Method, device, equipment, medium and program product for locating root cause of poor quality cell

Abstract

The present invention relates to the field of communication technology and provides a method, device, equipment, medium and program product for locating the root cause of poor-quality cells. The method comprises: obtaining full data such as perception data, coverage data, interference data, capacity data and quality data of each target cell and inputting the data into a pre-trained classification prediction model to identify the poor-quality cells; performing model interpretation on the poor-quality cells to determine the quality contribution of each feature data of the poor-quality cells; locating the poor-quality root cause of the poor-quality cells based on the quality contribution; the classification prediction model is a sample data set constructed based on the historical full data and quality labels of all cells in the entire network, and is obtained by iteratively training a preset basic classification prediction model. By constructing a classification prediction model to predict poor-quality cells, cells that may be of poor quality can be identified in advance, improving the timeliness of operation and maintenance, and accurately locating the poor-quality root cause based on the quality contribution, ensuring the effectiveness of operation and maintenance.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a method, device, equipment, medium and program product for locating the root cause of a poor-quality cell. Background Art

[0002] With the continued growth in the number of communications users, especially with the widespread adoption of internet applications, user demand for data services has increased significantly, placing higher demands on network speed, stability, and coverage. High-quality network coverage and performance are key factors in providing high-quality services to users. Emerging services such as the Internet of Things, autonomous driving, and telemedicine place even higher demands on network performance (such as low latency and high reliability). The presence of poor-quality cells not only impacts user services but can even hinder the implementation and development of emerging services. Therefore, it is necessary to optimize resource allocation within limited O&M resources to ensure the effective O&M of poor-quality cells.

[0003] Locating the root cause of poor-quality cells is an important part of network operation and maintenance, and has a significant impact on operation and maintenance efficiency, effectiveness, and even the health of the entire network environment. The core goal of operation and maintenance is to ensure the stable operation of the network, quickly respond to and resolve various faults and problems, and ensure business continuity and data security. In the operation and maintenance process, it is crucial to accurately and quickly find the root cause of the problem. Root cause location needs to reveal the deep-seated reasons behind the fault, rather than just staying at the superficial level. This helps to fundamentally solve the problem during the operation and maintenance stage, prevent recurrence of the fault, and improve the quality and efficiency of operation and maintenance. At the same time, directly maintaining the root cause of the fault can reduce operation and maintenance time, accelerate the fault repair process, and reduce business interruption time.

[0004] Currently, the root cause of poor-quality cells is usually located based on troubleshooting of cell faults. On the one hand, the root cause of poor-quality cells can only be located after they occur, resulting in a lag in operation and maintenance work and affecting the timeliness of operation and maintenance. On the other hand, the root cause of the poor-quality cell fault may be the result of the combined effect of multiple factors. It is difficult to locate the true root cause through troubleshooting, which easily affects the effectiveness of operation and maintenance work. Summary of the Invention

[0005] The present invention provides a method, device, equipment, medium and program product for locating the root cause of poor quality cells, which is used to solve the defect that the existing method of locating the root cause of poor quality cells through fault troubleshooting easily affects the timeliness and effectiveness of operation and maintenance.

[0006] The present invention provides a method for locating the root cause of poor cell quality, comprising:

[0007] Obtaining full cell data for each target cell; the full cell data includes perception data, coverage data, interference data, capacity data, and quality data;

[0008] Inputting the full amount of cell data into a pre-trained classification prediction model, and using the classification prediction model to identify poor quality cells among the target cells;

[0009] Performing a model interpretation on the poor-quality cell to determine a quality contribution of each feature data corresponding to the full cell data of the poor-quality cell;

[0010] Locating a root cause of the poor quality of the poor quality cell according to the poor quality contribution;

[0011] The classification prediction model is obtained by iteratively training a preset basic classification prediction model based on a sample data set constructed based on the historical full data and quality difference labels of all cells in the entire network.

[0012] According to the method for locating the root cause of poor quality cells provided by the present invention, the poor quality cells include a plurality of cells; and performing model interpretation on the poor quality cells to determine the quality contribution of each feature data corresponding to the full cell data of the poor quality cells includes:

[0013] Obtaining a scenario category corresponding to each of the poor-quality cells and a poor-quality prototype corresponding to the scenario category; the poor-quality prototype is obtained by extracting features from the entire historical data of the poor-quality cells under the scenario category;

[0014] Calculating the relative entropy between the poor-quality cell and the poor-quality prototype based on the full cell data of the poor-quality cell, and obtaining the prediction confidence of the poor-quality cell;

[0015] Calculating the product of the relative entropy and the prediction confidence to obtain the priority score of the poor quality cell;

[0016] Sorting the poor quality cells in descending order according to the priority;

[0017] Based on the sorting order of each of the poor quality cells, a model interpretation is performed on the first K poor quality cells to determine the quality contribution of each characteristic data corresponding to the full cell data of the target poor quality cell; the target poor quality cell is any one of the first K poor quality cells, and K is a positive integer greater than or equal to 1.

[0018] According to the method for locating the root cause of poor quality cells provided by the present invention, the method of inputting the full amount of cell data into a pre-trained classification prediction model and using the classification prediction model to identify the poor quality cells in the target cell includes:

[0019] Inputting the full cell data into a pre-trained classification prediction model, using the classification prediction model to identify the scene category of the target cell, and performing feature extraction on the full cell data to obtain feature data corresponding to the full cell data; the feature data includes perception features, coverage features, interference features, capacity features, and quality features;

[0020] Identify poor-quality cells among the target cells according to the scene category and the feature data.

[0021] According to the method for locating the root cause of poor quality cells provided by the present invention, after locating the root cause of the poor quality of the poor quality cell according to the poor quality contribution, the method further includes:

[0022] generating a cell maintenance suggestion based on the root cause of the poor quality;

[0023] The cell maintenance suggestion is sent to the operation and maintenance terminal corresponding to the operation and maintenance personnel for display.

[0024] According to the method for locating the root cause of poor cell quality provided by the present invention, after obtaining the full cell data of each target cell, the method further includes:

[0025] Filtering and cleaning missing data and abnormal data in the full data of the cell;

[0026] The filtered and cleaned cell data is subjected to structured processing; the structured processing includes data type conversion and data format conversion.

[0027] According to the root cause location method for poor quality cells provided by the present invention, the classification prediction model includes an input layer, a feedforward layer, a hidden layer, a classifier and an output layer; the feedforward layer includes a multi-layer deep neural network; the hidden layer includes multiple layers, and the multiple layers of the hidden layer contain different numbers of neurons.

[0028] The present invention also provides a device for locating the root cause of a poor-quality cell, comprising:

[0029] A data acquisition module is used to obtain full cell data of each target cell; the full cell data includes perception data, coverage data, interference data, capacity data and quality data;

[0030] A cell prediction module, configured to input the full amount of cell data into a pre-trained classification prediction model, and use the classification prediction model to identify cells with poor quality among the target cells;

[0031] A model interpretation module is used to perform model interpretation on the poor-quality cell to determine the quality contribution of each feature data corresponding to the full cell data of the poor-quality cell;

[0032] a root cause locating module, configured to locate a root cause of the poor quality of the poor quality cell according to the poor quality contribution;

[0033] The classification prediction model is obtained by iteratively training a preset basic classification prediction model based on a sample data set constructed based on the historical full data and quality difference labels of all cells in the entire network.

[0034] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the program, the steps of the method for locating the root cause of poor quality cells as described above are implemented.

[0035] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of any of the above-mentioned methods for locating the root cause of poor quality cells are implemented.

[0036] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of any of the above-mentioned methods for locating the root cause of poor quality cells.

[0037] The method, device, equipment, medium and program product for locating the root cause of poor quality cells provided by the present invention predict the full amount of data such as perception data, coverage data, interference data, capacity data and quality data of the cell by constructing a classification prediction model, identify the poor quality cells, and perform model interpretation on the poor quality cells, determine the quality difference contribution of each feature data, locate the poor quality root cause of the poor quality cell according to the quality difference contribution of each feature data, and locate the poor quality root cause based on the feature data closely related to the poor quality of the cell. By constructing a classification prediction model to identify poor quality cells, not only can the poor quality cells be predicted and cells that may be of poor quality be identified in advance, thereby improving the timeliness of operation and maintenance, but the poor quality root cause of the poor quality cell can also be accurately located based on the quality difference contribution using feature data closely related to the poor quality, thereby ensuring the effectiveness of operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 1 is a flow chart of the method for locating the root cause of poor quality cells provided by the present invention;

[0040] Figure 2This is a schematic diagram of the root cause location process for poor quality cells provided by the present invention;

[0041] Figure 3 2 is a schematic structural diagram of the root cause location device for poor quality cells provided by the present invention;

[0042] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0043] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0044] In the description of the embodiments of the present application, it should be noted that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the embodiments of the present application. In addition, the terms "first", "second", and "third" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0045] In the description of the embodiments of this application, it should be noted that, unless otherwise specified or limited, the terms "connected" and "connection" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium. Those skilled in the art will understand the specific meanings of the above terms in the embodiments of this application based on the specific circumstances.

[0046] In the embodiments of the present application, unless otherwise expressly specified or limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. Furthermore, a first feature being "above," "above," and "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below," "below," and "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0047] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiments of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0048] Reference Figure 1 The embodiment of the present invention provides a method for locating the root cause of poor cell quality. Figure 1 A flow chart of the method for locating the root cause of poor quality cells provided by an embodiment of the present invention, based on Figure 1 The method for locating the root cause of poor cell quality provided by an embodiment of the present invention includes:

[0049] Step 100: Acquire full cell data of each target cell; the full cell data includes perception data, coverage data, interference data, capacity data, and quality data;

[0050] Step 200: inputting the full amount of cell data into a pre-trained classification prediction model, and using the classification prediction model to identify cells with poor quality among the target cells;

[0051] Step 300: Perform model interpretation on the poor-quality cell to determine the quality contribution of each feature data corresponding to the full cell data of the poor-quality cell;

[0052] Step 400: locating the root cause of the poor quality of the poor quality cell according to the poor quality contribution;

[0053] The classification prediction model is obtained by iteratively training a preset basic classification prediction model based on a sample data set constructed based on the historical full data and quality difference labels of all cells in the entire network.

[0054] First, obtain full cell data for each target cell. This data includes perception data, coverage data, interference data, capacity data, and quality data. The target cell can be one or more cells requiring quality assessment, or one or more cells in a specific service scenario, without limitation.

[0055] Optionally, perception data include the proportion of highly perceived dissatisfied users, the proportion of poor web browsing quality, the proportion of poor instant messaging quality, the proportion of video freezes, the proportion of game freezes, the end-to-end connection rate and the comprehensive call drop rate; coverage data include downlink coverage, uplink coverage, overlapping coverage, over-coverage and over-coverage; interference data include average interference and the proportion of interference RB (Resource Block); capacity data include whether there is high load to be expanded, whether there is high load warning, uplink PRB (Physical Resource Block) utilization (from busy time) and downlink PRB utilization (from busy time); quality data include uplink poor quality, uplink CCE (Control Channel Element) allocation success rate, downlink CCE allocation success rate and uplink residual block error rate.

[0056] The obtained full cell data of each target cell is input into a pre-trained classification prediction model, and the classification prediction model is used to identify the poor quality cells in the target cells. The classification prediction model is constructed based on samples and data sets of the historical full cell data and poor quality labels of cells in the entire network, and is obtained by iteratively training the preset basic classification prediction model.

[0057] Optionally, in one embodiment, all cells in the entire network are labeled as having poor quality, and then a sample data set is constructed based on the historical full data of all cells in the entire network. A preset basic classification prediction model is iteratively trained, and the classification prediction model extracts features from the historical full data labeled as poor quality cells to form a poor quality prototype of the poor quality cells. When identifying the poor quality of each target cell, the full cell data of each target cell is input into a pre-trained classification prediction model. The classification prediction model can calculate the similarity between each target cell and the poor quality prototype based on the full cell data, thereby identifying a cell with a high similarity to the poor quality cell, that is, a poor quality cell, or a cell that is about to enter poor quality.

[0058] Furthermore, the identified poor quality cells may include one or more, and a model interpretation is performed on the identified poor quality cells to determine the quality contribution of each characteristic data corresponding to the full cell data of the poor quality cell, and locate the root cause of the poor quality of the poor quality cell based on the quality contribution.

[0059] Optionally, based on the quality contribution of each feature data corresponding to the full cell data, feature data with a quality contribution greater than a preset threshold is selected for locating the root cause of the quality difference; alternatively, the feature data is sorted in descending order based on the quality contribution, and a preset number of feature data are selected for locating the root cause of the quality difference. Locating the root cause of quality difference based on the contribution of each feature data is to locate the root cause of the cell's quality difference based on feature data that is closely related to the cell's quality difference, which is conducive to accurately locating the cell's quality difference.

[0060] In this embodiment, by constructing a classification prediction model, a full range of data such as perception data, coverage data, interference data, capacity data, and quality data of a cell is predicted to identify poor quality cells, and a model interpretation is performed on the poor quality cells to determine the quality contribution of each feature data. The root cause of the poor quality of the poor quality cell is located based on the quality contribution of each feature data, and the root cause of the poor quality can be located based on feature data closely related to the poor quality of the cell. By constructing a classification prediction model to identify poor quality cells, not only can poor quality cells be predicted and cells that may be of poor quality be identified in advance, thereby improving the timeliness of operation and maintenance, but the root cause of the poor quality of the poor quality cell can also be accurately located based on the quality contribution using feature data closely related to the poor quality, thereby ensuring the effectiveness of operation and maintenance.

[0061] Optionally, the identified poor-quality cells may include multiple cells. In step 300, a model interpretation is performed on the poor-quality cells to determine the quality contribution of each feature data corresponding to the full cell data of the poor-quality cells, specifically including:

[0062] Step 310: Obtain the scenario category corresponding to each poor-quality cell and the poor-quality prototype corresponding to the scenario category; the poor-quality prototype is obtained by extracting features from the entire historical data of the poor-quality cells under the scenario category;

[0063] Step 320: Calculate the relative entropy between the poor-quality cell and the poor-quality prototype based on the full cell data of the poor-quality cell, and obtain the prediction confidence of the poor-quality cell;

[0064] Step 330: Calculate the product of the relative entropy and the prediction confidence to obtain the priority score of the poor quality cell;

[0065] Step 340: sorting the poor quality cells in descending order according to the priority;

[0066] Step 350: Based on the sorting order of each of the poor quality cells, a model interpretation is performed on the first K poor quality cells to determine the quality contribution of each characteristic data corresponding to the full cell data of the target poor quality cell; the target poor quality cell is any one of the first K poor quality cells, and K is a positive integer greater than or equal to 1.

[0067] Obtain the scenario category corresponding to each poor-quality cell and the corresponding poor-quality prototype for that scenario category. The scenario category is determined by the poor-quality cell's service scenario. Different service scenarios have different requirements for the cell's network performance and focus on different network metrics, resulting in different criteria for determining poor cell quality. The poor-quality prototype is derived by extracting features from the entire historical data set of poor-quality cells within that scenario category.

[0068] Furthermore, based on the full cell data of the poor quality cell, the relative entropy between the poor quality cell and the poor quality prototype is calculated, and the prediction confidence of the poor quality cell is obtained. The prediction confidence is the probability that the classification prediction model predicts that the target cell is a poor quality cell. The product of the relative entropy of the poor quality cell and the prediction confidence is calculated as the priority score of the poor quality cell. According to the priority score, multiple poor quality cells are sorted in descending order. Based on the sorting order of the poor quality cells, the model interpretation is performed on the top K poor quality cells to determine the quality contribution of each feature data corresponding to the full cell data of the target poor quality cell, where the target poor quality cell is any one of the top K poor quality cells, and K is a positive integer greater than or equal to 1.

[0069] Optionally, after determining the quality contribution of the top K poor-quality cells, the root cause of the poor quality of the top K poor-quality cells is located based on the quality contribution. Further, after determining the root cause of the poor quality of the poor-quality cells, operation and maintenance can be performed on the poor-quality cells. The product of relative entropy and prediction confidence is used as the priority score of the poor-quality cells, which can be used as a reference for the operation and maintenance priority of the poor-quality cells, allowing operation and maintenance personnel to prioritize the cells that the model believes are most likely to be of poor quality and have more severe poor quality.

[0070] Optionally, in step 200, all cell data of the target cell is input into a pre-trained classification prediction model, and the classification prediction model is used to identify cells with poor quality in the target cell, specifically including:

[0071] Step 210: Input the full cell data into a pre-trained classification prediction model, use the classification prediction model to identify the scene category of the target cell, and perform feature extraction on the full cell data to obtain feature data corresponding to the full cell data; the feature data includes perception features, coverage features, interference features, capacity features, and quality features;

[0072] Step 220: Identify cells with poor quality among the target cells according to the scene category and the feature data.

[0073] The acquired full cell data is input into a trained classification prediction model, and the classification prediction model is used to identify the scene category of the target cell. Feature extraction is then performed on the full cell data to obtain feature data corresponding to the full cell data. The feature data includes perception features, coverage features, interference features, capacity features, and quality features.

[0074] Optionally, the target cells include multiple cells, and the full cell data input into the classification prediction model includes the full cell data of each target cell. The scene categories to which different target cells belong may be the same or different.

[0075] Based on the scene category to which each target cell belongs and the feature data extracted from the full data of the cell, the poor quality cells among the target cells are identified. Optionally, for any target cell, its feature data is analyzed for similarity with the poor quality prototype under the scene category to which it belongs. If the similarity is greater than a preset value, the target cell under the scene category is considered to be a poor quality cell, or there is a certain probability that it will become a poor quality cell.

[0076] Optionally, in step 400, after locating the root cause of the poor quality of the poor quality cell according to the poor quality contribution, the following steps may also be performed:

[0077] Step 401, generating a cell maintenance suggestion based on the root cause of the poor quality;

[0078] Step 402: Send the cell maintenance suggestion to the operation and maintenance personnel's corresponding operation and maintenance terminal for display.

[0079] Based on the root cause of the poor quality of the poor quality cell, a cell maintenance suggestion is generated and sent to the corresponding operation and maintenance terminal of the operation and maintenance personnel for display.

[0080] Optionally, in step 100, after obtaining the full cell data of the target cell, the following steps may also be included:

[0081] Step 101: filtering and cleaning missing data and abnormal data in the full data of the cell;

[0082] Step 102 : Structural processing is performed on the filtered and cleaned full cell data; the structural processing includes data type conversion and data format conversion.

[0083] Before inputting the acquired full cell data into a pre-trained classification prediction model to identify poor-quality cells, the data must be pre-processed. This primarily involves filtering and cleaning missing and abnormal data, and then performing structured processing on the filtered and cleaned data. Furthermore, structured processing includes data type conversion and data format conversion.

[0084] Optionally, in one embodiment, the classification prediction model includes an input layer, a feedforward layer, a hidden layer, a classifier, and an output layer; the feedforward layer includes a multi-layer deep neural network DNN, the hidden layer includes multiple layers, and the multiple layers of the hidden layer contain different numbers of neurons. Exemplarily, the feedforward layer of the classification prediction model includes a three-layer feedforward neural network DNN, the input is 41-dimensional full-scale cell data, and the output is 128-dimensional features and classification results. The dimension of the input data is 41, so the input layer contains 41 neurons, and the hidden layer includes two, the first hidden layer contains 64 neurons, and the second hidden layer contains 128 neurons, so the model output is a 128-dimensional feature, and the classifier chooses to use the nearest class mean classifier (NearestClassMean, NCM).

[0085] In one embodiment, referring to Figure 2 The root cause positioning process of poor quality cells shown in the figure, the root cause positioning method of poor quality cells provided by the embodiment of the present invention mainly includes two stages: model training and model application. Among them, the model training stage includes the construction of sample data sets and the construction of cell prototypes, and the model application stage mainly includes the acquisition and preprocessing of the full data of the target cell to be predicted, model identification and model interpretation.

[0086] Specifically, based on Figure 2 During the model training phase, the model first obtains historical full data for all cells in the network, along with quality labels for all cells in the network. These labels are used to identify cells with poor quality within the network. The historical full data for all cells in the network includes historical perception data, coverage data, interference data, capacity data, and quality data. The obtained historical full data is preprocessed, filtering and cleaning missing data and outliers. The filtered and cleaned historical full data is then structured. For example, the data formats for "Cell Name" and "Scenario Subcategory" are set to string types, while the data formats for fields such as "Downlink Coverage," "Percentage of Poor Instant Messaging Quality," and "Uplink PRB Utilization (from Busy Hour)" are set to floating-point numbers. The data formats for fields such as "Is High Load Expansion Required?" and "Is High Load Warning Alert Required" are set to category types. Furthermore, during the preprocessing process, standardization can be performed using methods such as the z-score method. The standardized data is used for regression analysis, which can reduce the impact of the indicator's dimensionality on quality analysis and regression prediction.

[0087] The acquired historical full data is preprocessed to construct a sample data set. During the model training phase, the preprocessed historical full data is divided into an 8:2 ratio, with 80% of the sample data used for model training and 20% of the sample data used for model testing.

[0088] Optionally, the classification prediction model is shown in the following formulas 1 and 2:

[0089]

[0090] y * =argmin c＝1,2 ‖f(x,θ)-μ c ||; (2)

[0091] Among them, n c is the number of samples of scene category c, 1{y i =c} is a category indicator. When the value in the brackets is true, the indicator data value is 1, otherwise it is 0. i is the true quality difference label of the i-th sample, y * is the predicted quality label of the i-th sample, f(·) is the feature extractor, that is, the feedforward layer deep neural network of the classification prediction model, x i represents the i-th sample data, μ c is the poor quality prototype of scene category c, argmin c＝1,2 ‖·‖ returns f(x,θ)-μ c The scene category c with the smallest value, θ is the model parameter of the feedforward layer deep neural network.

[0092] After completing the model training, the cell prototypes and classification prediction models under each scenario category are solidified. In the model application stage, the full data of the target cell to be predicted is first obtained, and the obtained full data is preprocessed. The preprocessing method for the full data of the target cell can be the same as the preprocessing method for the historical full data in the sample data set construction stage, which will not be repeated here.

[0093] Furthermore, the pre-processed full data is input into a pre-trained classification prediction model, which performs a similarity analysis between the full data and the cell prototype under the scene category corresponding to the target cell, thereby predicting the poor quality cell based on the input full data and identifying the poor quality cell. Among them, the cell prototypes under different scene categories calculated according to the scene category of the cell are shown in the following formula 3:

[0094]

[0095] In formula 3, λ j is the prototype of the cell of the jth scene category, N j is the number of samples under the j-th scene category, represents the i-th sample under the j-th scene category.

[0096] For poor-quality cells identified by the classification prediction model, the product of the prediction confidence and relative entropy is used as the operation and maintenance priority score of the poor-quality cell. Relative entropy, also known as KL divergence, is used to measure the similarity of the probability distributions of two random variables, as shown in Formula 4:

[0097]

[0098] Among them, D KL (f(x i ,θ)||λ) represents the KL divergence, n represents the number of possible values ​​of v, λ(v i ) is the prototype of the scene category in v i The value of the dimension, f(x i ,θ)(v i ) is the sample x i Embedded in v i The value on .

[0099] As described above, in this embodiment, the classification prediction model uses NCM as a classifier, so the feedforward layer neural network only outputs the high-dimensional embedding of each sample, but does not contain a confidence vector. Therefore, the distance between the high-dimensional embedding and the prototype of each scene category is used to calculate the confidence, as shown in Formula 5:

[0100]

[0101] In formula 5, ρ i is the sample x i The prediction confidence of μ1 is the prototype of the cell in scenario category 1, and μ2 is the prototype of the cell in scenario category 2. If scenario category 2 is a poor quality cell, the method shown in Formula 6 can be used to calculate the sample x i The weighted product of the prediction confidence and KL divergence is used as the priority score of the cell score(x i ):

[0102]

[0103] Furthermore, during the model interpretation phase, the poor-quality cells are sorted in descending order based on their priority scores. Based on the sorted order, the top K poor-quality cells are selected for model interpretation. This process determines the contribution of each feature data point to the prediction result, thereby determining the contribution of each type of data point in the full cell data set to the poor-quality prediction result for the poor-quality cell. Based on the quality contribution, feature data closely associated with the poor-quality cell is identified. This feature data is then used to locate the root cause of the poor-quality cell, accurately pinpointing the root cause.

[0104] Optionally, the model interpretation of poor quality cells can be implemented based on the SHAP interpreter and other methods. The priority score of the cell predicted to be poor quality is calculated by multiplying the prediction confidence and the relative entropy, and the cells are sorted in descending order according to the maintenance priority score. The SHAP interpreter is used to perform model interpretation on the poor quality cells with a priority score of TOPK, thereby determining the quality contribution of the characteristic data of each poor quality cell, and selecting the characteristic data closely related to the quality difference for locating the root cause of the quality difference.

[0105] In this embodiment, poor quality cells are predicted based on a deep neural network, and the operation and maintenance priority of the poor quality cells is evaluated by the relative entropy of the poor quality cells and the poor quality cell prototypes under the corresponding scenario category through the classification prediction confidence of the poor quality cells. The top K poor quality cells are modeled according to the operation and maintenance priority, and characteristic data closely related to the root cause of the poor quality of the poor quality cells are determined for root cause location. In this way, when operation and maintenance resources are limited, resource allocation can be optimized, and operation and maintenance can be given priority to cells that are more likely to become poor quality cells and have a higher degree of poor quality, thereby improving the overall operation and maintenance efficiency and level.

[0106] Furthermore, the greater the classification prediction confidence, the more likely the cell is to have poor quality. If the characteristic data of the cell is more similar to the prototype of the poor quality cell, the more serious the poor quality of the cell. The weighted product of the classification prediction confidence and relative entropy of each poor quality cell is used as the maintenance priority score of the poor quality cell. The greater the possibility of poor quality and the more serious the poor quality of the cell, the higher the operation and maintenance priority, which can better guide the operation and maintenance work and improve the utilization rate and efficiency of operation and maintenance resources.

[0107] The following describes the device for locating the root cause of a poor quality cell provided by the present invention. The device for locating the root cause of a poor quality cell described below and the method for locating the root cause of a poor quality cell described above can be referenced to each other.

[0108] Reference Figure 3 The embodiment of the present invention provides a device for locating the root cause of a poor-quality cell, including:

[0109] The data collection module 10 is used to obtain the full cell data of each target cell; the full cell data includes perception data, coverage data, interference data, capacity data and quality data;

[0110] The cell prediction module 20 is configured to input the full amount of cell data into a pre-trained classification prediction model, and use the classification prediction model to identify cells with poor quality among the target cells;

[0111] A model interpretation module 30 is configured to perform a model interpretation on the poor-quality cell to determine a quality contribution of each feature data corresponding to the full cell data of the poor-quality cell;

[0112] a root cause locating module 40, configured to locate a root cause of the poor quality of the poor quality cell according to the poor quality contribution;

[0113] The classification prediction model is obtained by iteratively training a preset basic classification prediction model based on a sample data set constructed based on the historical full data and quality difference labels of all cells in the entire network.

[0114] In one embodiment, the poor quality cells include a plurality of cells; the model interpretation module 30 is further configured to:

[0115] Obtaining a scenario category corresponding to each of the poor-quality cells and a poor-quality prototype corresponding to the scenario category; the poor-quality prototype is obtained by extracting features from the entire historical data of the poor-quality cells under the scenario category;

[0116] Calculating the relative entropy between the poor-quality cell and the poor-quality prototype based on the full cell data of the poor-quality cell, and obtaining the prediction confidence of the poor-quality cell;

[0117] Calculating the product of the relative entropy and the prediction confidence to obtain the priority score of the poor quality cell;

[0118] Sorting the poor quality cells in descending order according to the priority;

[0119] Based on the sorting order of each of the poor quality cells, a model interpretation is performed on the first K poor quality cells to determine the quality contribution of each characteristic data corresponding to the full cell data of the target poor quality cell; the target poor quality cell is any one of the first K poor quality cells, and K is a positive integer greater than or equal to 1.

[0120] In one embodiment, the cell prediction module 20 is further configured to:

[0121] Inputting the full cell data into a pre-trained classification prediction model, using the classification prediction model to identify the scene category of the target cell, and performing feature extraction on the full cell data to obtain feature data corresponding to the full cell data; the feature data includes perception features, coverage features, interference features, capacity features, and quality features;

[0122] Identify poor-quality cells among the target cells according to the scene category and the feature data.

[0123] In one embodiment, the poor quality cell root cause location device further includes a post-processing module configured to:

[0124] generating a cell maintenance suggestion based on the root cause of the poor quality;

[0125] The cell maintenance suggestion is sent to the operation and maintenance terminal corresponding to the operation and maintenance personnel for display.

[0126] In one embodiment, the poor quality cell root cause location device further includes a pre-processing module, which is configured to:

[0127] Filtering and cleaning missing data and abnormal data in the full data of the cell;

[0128] The filtered and cleaned cell data is subjected to structured processing; the structured processing includes data type conversion and data format conversion.

[0129] In one embodiment, the classification prediction model includes an input layer, a feedforward layer, a hidden layer, a classifier and an output layer; the feedforward layer includes a multi-layer deep neural network; the hidden layer includes multiple layers, and the multiple layers of the hidden layer contain different numbers of neurons.

[0130] Figure 4 An example of a physical structure diagram of an electronic device is shown below. Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute the steps of the method for locating the root cause of poor quality cells, which includes:

[0131] Obtaining full cell data for each target cell; the full cell data includes perception data, coverage data, interference data, capacity data, and quality data;

[0132] Inputting the full amount of cell data into a pre-trained classification prediction model, and using the classification prediction model to identify poor quality cells among the target cells;

[0133] Performing a model interpretation on the poor-quality cell to determine a quality contribution of each feature data corresponding to the full cell data of the poor-quality cell;

[0134] Locating a root cause of the poor quality of the poor quality cell according to the poor quality contribution;

[0135] The classification prediction model is obtained by iteratively training a preset basic classification prediction model based on a sample data set constructed based on the historical full data and quality difference labels of all cells in the entire network.

[0136] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling 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 method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0137] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of the method for locating the root cause of poor quality cells provided by the above methods. The method includes:

[0138] Obtaining full cell data for each target cell; the full cell data includes perception data, coverage data, interference data, capacity data, and quality data;

[0139] Inputting the full amount of cell data into a pre-trained classification prediction model, and using the classification prediction model to identify poor quality cells among the target cells;

[0140] Performing a model interpretation on the poor-quality cell to determine a quality contribution of each feature data corresponding to the full cell data of the poor-quality cell;

[0141] Locating a root cause of the poor quality of the poor quality cell according to the poor quality contribution;

[0142] The classification prediction model is obtained by iteratively training a preset basic classification prediction model based on a sample data set constructed based on the historical full data and quality difference labels of all cells in the entire network.

[0143] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for locating the root cause of poor quality cells provided by the above methods are implemented, the method comprising:

[0144] Obtaining full cell data for each target cell; the full cell data includes perception data, coverage data, interference data, capacity data, and quality data;

[0145] Inputting the full amount of cell data into a pre-trained classification prediction model, and using the classification prediction model to identify poor quality cells among the target cells;

[0146] Performing a model interpretation on the poor-quality cell to determine a quality contribution of each feature data corresponding to the full cell data of the poor-quality cell;

[0147] Locating a root cause of the poor quality of the poor quality cell according to the poor quality contribution;

[0148] The classification prediction model is obtained by iteratively training a preset basic classification prediction model based on a sample data set constructed based on the historical full data and quality difference labels of all cells in the entire network.

[0149] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0150] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for locating the root cause of poor cell quality, characterized in that: include: Obtaining full cell data for each target cell; the full cell data includes perception data, coverage data, interference data, capacity data, and quality data; Inputting the full amount of cell data into a pre-trained classification prediction model, and using the classification prediction model to identify poor quality cells among the target cells; Performing a model interpretation on the poor-quality cell to determine a quality contribution of each feature data corresponding to the full cell data of the poor-quality cell; Locating a root cause of the poor quality of the poor quality cell according to the poor quality contribution; The classification prediction model is obtained by iteratively training a preset basic classification prediction model based on a sample data set constructed based on the historical full data and quality difference labels of all cells in the entire network; The poor-quality cells include a plurality of cells; and performing a model interpretation on the poor-quality cells to determine a quality contribution of each feature data corresponding to the full cell data of the poor-quality cells includes: Obtaining a scenario category corresponding to each of the poor-quality cells and a poor-quality prototype corresponding to the scenario category; the poor-quality prototype is obtained by extracting features from the entire historical data of the poor-quality cells under the scenario category; Calculating the relative entropy between the poor-quality cell and the poor-quality prototype based on the full cell data of the poor-quality cell, and obtaining the prediction confidence of the poor-quality cell; Calculating the product of the relative entropy and the prediction confidence to obtain the priority score of the poor quality cell; Sorting the poor quality cells in descending order according to the priority; Based on the sorting order of each of the poor quality cells, a model interpretation is performed on the first K poor quality cells to determine the quality contribution of each characteristic data corresponding to the full cell data of the target poor quality cell; the target poor quality cell is any one of the first K poor quality cells, and K is a positive integer greater than or equal to 1.

2. The method for locating the root cause of poor cell quality according to claim 1, characterized in that: Inputting the full amount of cell data into a pre-trained classification prediction model, and using the classification prediction model to identify poor quality cells among the target cells, includes: Inputting the full cell data into a pre-trained classification prediction model, using the classification prediction model to identify the scene category of the target cell, and performing feature extraction on the full cell data to obtain feature data corresponding to the full cell data; the feature data includes perception features, coverage features, interference features, capacity features, and quality features; Identify poor-quality cells among the target cells according to the scene category and the feature data.

3. The method for locating the root cause of poor cell quality according to claim 1, wherein: After locating the root cause of the poor quality of the poor quality cell according to the poor quality contribution, the method further includes: generating a cell maintenance suggestion based on the root cause of the poor quality; The cell maintenance suggestion is sent to the operation and maintenance terminal corresponding to the operation and maintenance personnel for display.

4. The method for locating the root cause of poor cell quality according to claim 1, wherein: After obtaining the full amount of cell data of each target cell, the method further includes: Filtering and cleaning missing data and abnormal data in the full data of the cell; The filtered and cleaned cell data is subjected to structured processing; the structured processing includes data type conversion and data format conversion.

5. The method for locating the root cause of poor cell quality according to claim 1, wherein: The classification prediction model includes an input layer, a feedforward layer, a hidden layer, a classifier and an output layer; the feedforward layer includes a multi-layer deep neural network; the hidden layer includes multiple layers, and the multiple layers of the hidden layer contain different numbers of neurons.

6. A device for locating the root cause of poor cell quality, characterized in that: include: A data acquisition module is used to obtain full cell data of each target cell; the full cell data includes perception data, coverage data, interference data, capacity data and quality data; A cell prediction module, configured to input the full amount of cell data into a pre-trained classification prediction model, and use the classification prediction model to identify cells with poor quality among the target cells; A model interpretation module is used to perform model interpretation on the poor-quality cell to determine the quality contribution of each feature data corresponding to the full cell data of the poor-quality cell; a root cause locating module, configured to locate a root cause of the poor quality of the poor quality cell according to the poor quality contribution; The classification prediction model is obtained by iteratively training a preset basic classification prediction model based on a sample data set constructed based on the historical full data and quality difference labels of all cells in the entire network; The poor-quality cells include a plurality of cells; and performing a model interpretation on the poor-quality cells to determine a quality contribution of each feature data corresponding to the full cell data of the poor-quality cells includes: Obtaining a scenario category corresponding to each of the poor-quality cells and a poor-quality prototype corresponding to the scenario category; the poor-quality prototype is obtained by extracting features from the entire historical data of the poor-quality cells under the scenario category; Calculating the relative entropy between the poor-quality cell and the poor-quality prototype based on the full cell data of the poor-quality cell, and obtaining the prediction confidence of the poor-quality cell; Calculating the product of the relative entropy and the prediction confidence to obtain the priority score of the poor quality cell; Sorting the poor quality cells in descending order according to the priority; Based on the sorting order of each of the poor quality cells, a model interpretation is performed on the first K poor quality cells to determine the quality contribution of each characteristic data corresponding to the full cell data of the target poor quality cell; the target poor quality cell is any one of the first K poor quality cells, and K is a positive integer greater than or equal to 1.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method for locating the root cause of poor quality cells as described in any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for locating the root cause of poor quality cells according to any one of claims 1 to 5 are implemented.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method for locating the root cause of poor quality cells according to any one of claims 1 to 5 are implemented.

Images