the feature extraction is determined as the feature vector set.

7. The method according to claim 6, characterized in that The determining, based on the operator corresponding to the original service data, the spatial relationship between the base stations involved in the original service data, and constructing a spatial relationship graph of the base stations based on the spatial relationship, includes: Determine the operator information corresponding to the original service data, determine the base station information corresponding to the original service data based on the operator information, and determine the spatial relationship between the base stations based on the base station information; Abstract processing is performed based on the spatial relationship, and the spatial relationship of the base stations is determined as a spatial topological structure, and the spatial relationship graph corresponding to the spatial relationship of the base stations is constructed.

8. A device for determining the root cause of poor quality, characterized in that: include: An acquisition module, used to acquire original service data carrying the root cause annotation of poor quality, perform feature extraction on the original service data based on a preset feature extraction algorithm, and determine a corresponding feature vector set; A first processing module, configured to determine, based on the operator corresponding to the original service data, a spatial relationship between base stations involved in the original service data, and construct a spatial relationship graph of the base stations based on the spatial relationship; A second processing module is used to perform graph neural network training based on the feature vector set and the spatial relationship graph to determine a trained neural network model; The third processing module is used to determine model parameters based on the neural network model, perform aggregation calculation based on the model parameters to determine federal model parameters, update the neural network model based on the federal model parameters, and determine a graph federal model for determining the root cause of the quality difference.

9. A device for determining the root cause of poor quality, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method for determining the root cause of poor quality as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method for determining the root cause of quality difference according to any one of claims 1 to 7.

Images