Wireless cell poor quality root cause positioning method, device, equipment, medium and product

Abstract

The invention provides a poor-quality root cause positioning method and device of a wireless cell, equipment, a medium and a product, and relates to the technical field of wireless communication. The method comprises the steps that operation index data sets of multiple wireless cells, identification information of wireless cells with poor quality and a network topological graph are input into a poor-quality root cause positioning model, poor-quality root causes of the wireless cells with poor quality output by the poor-quality root cause positioning model are obtained, and the network topological graph comprises the connection relation of the multiple wireless cells. According to the invention, learning and extraction of the time characteristics of the operation indexes of the wireless cell are realized through the time sequence network. Meanwhile, through the network topological graph, the spatial characteristics among the wireless cells are obtained. According to the method, the time features and the space features of the operation indexes are combined, accurate positioning of the poor-quality root cause is achieved, and the accuracy of determining the poor-quality root cause of the wireless cell is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, and in particular to a wireless cell quality difference root cause positioning method, device, equipment, medium and product. BACKGROUND

[0002] With the large-scale commercialization of the fifth generation mobile communication technology (5G) and the continuous growth of user scale, the wireless access side of the mobile communication network has formed a complex network architecture with wide coverage and high device density. In order to meet the demand of users for high bandwidth and low latency services, the wireless network usually divides the geographical area into multiple logical cells, each cell is managed by an independent wireless device, and adjacent cells are seamlessly accessed through overlapping coverage. However, with the rapid advancement of urban construction and the dynamic changes of user distribution, the number of wireless cells in a single area has increased from hundreds in the early days to tens of thousands, and the neighbor relationship has become extremely complex. This large-scale and high-complexity network environment leads to frequent quality difference problems (such as weak coverage, high drop rate, low speed, frequent handover failure, etc.), and the base station retreat phenomenon is also on the rise. However, in the face of a large number of cells and complex neighbor interactions, it is difficult for operation and maintenance personnel to efficiently sort out the correlation between the running state of each cell and the quality difference, and troubleshooting often relies on experience and trial and error, resulting in long processing period and high repeat order rate, which not only increases the network operation and maintenance cost, but also causes the number of user complaints to rise due to the delay in solving the problem, seriously affecting user experience and service quality of the operator.

[0003] In view of the demand for wireless cell quality difference root cause positioning, the existing technology mainly proposes the following solutions: first, a text mining method based on operation and maintenance work order, which uses natural language processing (LLP) technology to perform word segmentation, entity recognition and sentiment analysis on the fault description (such as "user feedback slow rate" and "handover failure") in the historical work order, extracts high-frequency keywords and establishes a mapping relationship with the fault type, for example, "handover failure" is highly associated with "neighbor missing configuration" or "handover parameter configuration error". The second is a rule-based alarm linkage system, which checks the preset neighbor relationship rules (such as PCI uniqueness check and neighbor list integrity verification) and KPI threshold (such as RSRP≤-105dBm triggering weak coverage alarm), and when the rule violation or threshold is detected, it automatically associates the real-time indicators of the related cells and generates a preliminary diagnosis report.

[0004] In the existing wireless cell quality difference root cause positioning, many wireless cells have a large number of users and a large flow of users, making it difficult to process each cell user. Although some technical solutions analyze the running data of the wireless cell and the wireless operation and maintenance work order to a certain extent, these solutions only analyze the causal relationship between the indicators and do not obtain specific quality difference root causes, or simply extract the results directly from the wireless operation and maintenance work order. In summary, the existing wireless cell quality difference root cause positioning is not accurate. SUMMARY

[0005] The present application provides a wireless cell quality difference root cause positioning method, device, equipment, medium and product, to solve the defects of inaccurate wireless cell quality difference root cause positioning in the prior art, and to improve the accuracy of wireless cell quality difference root cause positioning.

[0006] In a first aspect, the present application provides a wireless cell quality difference root cause positioning method, comprising: inputting the running index data set of a plurality of wireless cells, the identification information of the wireless cell with quality difference, and the network topology graph into the quality difference root cause positioning model, and obtaining the quality difference root cause of the wireless cell with quality difference output by the quality difference root cause positioning model, wherein the network topology graph comprises the connection relationship between the plurality of wireless cells and the wireless cell with quality difference; The quality difference root cause positioning model comprises a time series network and a root cause positioning module, the time series network is used to obtain the running characteristics of each wireless cell based on the running index data set, and the root cause positioning module is used to determine the quality difference root cause based on the running characteristics of each wireless cell, the identification information and the network topology graph.

[0007] In one embodiment, the root cause positioning module comprises a graph neural network and a fully connected network: The graph neural network is used to obtain the comprehensive running characteristics of the wireless cell with quality difference based on the running characteristics of each wireless cell, the identification information and the network topology graph, wherein the comprehensive running characteristics comprise the running characteristics of the wireless cell with quality difference and the associated characteristics that have an impact on the running characteristics of the wireless cell with quality difference; The fully connected network is used to determine the quality difference root cause based on the comprehensive running characteristics.

[0008] In one embodiment, the running index data set comprises a running index data subset of each wireless cell, and the running index data subset comprises a plurality of historical running indexes carrying timestamps, and the time series network is used to: obtain the historical index time series of each wireless cell, wherein the historical index time series is obtained by sorting the historical running indexes in each running index data subset based on the timestamps; encode and decode the historical index time series of each wireless cell, and predict the future index time series of each wireless cell in a future time period; obtain the running characteristics of each wireless cell based on each future index time series.

[0009] In one embodiment, the graph neural network comprises a plurality of graph calculation layers connected in sequence, and the graph neural network is used to: For the first graph computing layer, a first layer output feature of the target wireless cell is determined based on a running feature of the target wireless cell and a first layer associated node feature of the target wireless cell; a first layer output feature of each wireless cell is obtained based on the first layer output features of the target wireless cells; the target wireless cell is any wireless cell, and the first layer associated node feature is determined based on a running feature of an associated wireless cell, the associated wireless cell being determined based on a network topology graph, and the associated wireless cell being connected to the target wireless cell; For the L+1th graph computing layer, an L+1th layer input feature of each wireless cell is obtained based on the Lth layer output features of the wireless cells; an L+1th layer output feature of the target wireless cell is determined based on the L+1th layer input feature of the target wireless cell and an L+1th layer associated node feature of the target wireless cell; an L+1th layer output feature of each wireless cell is obtained based on the L+1th layer output features of the target wireless cells; until a last layer output feature of each wireless cell is obtained, L≥1, and the L+1th layer associated node feature of the target wireless cell is determined based on an L+1th layer input feature of an associated wireless cell; In the last layer output features of the wireless cells, the last layer output feature of the wireless cell with the quality difference is taken as the comprehensive running feature based on the identification information.

[0010] In an embodiment, the fully connected network is used for: The comprehensive running feature is subjected to multiple feature extraction, feature fusion and feature conversion, and the probability of a specified root cause is obtained based on a last conversion result; The specified root cause with the maximum probability is taken as the quality difference root cause.

[0011] In an embodiment, the quality difference root cause positioning model is obtained based on the following manner: A sample running index data subset is obtained according to a running index data subset of a current time period and a running index data subset of a next time period of a sample wireless cell; The original time series network is trained according to the sample running index data subset until a first loss function of the original time series network meets a first set value, and a time series network is obtained, the first loss function being constructed based on a mean square error of a predicted running index data subset of the next time period output by the original time series network and the running index data subset of the next time period; A preset model is constructed based on the time series network and an initial root cause positioning module; The sample running index data set of the plurality of sample wireless cells, the identification information of the sample wireless cell with the quality difference and the sample network topology graph are marked according to the label of the sample quality difference root cause of the sample wireless cell with the quality difference, and a training sample carrying the label is obtained. The preset model is trained on the training samples with labels until the second loss function of the preset model meets the second set value, thus obtaining the poor quality root cause localization model. The second loss function is determined based on the cross-entropy loss value of the predicted poor quality root cause output by the preset model and the label of the poor quality root cause of the sample.

[0012] Secondly, this application provides a device for locating the root cause of poor wireless cell quality, comprising: The poor quality root cause localization module is used to input the operational indicator datasets of multiple wireless cells, the identification information of wireless cells with poor quality, and the network topology map into the poor quality root cause localization model, and obtain the poor quality root cause of the wireless cells with poor quality output by the poor quality root cause localization model. The network topology map includes the connection relationship between multiple wireless cells and wireless cells with poor quality. The poor quality root cause localization model includes a time series network and a root cause localization module. The time series network is used to obtain the operational characteristics of each wireless cell based on the operational indicator dataset, and the root cause localization module is used to determine the poor quality root cause based on the operational characteristics, identification information, and network topology map of each wireless cell.

[0013] Thirdly, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for locating the root cause of poor quality in any of the above-described wireless cells.

[0014] Fourthly, this application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for locating the root cause of poor quality in any of the wireless cells described above.

[0015] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements a method for locating the root cause of poor quality in any of the wireless cells described above.

[0016] This application provides a method, apparatus, device, medium, and product for locating the root causes of poor wireless cell quality. By inputting a dataset of operational indicators from multiple wireless cells, identification information of wireless cells with poor quality, and a network topology map into a quality-poor-cause localization model, the model outputs the root causes of poor quality in the wireless cells with poor quality. The network topology map includes the connections between multiple wireless cells and the wireless cells with poor quality. The quality-poor-cause localization model includes a time-series network and a root cause localization module. The time-series network is used to obtain the operational characteristics of each wireless cell based on the operational indicator dataset, and the root cause localization module is used to determine the root causes of poor quality based on the operational characteristics, identification information, and network topology map of each wireless cell. This application achieves the learning and extraction of the temporal characteristics of the operational indicators of wireless cells through a time-series network. Simultaneously, it obtains the spatial characteristics between wireless cells through the network topology map. By combining the temporal and spatial characteristics of the operational indicators, this application achieves accurate localization of the root causes of poor quality, improving the accuracy of determining the root causes of poor quality in wireless cells. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is one of the flowcharts illustrating the method for locating the root cause of poor wireless cell quality provided in this application.

[0019] Figure 2 This is the second flowchart of the method for locating the root cause of poor wireless cell quality provided in this application.

[0020] Figure 3 This is the third flowchart of the method for locating the root cause of poor quality in wireless cells provided in this application.

[0021] Figure 4 This is a schematic diagram of the structure of the wireless cell poor quality root cause localization device provided in this application.

[0022] Figure 5 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] The application areas of this application include the root cause localization of poor wireless cell quality. For ease of understanding, the relevant terms and concepts that may be involved in the embodiments of this application will be introduced below.

[0025] A wireless cell is a logical coverage area defined by wireless receiving and transmitting equipment (such as base stations, antennas, etc.) at the end of a mobile communication network (e.g., the wireless access side). It is the basic unit by which the network provides wireless connectivity services to users.

[0026] Time series networks are a type of neural network architecture specifically designed for processing and analyzing time series data. They capture the temporal dependencies between data points, meaning that the data at the current moment is not only related to the input at the current moment but also to data from a past period.

[0027] Fully connected networks, also known as multilayer perceptrons (MLPs), are a fundamental and widely used artificial neural network architecture. A fully connected network consists of an input layer, one or more hidden layers, and an output layer. Between adjacent network layers, each neuron is connected to all neurons in the next layer via learnable weight parameters, forming a dense connection structure.

[0028] The following is combined with Figures 1-5 This application describes a method, apparatus, and electronic device for locating the root cause of poor quality in wireless cells.

[0029] Figure 2 This is the second flowchart illustrating the method for locating the root cause of poor wireless cell quality provided in this application. Figure 2 As shown, the method for locating the root cause of poor quality in a wireless cell includes step S100.

[0030] S100: Input the operational indicator datasets of multiple wireless cells, the identification information of wireless cells with poor quality, and the network topology map into the poor quality root cause localization model, and obtain the poor quality root cause of the wireless cells with poor quality output by the poor quality root cause localization model. The network topology map includes the connection relationship between multiple wireless cells and wireless cells with poor quality.

[0031] The poor quality root cause localization model includes a time series network and a root cause localization module. The time series network is used to obtain the operational characteristics of each wireless cell based on the operational indicator dataset, and the root cause localization module is used to determine the root cause of poor quality based on the operational characteristics, identification information and network topology of each wireless cell.

[0032] Poor quality refers to the poor quality state exhibited by a wireless cell during operation, which may include poor communication quality, poor network performance, and poor user experience.

[0033] Multiple wireless cells include those with poor signal quality and those connected to them. The operational metrics dataset consists of a subset of operational metrics data for each wireless cell. This subset includes historical operational metrics for the wireless cell over a given time period. Historical operational metrics include various indicators such as operational performance, engineering parameters, and interference parameters, totaling over 1000 fields. These historical metrics are divided according to the wireless cell's identifier to obtain the operational metrics data subset for each wireless cell.

[0034] like Figure 1 As shown, the operational metric dataset is preprocessed. The wireless maintenance work order, the preprocessed operational metric dataset, and the network topology map are input into the poor quality root cause localization model to obtain the root causes of poor quality in the wireless cells output by the model. The wireless maintenance work order carries the identification information of the wireless cells with poor quality. The wireless maintenance work order is used to enable the poor quality root cause localization model to identify wireless cells with poor quality.

[0035] A network topology graph is used to represent the connections between multiple wireless cells and wireless cells with poor quality. A network topology graph includes multiple nodes, each representing a wireless cell. When two wireless cells are connected, they are connected by an edge. The network topology graph is obtained based on the multiple nodes and edges.

[0036] The operational metric dataset and network topology graph are input into the poor quality root cause localization model. The time series network of the poor quality root cause localization model learns the temporal features of each subset of operational metric data, thereby obtaining the operational features of each radio cell. The time series network sends the operational features of each radio cell to the root cause localization module (including graph neural networks and fully connected networks).

[0037] The root cause localization module obtains the operational characteristics of wireless cells with poor quality based on the operational characteristics of each wireless cell. It also extracts related features (spatial features) that influence the operational characteristics of wireless cells with poor quality, based on the operational characteristics of each wireless cell and the network topology. Finally, the root cause localization module analyzes the root causes of the poor quality in these wireless cells based on their operational characteristics and related features.

[0038] Furthermore, based on the root causes of the quality issues identified through analysis, corresponding handling measures are determined to achieve efficient processing of wireless maintenance work orders.

[0039] The method for locating the root cause of poor quality in wireless cells provided in this application involves inputting a dataset of operational indicators from multiple wireless cells, identification information of wireless cells with poor quality, and a network topology map into a quality-poor-quality root cause localization model. The model then outputs the root causes of poor quality in the wireless cells with poor quality. The network topology map includes the connections between multiple wireless cells and the wireless cells with poor quality. The quality-poor-quality root cause localization model includes a time-series network and a root cause localization module. The time-series network is used to obtain the operational characteristics of each wireless cell based on the operational indicator dataset. The root cause localization module is used to determine the root cause of poor quality based on the operational characteristics, identification information, and network topology map of each wireless cell. This application achieves the learning and extraction of the temporal characteristics of the operational indicators of wireless cells through a time-series network. Simultaneously, it obtains the spatial characteristics between wireless cells through the network topology map. By combining the temporal and spatial characteristics of the operational indicators, this application achieves accurate localization of the root cause of poor quality, improving the accuracy of determining the root cause of poor quality in wireless cells.

[0040] Based on the above embodiments, the operational indicator dataset includes a subset of operational indicator data for each wireless cell. This subset includes multiple historical operational indicators carrying timestamps. The time-series network is used to perform the following operations: The historical indicator time series of each wireless cell is obtained. The historical indicator time series is obtained by sorting the historical operating indicators in each subset of operating indicator data based on timestamps. The historical time series indicators of each wireless cell are encoded and decoded to predict the future time series indicators of each wireless cell in the future time period. Based on the time series of various future indicators, the operational characteristics of each wireless cell are obtained.

[0041] The time-series network of this application includes an encoder-decoder based transformer structure. For example, the time-series network of this application includes a 12-layer encoder and a 12-layer decoder.

[0042] Each radio cell's operational metric data subset includes multiple historical operational metrics for the radio cell, each with a timestamp. For each subset, historical operational metrics at the same point in time are concatenated into a historical metric vector based on their timestamps. For each subset, the historical metric vectors are sorted by timestamps to obtain the historical metric time series for the radio cell.

[0043] Furthermore, if there are missing historical operational indicators in the historical indicator time series of the wireless cell, the time period of the missing historical operational indicator is removed to obtain a preprocessed historical indicator time series. The preprocessed historical indicator time series is then input into the time series network.

[0044] The encoder of the time series network encodes the preprocessed historical indicator time series to obtain the encoded historical indicator time series. Based on the decoder of the final layer of the time series network, the future indicator time series of the wireless cell within a future time period is obtained. This future indicator time series is then used as the operational characteristic of the wireless cell.

[0045] Time series networks are used to predict the future time series of indicators of a wireless cell within a future time period based on the historical time series of indicators of the cell. For example, a time series network can predict the future time series of indicators of a wireless cell for the next day based on the historical time series of indicators of the cell for the previous 4 days.

[0046] Based on the time series of various future indicators, the operational characteristics of each wireless cell are obtained.

[0047] This application uses a time series network to encode and decode historical indicator time series, thereby achieving accurate extraction of the time characteristics of a subset of operational indicator data from wireless cells.

[0048] Based on the above embodiments, the root cause localization module includes a graph neural network and a fully connected network.

[0049] Graph neural networks are used to obtain the comprehensive operational characteristics of wireless cells with poor quality based on the operational characteristics, identification information, and network topology of each wireless cell. The comprehensive operational characteristics include the operational characteristics of wireless cells with poor quality and the correlation characteristics that affect the operational characteristics of wireless cells with poor quality.

[0050] Fully connected networks are used to identify the root causes of poor quality based on comprehensive operational characteristics.

[0051] The graph neural network performs multi-layer feature fusion and extraction on the operational and related features of each wireless cell to obtain the comprehensive operational features of each wireless cell.

[0052] Based on the wireless operation and maintenance work order, the characteristics (identification information) of wireless cells with poor quality are determined. These characteristics are then input into a graph neural network. The graph neural network identifies the wireless cells with poor quality based on their characteristics and sends the comprehensive operational characteristics of these cells to the fully connected network.

[0053] The fully connected network identifies the root cause of poor wireless quality based on the comprehensive operational characteristics of the poor-quality wireless cells.

[0054] This application uses graph neural networks to obtain comprehensive operational characteristics, which considers both the operational characteristics of radio cells with poor quality and the related characteristics that affect the operational characteristics of radio cells. This improves the comprehensiveness of the comprehensive operational characteristics and helps to improve the accuracy of subsequent determination of the root causes of poor quality.

[0055] Based on the above embodiments, the graph neural network includes multiple sequentially connected graph computation layers, and the graph neural network is used for: For the first graph computation layer, the first-layer output characteristics of the target wireless cell are determined based on the operational characteristics of the target wireless cell and the first-layer associated node characteristics of the target wireless cell; the first-layer output characteristics of each wireless cell are obtained based on the first-layer output characteristics of each target wireless cell; the target wireless cell is any wireless cell, the first-layer associated node characteristics are determined based on the operational characteristics of the associated wireless cells, and the associated wireless cells are determined based on the network topology graph and are wireless cells connected to the target wireless cell. For the (L+1)th graph computation layer, based on the Lth layer output features of each wireless cell, the (L+1)th layer input features of each wireless cell are obtained; based on the (L+1)th layer input features of the target wireless cell and the (L+1)th layer associated node features of the target wireless cell, the (L+1)th layer output features of the target wireless cell are determined; based on the (L+1)th layer output features of each target wireless cell, the (L+1)th layer output features of each wireless cell are obtained; until the last layer output features of each wireless cell are obtained, L≥1, and the (L+1)th layer associated node features of the target wireless cell are determined based on the (L+1)th layer input features of the associated wireless cells; Based on the identification information, the last-layer output characteristics of wireless cells with poor quality are used as comprehensive operational characteristics in the last-layer output characteristics of each wireless cell.

[0056] Graph neural networks are designed based on network topology graphs. A graph neural network includes multiple sequentially connected graph computation layers. For example, a graph neural network may include three (L=2) sequentially connected graph computation layers. This embodiment will illustrate this using a graph neural network with three sequentially connected graph computation layers as an example.

[0057] The operational characteristics of each wireless cell are used as the first-layer input features for each wireless cell. The input data for the first graph computation layer includes the first-layer input features (operational characteristics) of each wireless cell. For example... Figure 3 As shown, the first graph computation layer constructs the initial vector of each wireless cell in the network topology graph based on the operational characteristics of each wireless cell. Figure 3 In this context, the wireless cells include H1, H2, H3, H4, H5, H6, H7, and H8. H4 is a wireless cell with poor quality, identified based on a wireless maintenance work order. For each target wireless cell, the associated wireless cells connected to the target wireless cell are obtained in the network topology diagram. Based on the operational characteristics (initial vector) of the associated wireless cells, the first-layer associated node characteristics of the target wireless cell are calculated. Based on the operational characteristics (first-layer input characteristics) and the first-layer associated node characteristics of each target wireless cell, the first-layer output characteristics of each target wireless cell are determined, until the first-layer output characteristics of all wireless cells are obtained. For example, with H4 as the target wireless cell, the associated wireless cells of H4 include H1, H2, H3, H5, and H7. Based on the operational characteristics of H1, H2, H3, H5, and H7, the first-layer associated node characteristics of H4 are calculated. Based on the operational characteristics of H4 and the first-layer associated node characteristics of H4, the first-layer output characteristics of H4 are calculated. Taking H7 as the target radio cell, its associated radio cells include H4, H6, and H8. Based on the operational characteristics of H4, H6, and H8, the first-layer associated node characteristics of H7 are calculated. Based on the operational characteristics of H7 and the first-layer associated node characteristics of H7, the first-layer output characteristics of H7 are calculated. Following the above method, the first-layer output characteristics of each radio cell are calculated.

[0058] The first-layer output features of each wireless cell are used as the second-layer input features. The input data for the second graph calculation layer includes the second-layer input features of each wireless cell. For each target wireless cell, the second-layer associated node features of the target wireless cell are calculated based on the second-layer input features of the associated wireless cells. The second-layer output features of each target wireless cell are calculated based on the second-layer input features and the second-layer associated node features of each target wireless cell, until the second-layer output features of each wireless cell are obtained.

[0059] The second-layer output features of each wireless cell are used as the third-layer input features of each wireless cell. The input data for the third graph calculation layer includes the third-layer input features of each wireless cell. For each target wireless cell, the third-layer associated node features of the target wireless cell are calculated based on the third-layer input features of the associated wireless cells. Based on the third-layer input features and the third-layer associated node features of each target wireless cell, the third-layer output features of each target wireless cell are calculated until the third-layer output features of each wireless cell are obtained.

[0060] The (L+1)th layer associated node features of the target wireless cell are determined based on the (L)th layer output features of its associated wireless cells. The (L+1)th layer output features of the target wireless cell are determined based on its (L+1)th layer input features and (L+1)th layer associated node features. The calculation formulas for the (L+1)th layer associated node features and (L+1)th layer output features of the target wireless cell are as follows.

[0061] ; in, For the first The L+1 layer input features of the associated wireless cells For the first The Lth layer output features of the associated wireless cells For the target wireless cell (the first) The L+1 layer input features of (multiple wireless cells) Let L be the output feature of the target wireless cell at layer L. The number of the associated wireless cell of the target wireless cell. The total number of associated wireless cells of the target wireless cell. Calculate the first weight matrix in the layer for the (L+1)th graph. The L+1 layer associated node features of the target wireless cell, To compute the second weight matrix in the L+1th graph layer, Calculate the bias vector for the (L+1)th graph. For ReLU functions, This is the sigmoid function.

[0062] Based on the final graph computation layer, the final layer output features of each wireless cell are obtained. The graph neural network identifies wireless cells with poor quality based on wireless operation and maintenance work orders. The final graph computation layer finds the final layer output features of these poor-quality wireless cells, which are then used as the comprehensive operational features.

[0063] In the multiple sequentially connected graph computation layers of this application, the output feature of the Lth layer is the input feature of the L+1th layer, realizing multi-layer extraction of the operational features of each wireless cell and the features of the network topology graph. This ensures that the comprehensive operational features include the operational features of wireless cells with poor quality and comprehensive correlation features, which is beneficial to improving the accuracy of subsequent determination of the root causes of poor quality based on the comprehensive operational features.

[0064] Based on the above embodiments, the fully connected network is used to perform the following operations: Multiple feature extractions, feature fusions, and feature transformations are performed on the comprehensive operational characteristics. Based on the result of the last transformation, the probabilities of multiple specified root causes are obtained. The root cause with the highest probability is selected as the poor quality root cause.

[0065] like Figure 3 As shown, a fully connected network comprises several fully connected layers. This application uses a fully connected network comprising two fully connected layers as an example for illustration.

[0066] Fully connected networks can perform deep feature extraction and integration of the comprehensive operational features of the input, conduct comprehensive analysis on the integrated features, explore the intrinsic relationships and high-level semantic information between them, and map the comprehensive operational features to a feature space that is more conducive to classification.

[0067] The first fully connected layer performs feature extraction, feature fusion, and feature transformation on the comprehensive operational features to obtain the target feature vector. The second fully connected layer performs feature extraction, feature fusion, and feature transformation on the target feature vector to obtain the final feature vector. The dimension of the final feature vector is equal to the number of specified root causes. After normalizing the final feature vector through a normalized exponential function (Softmax) layer, the final N-dimensional vector is obtained, where each dimension represents the probability of a specified root cause. The probabilities of multiple specified root causes satisfy the following formula.

[0068] ; in, For the specified number of root causes, The number of the specified root cause. For the first The probability of a given root cause.

[0069] The root causes specified in this application include seven categories: capacity-related, interference-related, coverage structure-related, parameter anomaly-related, fault-related, temporary disturbance, and others.

[0070] The most probable root cause is selected as the quality defect root cause. Each quality defect root cause has a pre-defined corresponding handling plan. When a quality defect occurs, after determining the root cause, the system outputs the root cause and the corresponding handling plan to the maintenance personnel for processing, completing the closed loop of the wireless maintenance work order.

[0071] This application realizes the mapping of comprehensive operational characteristics to the probabilities of multiple specified root causes through a fully connected network, realizes intelligent analysis and processing of comprehensive operational characteristics, and improves the efficiency of locating the root causes of poor quality in wireless cells.

[0072] Based on the above embodiments, the poor quality root cause localization model is trained in the following manner: Based on the current time period's operational indicator data subset and the next time period's operational indicator data subset of the sample wireless cell, obtain the sample operational indicator data subset; The original time series network is trained based on a subset of sample operational index data until the first loss function of the original time series network satisfies a first set value, thus obtaining the time series network. The first loss function is constructed based on the predicted operational index data subset for the next time period output by the original time series network and the mean square error of the operational index data subset for the next time period. A pre-defined model is constructed based on time series networks and an initial root cause localization module; Based on the labels of the root causes of poor sample quality in sample wireless cells, the sample operation index datasets of multiple sample wireless cells, the identification information of sample wireless cells with poor quality, and the sample network topology are labeled to obtain labeled training samples. The preset model is trained on the training samples with labels until the second loss function of the preset model meets the second set value, thus obtaining the poor quality root cause localization model. The second loss function is determined based on the cross-entropy loss value of the predicted poor quality root cause output by the preset model and the label of the poor quality root cause of the sample.

[0073] In the training process of the entire poor quality root cause localization model, the original time-series network is first trained unsupervised using a subset of unsupervised sample performance index data from a single wireless cell. Then, in the supervised training phase, the time-series network, the initial graph neural network, and the initial fully connected network are integrated into a pre-defined model for end-to-end training. That is, the parameters of the time-series network can be further adjusted while training the initial graph neural network and the initial fully connected network.

[0074] A subset of operational indicator data from sample wireless cells is collected at fixed time intervals, for example, sampling at 15-minute granularities. A sample operational indicator data subset is obtained based on the operational indicator data subsets for the current time period and the next time period. During the training of the time series network, the original time series network predicts the predicted operational indicator data subset for the next time period based on the operational indicator data subset for the current time period. During training, a first loss function is constructed based on the mean squared error of the predicted operational indicator data subset and the operational indicator data subset for the next time period. During training, the parameters of the original time series network are updated according to the first loss function until the output value of the first loss function meets a first preset value, thus obtaining the time series network. For example, the expression for the first loss function is as follows.

[0075] ; in, For the first loss function, The number of sample operation indicator data in the subset of sample operation indicator data. The numbering of the sample operation index data, The time point number, For the position located at the The first time point The actual values ​​of the sample's operational indicator data (obtained from a subset of operational indicator data for the next actual time period). For the position located at the The first time point The predicted values ​​of the sample operation index data (obtained from a subset of the predicted operation index data for the next time period).

[0076] Since original time series networks typically have a large number of model parameters, and supervised training samples are difficult to collect and label, this application employs an unsupervised training method to train time series networks. This unsupervised training method does not require manual labeling and can fully utilize a large subset of sample index data to train the parameters of the time series network, thereby improving the training efficiency and accuracy of the time series network.

[0077] In the supervised training process, the initial root cause localization module includes an initial graph neural network and an initial fully connected network. A pre-defined model is constructed based on the time-series network, the initial graph neural network, and the initial fully connected network. Based on the labels of the root causes of poor quality in sample wireless cells, the sample operational index datasets and sample network topologies of multiple sample wireless cells are labeled to obtain labeled training samples. The pre-defined model is trained using these labeled training samples until the second loss function of the pre-defined model satisfies a second preset value, thus obtaining the poor quality root cause localization model.

[0078] The second loss function is determined based on the cross-entropy loss value of the predicted poor quality root causes and the labels of the sample poor quality root causes, as output by the pre-defined model. The formula for calculating the second loss function is as follows.

[0079] (4); in, For the second loss function, For the specified total number of root causes, Labels for the root causes of poor sample quality (belonging to the first category) The label value of a specified root cause, which is 1 or 0. To predict the root causes of poor quality (predicted to belong to the first category) (the probability of a given root cause). The number of the specified root cause.

[0080] Optionally, the root causes of poor sample quality in a sample radio cell are marked as 1, and non-poor sample root causes are marked as 0, thus obtaining the label. For example, if the root cause of poor sample quality in a sample radio cell is of the capacity category, then the capacity category is marked as 1, and the interference category, coverage structure category, parameter anomaly category, fault category, temporary disturbance category, and others are all marked as 0, thus obtaining the label of the root cause of poor sample quality.

[0081] The predicted root causes of poor quality include the probability of each root cause. For example, the predicted root causes of poor quality include a probability of 0.9 for capacity, 0.02 for disturbance, 0.02 for coverage structure, 0.02 for parameter anomalies, 0.02 for faults, 0.01 for temporary disturbances, and 0.01 for others.

[0082] This application acquires a time-series network through unsupervised training, fully utilizing the operational indicator data of sample wireless cells, thus solving the problem of limited supervised data and the inability to fully utilize a large amount of operational indicator data. This application integrates the time-series network, the initial graph neural network, and the initial fully connected network into a pre-defined model, simultaneously learning the operational indicator characteristics and neighbor cell relationship characteristics of the wireless cell itself. By combining the temporal and spatial information of the wireless cell, end-to-end training and testing can be achieved, effectively improving the accuracy of root cause localization of quality defects. Through end-to-end training, the parameters in each network (including the time-series network, the initial graph neural network, and the initial fully connected network) can be automatically adjusted based on the training data, resulting in greater adaptability and eliminating the need for manual rule setting.

[0083] The wireless access side plays a crucial role in the entire communication network, providing users with high-quality network connectivity services. Therefore, the rapid handling of poor wireless quality issues is essential. However, when poor wireless quality issues occur in a wireless cell, it is often necessary to analyze the causes of the quality problems by combining a large amount of operational indicator data, making root cause identification difficult. This application can solve the pain points in the wireless operation and maintenance process, helping operation and maintenance personnel to quickly analyze the causes of quality problems from a large number of wireless operation and maintenance work orders and take corresponding measures, thereby improving the efficiency of wireless operation and maintenance work order processing and enhancing customer perception and satisfaction. This application anticipates that after implementing automated root cause identification of quality problems, at least 10% of wireless operation and maintenance work orders can achieve automated closed-loop management. This application will also become a standardized intelligent tool for handling poor wireless quality problems.

[0084] The following describes the device for locating the root cause of poor quality in a wireless cell provided in this application. The device for locating the root cause of poor quality in a wireless cell described below can be referred to in correspondence with the method for locating the root cause of poor quality in a wireless cell described above.

[0085] like Figure 4 As shown, a device for locating the root cause of poor wireless cell quality includes: The poor quality root cause localization module 401 is used to input the operational indicator datasets of multiple wireless cells, the identification information of wireless cells with poor quality, and the network topology map into the poor quality root cause localization model, and obtain the poor quality root cause of the wireless cells with poor quality output by the poor quality root cause localization model. The network topology map includes the connection relationship between multiple wireless cells and wireless cells with poor quality. The poor quality root cause localization model includes a time series network and a root cause localization module. The time series network is used to obtain the operational characteristics of each wireless cell based on the operational indicator dataset, and the root cause localization module is used to determine the poor quality root cause based on the operational characteristics, identification information, and network topology map of each wireless cell.

[0086] The wireless cell quality poor root cause localization device provided in this application embodiment inputs a dataset of operational indicators of multiple wireless cells, identification information of wireless cells with poor quality, and a network topology map into a quality poor root cause localization model. The model then obtains the quality poor root causes of the wireless cells with poor quality, as output by the model. The network topology map includes the connection relationships between multiple wireless cells and the wireless cells with poor quality. The quality poor root cause localization model includes a time-series network and a root cause localization module. The time-series network is used to obtain the operational characteristics of each wireless cell based on the operational indicator dataset. The root cause localization module is used to determine the quality poor root causes based on the operational characteristics, identification information, and network topology map of each wireless cell. This application achieves the learning and extraction of the temporal characteristics of the operational indicators of wireless cells through a time-series network. Simultaneously, it obtains the spatial characteristics between wireless cells through the network topology map. By combining the temporal and spatial characteristics of the operational indicators, this application achieves accurate localization of the quality poor root causes, improving the accuracy of determining the quality poor root causes of wireless cells.

[0087] All relevant content of each step involved in the above method embodiments can be referenced from the functional description of the corresponding functional module, and will not be repeated here.

[0088] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communication bus 540. The processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a method for locating the root cause of poor quality in wireless cells. This method includes: inputting a dataset of operational indicators from multiple wireless cells, identification information of wireless cells with poor quality, and a network topology map into a model for locating the root cause of poor quality; obtaining the root cause of poor quality in the wireless cells output by the model; the network topology map includes the connection relationships between multiple wireless cells and the wireless cells with poor quality; wherein the model for locating the root cause of poor quality includes a time-series network and a root cause localization module. The time-series network is used to obtain the operational characteristics of each wireless cell based on the operational indicator dataset, and the root cause localization module is used to determine the root cause of poor quality based on the operational characteristics, identification information, and network topology map of each wireless cell.

[0089] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0090] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the wireless cell quality poor root cause localization method provided by the above methods. The method includes: inputting a dataset of operating indicators of multiple wireless cells, identification information of wireless cells with poor quality, and a network topology map into a quality poor root cause localization model, and obtaining the quality poor root cause of the wireless cells with poor quality output by the quality poor root cause localization model. The network topology map includes the connection relationship between multiple wireless cells and wireless cells with poor quality. The quality poor root cause localization model includes a time series network and a root cause localization module. The time series network is used to obtain the operating characteristics of each wireless cell based on the operating indicator dataset, and the root cause localization module is used to determine the quality poor root cause based on the operating characteristics, identification information, and network topology map of each wireless cell.

[0091] In another aspect, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the method for locating the root cause of poor quality in wireless cells provided by the above methods. The method includes: inputting a dataset of operational indicators of multiple wireless cells, identification information of wireless cells with poor quality, and a network topology map into a model for locating the root cause of poor quality; obtaining the root cause of poor quality in the wireless cells with poor quality output by the model; the network topology map includes the connection relationship between multiple wireless cells and the wireless cells with poor quality; wherein, the model for locating the root cause of poor quality includes a time series network and a root cause location module. The time series network is used to obtain the operational characteristics of each wireless cell based on the dataset of operational indicators, and the root cause location module is used to determine the root cause of poor quality based on the operational characteristics, identification information, and network topology map of each wireless cell.

[0092] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0093] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for locating the root cause of poor wireless cell quality, characterized in that, include: The system inputs the operational metric datasets of multiple wireless cells, the identification information of wireless cells with poor quality, and the network topology map into the poor quality root cause localization model to obtain the poor quality root cause of the wireless cells with poor quality output by the poor quality root cause localization model. The network topology map includes the connection relationship between the multiple wireless cells and the wireless cells with poor quality. The poor quality root cause localization model includes a time series network and a root cause localization module. The time series network is used to obtain the operational characteristics of each wireless cell based on the operational indicator dataset. The root cause localization module is used to determine the poor quality root cause based on the operational characteristics of each wireless cell, the identification information, and the network topology map.

2. The method for locating the root cause of poor quality in a wireless cell according to claim 1, characterized in that, The root cause localization module includes a graph neural network and a fully connected network: The graph neural network is used to obtain the comprehensive operating characteristics of the wireless cells with poor quality based on the operating characteristics of each wireless cell, the identification information and the network topology. The comprehensive operating characteristics include the operating characteristics of the wireless cells with poor quality and the associated characteristics that affect the operating characteristics of the wireless cells with poor quality. The fully connected network is used to determine the root causes of the quality problems based on the comprehensive operational characteristics.

3. The method for locating the root cause of poor quality in a wireless cell according to claim 1, characterized in that, The operational metric dataset includes a subset of operational metric data for each of the wireless cells. This subset includes multiple historical operational metrics with timestamps. The time-series network is used for: Obtain the historical indicator time series of each wireless cell. The historical indicator time series is obtained by sorting the historical operating indicators in each subset of the operating indicator data based on the timestamp. Encode and decode the historical indicator time series of each wireless cell to predict the future indicator time series of each wireless cell in the future time period; Based on the time series of each of the aforementioned future indicators, the operational characteristics of each of the aforementioned wireless cells are obtained.

4. The method for locating the root cause of poor quality in a wireless cell according to claim 2, characterized in that, The graph neural network includes multiple sequentially connected graph computation layers, and the graph neural network is used for: For the first graph computation layer, the first layer output characteristics of the target wireless cell are determined based on the operational characteristics of the target wireless cell and the characteristics of the first layer associated nodes of the target wireless cell. Based on the first-layer output features of each target wireless cell, obtain the first-layer output features of each wireless cell; The target wireless cell is any of the wireless cells mentioned above. The characteristics of the first layer associated nodes are determined based on the operating characteristics of the associated wireless cells. The associated wireless cells are determined based on the network topology map and are wireless cells connected to the target wireless cell. For the (L+1)th graph computation layer, the (L+1)th layer input features of each wireless cell are obtained based on the output features of the Lth layer of each wireless cell. Based on the L+1 layer input features of the target wireless cell and the L+1 layer associated node features of the target wireless cell, determine the L+1 layer output features of the target wireless cell; Based on the L+1 layer output features of each target wireless cell, obtain the L+1 layer output features of each wireless cell; Until the last layer output features of each wireless cell are obtained, L≥1, the L+1 layer associated node features of the target wireless cell are determined based on the L+1 layer input features of the associated wireless cell; Based on the identification information, the last-layer output features of the wireless cells with poor quality are used as the comprehensive operating features.

5. The method for locating the root cause of poor quality in a wireless cell according to claim 2, characterized in that, The fully connected network is used for: The comprehensive operational features are subjected to multiple feature extractions, feature fusions, and feature transformations. Based on the result of the last transformation, the probabilities of multiple specified root causes are obtained. The specified root cause with the highest probability is taken as the poor quality root cause.

6. The method for locating the root cause of poor quality in a wireless cell according to claim 1, characterized in that, The poor quality root cause localization model was trained in the following way: Based on the current time period's operational indicator data subset and the next time period's operational indicator data subset of the sample wireless cell, obtain the sample operational indicator data subset; The original time series network is trained based on the sample operational indicator data subset until the first loss function of the original time series network satisfies the first set value, thereby obtaining the time series network. The first loss function is constructed based on the predicted operational indicator data subset for the next time period output by the original time series network and the mean square error of the operational indicator data subset for the next time period. Based on the time series network and the initial root cause localization module, a preset model is constructed; Based on the labels of the root causes of poor sample quality in sample wireless cells, the sample operation index datasets of multiple sample wireless cells, the identification information of sample wireless cells with poor quality, and the sample network topology are labeled to obtain labeled training samples. The preset model is trained on the labeled training samples until the second loss function of the preset model satisfies the second set value, thereby obtaining the poor quality root cause localization model. The second loss function is determined based on the cross-entropy loss value of the predicted poor quality root cause output by the preset model and the label of the poor quality root cause of the sample.

7. A device for locating the root cause of poor wireless cell quality, characterized in that, include: The poor quality root cause localization module is used to input the operational indicator datasets of multiple wireless cells, the identification information of wireless cells with poor quality, and the network topology map into the poor quality root cause localization model, and obtain the poor quality root cause of the wireless cells with poor quality output by the poor quality root cause localization model. The network topology map includes the connection relationship between the multiple wireless cells and the wireless cells with poor quality. The poor quality root cause localization model includes a time series network and a root cause localization module. The time series network is used to obtain the operational characteristics of each wireless cell based on the operational indicator dataset. The root cause localization module is used to determine the poor quality root cause based on the operational characteristics of each wireless cell, the identification information, and the network topology map.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for locating the root cause of poor quality in a wireless cell as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for locating the root cause of poor quality in a wireless cell as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for locating the root cause of poor quality in a wireless cell as described in any one of claims 1 to 6.

Images