Network poor quality prediction method and device, nonvolatile storage medium and electronic equipment

Abstract

The invention discloses a poor network quality prediction method and device, a nonvolatile storage medium and electronic equipment. The method comprises the following steps: acquiring a first poor-quality feature from original signaling data, the first poor-quality feature being a feature representing that an abnormal communication scene exists in a communication network; according to the first poor-quality feature, determining a second poor-quality feature which is a statistic feature of the first poor-quality feature; the first poor-quality feature and the second poor-quality feature are screened, a poor-quality prediction model is trained according to a target poor-quality feature obtained after screening, the poor-quality prediction model is used for predicting a poor-quality object, and the poor-quality object is an object with the probability of encountering an abnormal communication scene meeting a preset requirement. According to the method and the device, the technical problems of inaccuracy and low efficiency of poor quality prediction caused by the fact that the poor quality reason cannot be accurately found by the existing network poor quality judgment technology are solved.

Description

Technical Field

[0001] The present application relates to the field of mobile communication network technology, and in particular to a network quality difference prediction method, device, non-volatile storage medium and electronic device. Background Art

[0002] In the existing technology, the reliability of the list of poor quality users obtained based on traditional rules (manual rules, simple judgments and rule algorithms) cannot be guaranteed, resulting in a large number of users on the network who perceive poor quality but have not complained being unable to be discovered and resolved in advance, causing complaints to be unable to be effectively controlled and reduced, while also objectively increasing the dissatisfaction of such users and the risk of leaving the network; operator network planning has long been affected by resources, and accurate planning and investment face challenges; customer care has a blanket model, and it is difficult to focus on effective services for potentially dissatisfied customers.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present application provide a network quality difference prediction method, device, non-volatile storage medium and electronic device to at least solve the technical problem of inaccurate quality difference prediction and low efficiency caused by the inability of existing network quality difference judgment technology to accurately find the cause of quality difference.

[0005] According to one aspect of an embodiment of the present application, a network quality difference prediction method is provided, comprising: obtaining a first quality difference feature from original signaling data, wherein the first quality difference feature is a feature that characterizes the existence of an abnormal communication scenario in the communication network; determining a second quality difference feature based on the first quality difference feature, wherein the second quality difference feature is a statistical feature of the first quality difference feature, and the statistical type of the statistical feature includes at least one of the following: mean, range, variance, median, quantile; screening the first quality difference feature and the second quality difference feature, and training a quality difference prediction model based on the target quality difference feature obtained after screening, wherein the quality difference prediction model is used to predict a quality difference object, and the quality difference object is an object whose probability of encountering an abnormal communication scenario meets preset requirements.

[0006] Optionally, screening the first quality difference feature and the second quality difference feature includes: screening the first quality difference feature according to the missing rate of the first quality difference feature, and screening the second quality difference feature according to the missing rate of the second quality difference feature, to obtain a first candidate feature set; determining a difference index determination method matching the feature type according to the feature type of each candidate feature in the first candidate feature set, and determining the difference index of the candidate feature according to the difference index determination method, wherein the difference index is used to reflect the degree of discrimination of the quality difference feature for the quality difference target; screening the candidate features in the first candidate feature set according to the difference index to obtain a second candidate feature set; determining a correlation index between each candidate feature in the second candidate feature set and other candidate features, and screening the candidate features in the second candidate feature set according to the correlation index to obtain a third candidate feature set, wherein the correlation index includes covariance; determining an importance index of each candidate feature in the third candidate feature set, and determining the target quality difference feature from the third candidate feature set according to the importance index, wherein the importance index includes at least one of the following: information gain value, information value.

[0007] Optionally, the missing rate includes a positive sample missing rate and a negative sample missing rate; the first quality difference feature and the second quality difference feature are screened according to the missing rate to obtain a first candidate feature set including: when the positive sample missing rate of the first quality difference feature or the second quality difference feature is higher than a preset missing rate threshold, the quality difference feature is removed; the positive sample missing rate of each of the first quality difference feature and the second quality difference feature is determined, and the first quality difference feature whose positive sample missing rate is higher than the preset missing rate threshold is removed, and the second quality difference feature whose positive sample missing rate is higher than the preset missing rate threshold is removed; when the positive sample missing rate of the first quality difference feature is lower than the preset missing rate threshold and the negative sample missing rate of the quality difference feature is higher than the preset missing rate threshold, the first quality difference feature is data filled; and when the positive sample missing rate of the second quality difference feature is lower than the preset missing rate threshold and the negative sample missing rate of the quality difference feature is higher than the preset missing rate threshold, the second quality difference feature is data filled.

[0008] Optionally, training the quality difference prediction model according to the target quality difference feature obtained after screening includes: determining a training sample set, wherein each sample in the training sample set includes the target quality difference feature and a feature value corresponding to the target quality difference feature; adjusting the weights of positive samples and negative samples in the sample set by an extreme gradient boosting model; training the quality difference prediction model by using the following loss function and the samples in the sample set after weight adjustment; FL(p t )=-α t (1-p t ) γ log(p t) where pt is the probability predicted by the quality difference prediction model, αt is the balance factor, and γ is the focus parameter. αt and γ are used to adjust the weights of positive and negative samples again.

[0009] Optionally, after training the quality difference prediction model, the method also includes: obtaining the original signaling details data to be predicted, and determining the third quality difference feature based on the original signaling details data to be predicted; inputting the third quality difference feature into the quality difference prediction model, and obtaining the quality difference prediction result output by the quality difference prediction model, wherein the quality difference prediction result includes the quality difference probability indicated by each third quality difference feature; determining the quality difference object based on the quality difference prediction result, wherein the quality difference object includes quality difference users and quality difference cells.

[0010] Optionally, after obtaining the quality difference prediction result of the quality difference prediction model, the method also includes: sorting each third quality difference feature in descending order according to the corresponding quality difference probability, and determining a preset number of third quality difference features in the front order as the root cause of the quality difference.

[0011] Optionally, the abnormal communication scenario includes at least one of the following: being unable to access the network, the access time consumed for accessing the network being greater than a preset time, and the signal strength within the coverage area of ​​the base station being lower than a preset strength.

[0012] According to another aspect of an embodiment of the present application, a network quality difference prediction device is also provided, including: an acquisition module, used to acquire a first quality difference feature from original signaling data, wherein the first quality difference feature is a feature that characterizes the existence of an abnormal communication scenario in the communication network; a determination module, used to determine a second quality difference feature based on the first quality difference feature, wherein the second quality difference feature is a statistical feature of the first quality difference feature, and the statistical type of the statistical feature includes at least one of the following: mean, range, variance, median, quantile; a screening module, used to screen the first quality difference feature and the second quality difference feature, and train a quality difference prediction model based on the target quality difference feature obtained after screening, wherein the quality difference prediction model is used to predict a quality difference object, and the quality difference object is an object whose probability of encountering an abnormal communication scenario meets preset requirements.

[0013] Optionally, according to another aspect of an embodiment of the present application, a non-volatile storage medium is further provided, in which a program is stored, wherein when the program is running, a device where the non-volatile storage medium is located is controlled to execute the network quality difference prediction method.

[0014] According to another aspect of an embodiment of the present application, there is also provided an electronic device, including: a memory and a processor, the processor being configured to run a program stored in the memory, wherein the network quality difference prediction method is executed when the program is run.

[0015] According to another aspect of an embodiment of the present application, a computer program product is also provided, including a computer program, and when the computer program is executed by a processor, a network quality difference prediction method is implemented.

[0016] In an embodiment of the present application, a first quality difference feature is obtained from original signaling data, wherein the first quality difference feature is a feature that characterizes the existence of abnormal communication scenarios in the communication network; based on the first quality difference feature, a second quality difference feature is determined, wherein the second quality difference feature is a statistical feature of the first quality difference feature, and the statistical type of the statistical feature includes at least one of the following: mean, range, variance, median, quantile; the first quality difference feature and the second quality difference feature are screened, and a quality difference prediction model is trained based on the target quality difference feature obtained after screening, wherein the quality difference prediction model is used to predict a quality difference object, and the quality difference object is an object whose probability of encountering an abnormal communication scenario meets preset requirements. By acquiring the quality difference feature based on the original signaling data to establish a quality difference prediction model, the purpose of accurately predicting the quality difference object is achieved, thereby achieving the technical effect of improving the efficiency and accuracy of quality difference feature processing and quality difference prediction model prediction, and further solving the technical problem of inaccurate quality difference prediction and low efficiency caused by the inability of existing network quality difference judgment technology to accurately find the cause of quality difference. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 It is a structural schematic diagram of a computer terminal (mobile device) provided according to an embodiment of the present application;

[0019] Figure 2 It is a flowchart of a network quality difference prediction method provided according to an embodiment of the present application;

[0020] Figure 3 It is a schematic diagram of a network quality difference prediction data source provided according to an embodiment of the present application;

[0021] Figure 4 is a schematic diagram of a first quality difference feature provided according to an embodiment of the present application;

[0022] Figure 5 It is a schematic diagram of a quality difference feature screening process provided according to an embodiment of the present application;

[0023] Figure 6 is a schematic diagram of a quality difference feature type provided according to an embodiment of the present application;

[0024] Figure 7is a schematic diagram of an information gain value provided according to an embodiment of the present application;

[0025] Figure 8 It is a schematic diagram of an XGBoost model principle provided according to an embodiment of the present application;

[0026] Fig. 9 is a schematic diagram of a poor quality object determination process provided according to an embodiment of the present application;

[0027] Fig.10 is a schematic diagram of a user feature contribution provided according to an embodiment of the present application;

[0028] Fig.11 It is a schematic diagram showing the difference between an existing quality difference prediction rule provided in an embodiment of the present application and the present application;

[0029] Fig.12 It is a schematic diagram of a network quality difference prediction method process provided according to an embodiment of the present application;

[0030] Fig.13 is a schematic diagram of a sample feature construction provided in an embodiment of the present application;

[0031] Fig.14 is a schematic diagram of a model training provided according to an embodiment of the present application;

[0032] Fig.15 It is a structural schematic diagram of a network quality difference prediction device provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] In order to better understand the embodiments of the present application, the technical terms involved in the embodiments of the present application are explained as follows:

[0036] CCM: Customer Complaint Mining, customer complaint mining.

[0037] In related technologies, operators generally have customer satisfaction assessments but lack effective means. This is often reflected in the fact that indicators are optimized but complaints do not decrease. It is impossible to accurately find and optimize the poor quality events related to complaints. The current identification of potential complaining users mainly faces the following pain points:

[0038] 1) There is a lack of mapping relationship between operational indicators and actual customer perception. The traditional KPI (Key Performance Indicator) / KQI (Key Quality Indicator) system cannot accurately determine which areas, businesses, and indicators to optimize in order to effectively reduce user complaints and improve customer satisfaction.

[0039] 2) Lack of evaluation system for complaining customers. The subjective feedback from some complaining users is quite different from their objective experience. How can we evaluate customers’ real network perception and identify customers with poor network experience to focus on optimization?

[0040] 3) Satisfaction sampling surveys are highly random and have a small sample size, making it difficult to fully guide quality optimization. Traditional model building based on satisfaction scoring data cannot fully utilize the advantages of machine learning due to insufficient sample data, resulting in low model accuracy and failure to meet customer needs.

[0041] In order to solve the above problems, relevant solutions are provided in the embodiments of the present application, which are described in detail below.

[0042] According to an embodiment of the present application, a method embodiment of a network quality difference prediction method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0043] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal for implementing a network quality difference prediction method. Figure 1 As shown, the computer terminal 10 may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0044] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0045] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the network quality difference prediction method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the above-mentioned network quality difference prediction method is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0046] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0047] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 .

[0048] In the above operating environment, the embodiment of the present application provides a network quality difference prediction method, such as Figure 2 As shown, the method comprises the following steps:

[0049] Step S202: Acquire a first quality difference feature from the original signaling data, wherein the first quality difference feature is a feature characterizing the existence of an abnormal communication scenario in the communication network.

[0050] Optionally, the abnormal communication scenario includes at least one of the following: being unable to access the network, the access time consumed for accessing the network being greater than a preset time, and the signal strength within the coverage area of ​​the base station being lower than a preset strength.

[0051] Optionally, the service data of the complaining user is taken as the research object, based on the signaling principle and the actual dialing test experience, combined with the user's original signaling data, the first quality difference feature is obtained, such as Figure 3 As shown, Figure 3The data source, i.e., the user's original signaling details, is shown. These details record the user's communication protocol process, complaint form, dial test form, and normal form in 4G and 5G networks. These data are used for feature construction, including determining the first quality difference feature, the second quality difference feature, and quality difference feature screening.

[0052] Optionally, the 3Sigma principle can be applied to clean the data and separate invalid samples; a set of data is calculated to obtain the standard deviation, and an interval is determined according to a certain probability. Any error exceeding this interval is considered an abnormal sample and should be eliminated. The detailed algorithm is as follows:

[0053] The 3σ criterion is also known as the Laida criterion. It assumes that a set of test data contains only random errors, calculates and processes them to obtain the standard deviation, and determines an interval according to a certain probability. It is believed that any error exceeding this interval is not a random error but a gross error, and the data containing this error should be eliminated. And 3σ is suitable when there are more sets of data.

[0054] The 3σ principle is:

[0055] The probability that the value is distributed in (μ-σ,μ+σ) is 0.6827;

[0056] The probability that the value is distributed in (μ-2σ,μ+2σ) is 0.9545;

[0057] The probability that the value is distributed in (μ-3σ,μ+3σ) is 0.9973;

[0058] It can be considered that the values ​​of Y are almost all concentrated in the interval (μ-3σ,μ+3σ), and the possibility of exceeding this range is less than 0.3%.

[0059] The 3σ criterion is based on the normal distribution of repeated measurements with equal precision, and the interference or noise caused by singular data is difficult to satisfy the normal distribution. If the absolute value of the residual error of a measurement value in a set of measurement data is νi>3σ, then the measurement value is a bad value and should be eliminated.

[0060] Usually the error equal to ±3σ is taken as the limit error. For the random error of normal distribution, the probability of falling outside ±3σ is only 0.27%. It is very unlikely to occur in a finite number of measurements, so there is a 3σ criterion. The 3σ criterion is the most commonly used and simplest criterion for gross error judgment. It is generally used when the number of measurements is sufficient (n≥30) or when n>10 for rough judgment.

[0061] Optionally, Figure 4Some first quality-poor features obtained from the original signaling data are shown: Combined with user complaints about poor service quality scenarios such as "unable to access the network, unable to access the Internet, slow Internet access, and weak coverage", the signaling plane "access, switching, and release" service documents are selected as the data source: including the main communication protocol processes for users staying in the 45G network, involving "Attach, UE-Initiated Detach Procedure, Network-Initiated Detach Procedure, Tracking area updating, Network triggered Service Request, UE requested PDN connectivity, UE requested PDN disconnect, Network Initiated Dedicated EPS bearer context activation, X2-Based Handover, S1-Based Handover, E-RAB Release, UE The "ContextRelease" process protocol; the user-side "web, video, game, instant messaging" business documents include the main network communication processes of users surfing the Internet and the evaluation information of users' Internet experience, mainly including "DNS connection, TCP handshake establishment, first GET success, web page response speed, web page opening speed, web page packet loss, video buffering speed, video freeze, video clarity, game ping stability, instant messaging text and picture delay speed" to represent the end-to-end Internet process; wireless service documents include the actual network conditions of users using the network, including "uplink and downlink interference, RRC connection, uplink and downlink coverage"; the documents of the three stages from network coverage, user network to user using the network to do business are used as the data source for building business features.

[0062] Step S204, determining a second quality difference feature based on the first quality difference feature, wherein the second quality difference feature is a statistical feature of the first quality difference feature, and the statistical type of the statistical feature includes at least one of the following: mean, range, variance, median, and quantile.

[0063] Optionally, the original detailed order data belongs to asynchronous event data in the continuous time domain. Feature extraction is the difficulty and key point. In addition to the expert-defined features (i.e., the first quality difference features), it is also necessary to use algorithms to extract features (i.e., the second quality difference features). Use featuretools to determine the second quality difference features. In addition to using general statistics such as mean and median, business-significant statistics such as range, variance, and quantile are also used. After adding featuretools to extract features, 80+ second quality difference features are added. Taking variance as an example, 30 second quality difference features are added:

[0064]

[0065]

[0066] Step S206, screening the first quality difference feature and the second quality difference feature, and training the quality difference prediction model based on the target quality difference feature obtained after screening, wherein the quality difference prediction model is used to predict the quality difference object, and the quality difference object is an object whose probability of encountering an abnormal communication scenario meets the preset requirements.

[0067] Optionally, useless features can be removed by using missing rate screening, variance screening / chi-square test / T test, correlation screening, and importance screening steps, and then the tsfresh feature algorithm is used to carefully screen weakly correlated features and significantly correlated features, and select the target quality difference that is strongly correlated with the quality difference prediction model to improve the model execution performance and effect. The number of all features (including the first quality difference features and the second quality difference features) of the feature library constructed based on expert experience and featuretools is greater than 300, and there are a large number of redundant features, which affect the training and reasoning efficiency. It is necessary to optimize the feature screening and select features that are strongly correlated with the quality difference prediction model to improve the model execution performance and effect. The feature selection of this model is distinguished according to the following three categories, and the relevant features are screened and added to the model:

[0068] 1) Useless features: They have no positive effect on the accuracy of the quality recognition model and may bring noise data;

[0069] 2) Weakly correlated features: They are slightly correlated with the accuracy of the quality difference recognition model, and removing these features has little impact on the model;

[0070] 3) Significantly correlated features: These features are strongly correlated with the accuracy of the quality difference identification model, especially features that have business significance and increase the explainability of experience quality differences.

[0071] Generally, feature screening is done by first screening roughly and then finely screening. Useless features are first removed, and then weakly correlated features and significantly correlated features are carefully screened. The main screening steps are as follows: Figure 5 As shown: missing rate screening, variance screening / chi-square test / T test, correlation screening and importance screening are performed in sequence.

[0072] Optionally, screening the first quality difference feature and the second quality difference feature includes: screening the first quality difference feature according to the missing rate of the first quality difference feature, and screening the second quality difference feature according to the missing rate of the second quality difference feature, to obtain a first candidate feature set; determining a difference index determination method matching the feature type according to the feature type of each candidate feature in the first candidate feature set, and determining the difference index of the candidate feature according to the difference index determination method, wherein the difference index is used to reflect the degree of discrimination of the quality difference feature for the quality difference target; screening the candidate features in the first candidate feature set according to the difference index to obtain a second candidate feature set; determining a correlation index between each candidate feature in the second candidate feature set and other candidate features, and screening the candidate features in the second candidate feature set according to the correlation index to obtain a third candidate feature set, wherein the correlation index includes covariance; determining an importance index of each candidate feature in the third candidate feature set, and determining the target quality difference feature from the third candidate feature set according to the importance index, wherein the importance index includes at least one of the following: information gain value, information value.

[0073] Optionally, feature types such as Figure 6 As shown, feature types can be divided into qualitative data and quantitative data according to their nature.

[0074] Optionally, a method for determining a difference index matching the feature type is as follows: Fig. 9 And the following table shows:

[0075] Feature data types Feature data group Analytical methods Classification and Quantification 2 or more groups Variance test Classification and Quantification Only two groups T-test Classification 2 or more groups Chi-square test

[0076] Specifically, the variance test is used to evaluate the overall distribution and variance relationship between batch user complaints and user wireless coverage RSRP; the T test is used to evaluate the service quality of users when the coverage RSRP is good or bad, as well as the correlation of complaints; and the chi-square test is used to evaluate whether users will complain at a certain coverage RSRP value.

[0077] Optionally, the correlation check mainly uses covariance as the main reference standard. Covariance is a value used to measure the degree of correlation between random variables. In this model, the "Pearson" and "Spearman" methods are mainly used to calculate covariance, and the quality difference features with low correlation with other quality difference features are removed according to the covariance. "Pearson" is mainly for linear correlation checks, while "Spearman" is for nonlinear correlation.

[0078] Optionally, importance screening is mainly to detect the predictive ability (i.e. contribution) of poor quality features. The IV value (information value) and information gain (gain) are mainly used as reference standards. Information gain (gain) is mainly tested on O-domain data. The original detailed list belongs to asynchronous event data in the continuous time domain, which intuitively reflects the quality of user experience at the corresponding time point, and the difference in the amount of positive and negative sample data is huge, so information gain (gain) can be used for detection. If the IV value algorithm is used, it may blur the boundary between "good" and "bad". At the same time, since the IV value is related to the variable saturation, it may mis-estimate the importance of poor quality features if the difference is too large. Therefore, it is more appropriate to use information gain (gain) as the criterion for judging the contribution of the model.

[0079] Specifically, information gain represents the average gain that can be achieved by using this poor quality feature in the model. When training the model, each poor quality feature is ranked according to its importance in the model. Figure 7 As shown in Figure 1, this ranking is calculated based on the contribution of the features to the model prediction results during training. These values ​​can be used to evaluate the role of each feature in the model and for tasks such as feature selection.

[0080] Optionally, the missing rate includes a positive sample missing rate and a negative sample missing rate; the first quality difference feature and the second quality difference feature are screened according to the missing rate to obtain a first candidate feature set including: when the positive sample missing rate of the first quality difference feature or the second quality difference feature is higher than a preset missing rate threshold, the quality difference feature is removed; the positive sample missing rate of each of the first quality difference feature and the second quality difference feature is determined, and the first quality difference feature whose positive sample missing rate is higher than the preset missing rate threshold is removed, and the second quality difference feature whose positive sample missing rate is higher than the preset missing rate threshold is removed; when the positive sample missing rate of the first quality difference feature is lower than the preset missing rate threshold and the negative sample missing rate of the quality difference feature is higher than the preset missing rate threshold, the first quality difference feature is data filled; and when the positive sample missing rate of the second quality difference feature is lower than the preset missing rate threshold and the negative sample missing rate of the quality difference feature is higher than the preset missing rate threshold, the second quality difference feature is data filled.

[0081] Specifically, the higher the missing rate of feature values, the less information they carry and the weaker the predictive ability, so features with higher missing rates should be screened out, that is, features with missing rates <0.9. Since the current ratio of negative samples to positive samples is too large, it is an extremely unbalanced state, and the time cost of collecting negative sample data is high, for the first or second poor quality features of data with high negative sample missing rates, direct filling can be used.

[0082] Optionally, training the quality difference prediction model according to the target quality difference feature obtained after screening includes: determining a training sample set, wherein each sample in the training sample set includes the target quality difference feature and a feature value corresponding to the target quality difference feature; adjusting the weights of positive samples and negative samples in the sample set by an extreme gradient boosting model; training the quality difference prediction model by using the following loss function and the samples in the sample set after weight adjustment; FL(p t )=-α t (1-p t ) γ log(p t ) where pt is the probability predicted by the quality difference prediction model, αt is the balance factor, and γ is the focus parameter. αt and γ are used to adjust the weights of positive and negative samples again.

[0083] Optionally, for the unbalanced classification problem of poor quality user identification, considering that unbalanced data classification is prone to overfitting and the characteristics of mobile network data are sparse, the ensemble learning XGBoost model is used to solve the problem of unbalanced data identification of poor quality users. The ratio of poor quality users to non-poor quality users in the existing network is extremely unbalanced, and non-poor quality users are usually more than 20 times the number of poor quality users, which will cause the probability predicted by the model to deviate from the actual probability. XGBoost is selected as the classification model, and Focalloss is used to solve the problem of sample imbalance.

[0084] Optionally, Figure 8 A schematic diagram of an XGBoost model is shown, Figure 8 As shown, XGBoost calculates the gradient and second-order derivative of the loss function for each sample, and builds a new decision tree based on this information. It gradually approaches the optimal solution by iteratively adding new decision trees to fit the residual of the last prediction.

[0085] Each decision tree is an additive model, and the values ​​of its leaf nodes are used as the final prediction results. When training is completed and k trees are obtained, we need to predict the score of a sample. In fact, according to the characteristics of this sample, it will fall into a corresponding leaf node in each tree. Each leaf node corresponds to a score. Finally, we only need to add up the scores corresponding to each tree to get the predicted value of the sample, and control the model complexity through regularization terms to avoid overfitting.

[0086] Focal Loss mainly reduces the weight of easy-to-classify samples by adding a modulation factor and strengthens the focus on difficult-to-classify samples.

[0087] FL(p_t) = -α_t(1-p_t)^γlog(p_t), where p_t is the probability predicted by the model, α_t is the balancing factor, and γ is the focus parameter used to adjust the weight reduction of easy-to-classify samples. The focus parameter γ smoothly adjusts the loss so that even if the correct classification probability of easy-to-classify samples is high, they can continue to participate in training but contribute lower losses.

[0088] XGBoost has significant advantages in dealing with positive and negative sample classification problems. It is based on gradient boosting decision trees, and through efficient parallel training and optimized memory usage, it can quickly process large-scale data sets while maintaining high accuracy. When dealing with unbalanced data sets, XGBoost can adjust sample weights through the Focal loss function, which can make the model pay more attention to poor quality user samples, thereby improving the ability to identify poor quality users.

[0089] Optionally, after training the quality difference prediction model, the method also includes: obtaining the original signaling details data to be predicted, and determining the third quality difference feature based on the original signaling details data to be predicted; inputting the third quality difference feature into the quality difference prediction model, and obtaining the quality difference prediction result output by the quality difference prediction model, wherein the quality difference prediction result includes the quality difference probability indicated by each third quality difference feature; determining the quality difference object based on the quality difference prediction result, wherein the quality difference object includes quality difference users and quality difference cells.

[0090] Optionally, Fig. 9 The process of determining poor quality objects is shown in FIG. Fig. 9 As shown in the figure, based on the inference prediction of poor quality users, the poor quality probability of poor quality users is aggregated in multiple dimensions such as community, POI, city, district, county and province, and the situation of poor quality users and poor quality communities in the whole network is output to guide the existing network to carry out network optimization work accurately.

[0091] Specifically, the third quality difference feature is determined based on the original signaling detail data to be predicted; the third quality difference feature is input into the quality difference prediction model, and the quality difference prediction result output by the quality difference prediction model is obtained, wherein the quality difference prediction result includes the quality difference probability indicated by each third quality difference feature; the quality difference object is determined based on the quality difference prediction result, wherein the quality difference object includes poor quality users and poor quality cells, and network optimization is performed based on the quality difference prediction result (i.e., the evaluation result), thereby improving the network quality.

[0092] Optionally, after obtaining the quality difference prediction result of the quality difference prediction model, the method also includes: sorting each third quality difference feature in descending order according to the corresponding quality difference probability, and determining a preset number of third quality difference features in the front order as the root cause of the quality difference.

[0093] Optionally, the user characteristics of the entire network (i.e., the third quality difference characteristic) are calculated, and the quality difference model is called to match the user characteristics of the entire network to generate an inference result. Based on the inferred quality difference result, the TreeSHAP method is used to interpret the inference result, and the TopN quality difference reasons of each quality difference user are provided based on the contribution of the quality difference characteristic, such as Fig.10 As shown, red feature values ​​increase the predicted value, while blue feature values ​​decrease the predicted value.

[0094] Optionally, the contribution of each poor quality feature is used to calculate the poor quality probability, and the probability of poor quality for users in the entire network and the location of the root cause are predicted and calculated, and a list of concentrated poor quality areas is formed for network optimization and construction, wherein poor quality features are features that lead to poor user service perception, such as low network coverage RSRP value, large service TCP link establishment delay, etc.; non-poor quality features are features that do not lead to poor user service perception, such as good network coverage RSRP, small service TCP link establishment delay, etc.

[0095] Through the above steps, it is possible to identify poor quality users with highly similar features to the complaining users from the entire network, and calculate the feature contribution according to the algorithm to perform feature convergence, achieve feature dimensionality reduction, improve the efficiency and accuracy of feature processing and model reasoning, and solve the problem that traditional rules are unreliable in identifying poor quality users. The difference between the poor quality model prediction rule of this application and the prediction rule of the prior art is as follows: Fig.11As shown: Traditional rule method: The left part of the figure shows how the traditional rule method determines whether a user encounters poor network quality based on a single threshold judgment or a simple rule condition. The limitation of this method is that it may rely too much on preset thresholds, which may not fully reflect the user's actual network experience, and when dealing with complex situations, the setting of rules may become very cumbersome and inflexible. The proposal of this application is to use poor quality features and poor quality prediction models to comprehensively evaluate and predict whether a user encounters poor network quality problems. This includes models trained based on all significant poor quality features, which can process complex network data and user feedback, and identify the deep-seated reasons for poor user experience. For example, the model may find that "the number of data with a TCP second handshake delay (server side) greater than or equal to x seconds for users' online streaming media exceeds n times within one hour" This complex pattern is strongly correlated with poor network quality, and this correlation may not be captured by simple rules. This method outputs a list of poor quality users, combines network conditions to perform poor quality incident cell aggregation analysis and locate the causes of user problems, and then guides the customer service department to focus on potential complaining users for proactive care to achieve the goal of reducing complaints. It can also guide the optimization department to accurately optimize specific problems in problem cells to achieve the goal of optimizing and improving the network, ensuring priority investment of resources and rapid problem resolution, thereby ensuring efficient improvement of customer perception and enhancing the mobile brand image. Different from the existing technology that constructs a poor quality rule system based on a single field of the original signaling detail list and a network KPI indicator method, this application has the following advantages:

[0096] 1. Through the expert feature library based on the featuretools algorithm, the single field is transformed into user network perception features, combined with AI training to generate user face features, and the recognition rate of poor quality users is increased by more than 10%;

[0097] 2. Based on the TreeSHAP algorithm, the most relevant features are selected according to their importance, and feature dimension reduction is achieved while ensuring model accuracy, improving computing efficiency, reducing training time by 60% and improving feature computing efficiency by more than 30%;

[0098] 3. The sample ratio of normal users and complaining users in the current network is more than 20 times, and the sample ratio is extremely unbalanced. This application is based on the XGBoost classification model combined with the Focal loss correction function to solve the sample imbalance problem between poor-quality users and high-quality users, and the model accuracy is improved by more than 10%.

[0099] The present application embodiment provides a network quality difference prediction method, the process is as follows Fig.12 As shown, Fig.12As shown: Combined with the actual data of the existing network, the user feature model is constructed and the poor quality users of the whole network are inferred. By constructing user features for complaining users and non-complaining users, training user identification models, and archiving them as feature model packages; calling the feature model package for reasoning on the user features of the whole network, identifying poor quality users, converging the main poor quality features based on the contribution of poor quality features, and matching and outputting the reasons for poor quality.

[0100] Specifically, the method includes the following steps:

[0101] Step 1: Complaint and non-complaint samples input.

[0102] Based on the actual complaint samples obtained, non-complaint samples (i.e. negative samples) are screened based on a 1:30 ratio through the user's original signaling detail data source, and IoT users are eliminated. Some sample examples are shown in the following table:

[0103]

[0104] Step 2: Complaint and non-complaint sample feature construction and model training.

[0105] Based on complaint and non-complaint samples, sample feature construction (target quality difference feature) and model training are performed: Sample feature construction is as follows: Fig.13 As shown, the model training is as follows Fig.14 shown.

[0106] Step 3: Identify the user characteristics of the entire network (the third quality difference characteristic).

[0107] Step 4: Inference results of users in the entire network: Based on the characteristics of users in the entire network, call the training result model package to infer the poor quality users in the entire network.

[0108] Through the above steps, featuretools is used to extract sample features, and 80+ business features can be added to improve model accuracy; then the most relevant features are screened according to importance based on the TreeSHAP algorithm to solve the problem of feature calculation efficiency. In addition: by using the classification model based on XGBoost combined with the Focal loss function to solve the sample imbalance problem of poor quality users and high quality users, the model pays more attention to the samples of poor quality users, thereby improving the ability to identify poor quality users, making the probability predicted by the model closer to the actual probability, and solving the problem of imbalanced model training samples.

[0109] The present application provides a network quality prediction device. Fig.15 is a schematic diagram of the structure of the device, such as Fig.15As shown, the device includes: an acquisition module 150, which is used to acquire a first quality difference feature from original signaling data, wherein the first quality difference feature is a feature that characterizes the existence of an abnormal communication scenario in a communication network; a determination module 152, which is used to determine a second quality difference feature based on the first quality difference feature, wherein the second quality difference feature is a statistical feature of the first quality difference feature, and the statistical type of the statistical feature includes at least one of the following: mean, range, variance, median, quantile; a screening module 154, which is used to screen the first quality difference feature and the second quality difference feature, and train a quality difference prediction model based on the target quality difference feature obtained after screening, wherein the quality difference prediction model is used to predict a quality difference object, and the quality difference object is an object whose probability of encountering an abnormal communication scenario meets preset requirements.

[0110] In some embodiments of the present application, the screening module 154 screens the first quality difference feature and the second quality difference feature, including: screening the first quality difference feature according to the missing rate of the first quality difference feature, and screening the second quality difference feature according to the missing rate of the second quality difference feature, to obtain a first candidate feature set; determining a difference index determination method that matches the feature type according to the feature type of each candidate feature in the first candidate feature set, and determining the difference index of the candidate feature according to the difference index determination method, wherein the difference index is used to reflect the degree of discrimination of the quality difference feature for the quality difference target; determining the difference index according to the difference index determination method; The method comprises the steps of: screening the candidate features in the first candidate feature set according to the correlation index to obtain the second candidate feature set; determining the correlation index between each candidate feature in the second candidate feature set and other candidate features, and screening the candidate features in the second candidate feature set according to the correlation index to obtain the third candidate feature set, wherein the correlation index includes the covariance; determining the importance index of each candidate feature in the third candidate feature set, and determining the target quality difference feature from the third candidate feature set according to the importance index, wherein the importance index includes at least one of the following: information gain value, information value.

[0111] In some embodiments of the present application, the missing rate includes a positive sample missing rate and a negative sample missing rate; the first quality difference feature and the second quality difference feature are screened according to the missing rate to obtain a first candidate feature set, including: when the positive sample missing rate of the first quality difference feature or the second quality difference feature is higher than a preset missing rate threshold, the quality difference feature is removed; the positive sample missing rate of each of the first quality difference feature and the second quality difference feature is determined, and the first quality difference feature whose positive sample missing rate is higher than the preset missing rate threshold is removed, and the second quality difference feature whose positive sample missing rate is higher than the preset missing rate threshold is removed; when the positive sample missing rate of the first quality difference feature is lower than the preset missing rate threshold and the negative sample missing rate of the quality difference feature is higher than the preset missing rate threshold, the first quality difference feature is data filled; and when the positive sample missing rate of the second quality difference feature is lower than the preset missing rate threshold and the negative sample missing rate of the quality difference feature is higher than the preset missing rate threshold, the second quality difference feature is data filled.

[0112] In some embodiments of the present application, the screening module 154 trains the quality difference prediction model according to the target quality difference feature obtained after screening, including: determining a training sample set, wherein each sample in the training sample set includes the target quality difference feature and the feature value corresponding to the target quality difference feature; adjusting the weights of the positive samples and the negative samples in the sample set by the extreme gradient boosting model; training the quality difference prediction model by the samples in the sample set after the weight adjustment using the following loss function; FL(p t )=-α t (1-p t ) γ log(p t ) where pt is the probability predicted by the quality difference prediction model, αt is the balance factor, and γ is the focus parameter. αt and γ are used to adjust the weights of positive and negative samples again.

[0113] In some embodiments of the present application, after training the quality difference prediction model, the method also includes: obtaining the original signaling details data to be predicted, and determining the third quality difference feature based on the original signaling details data to be predicted; inputting the third quality difference feature into the quality difference prediction model, and obtaining the quality difference prediction result output by the quality difference prediction model, wherein the quality difference prediction result includes the quality difference probability indicated by each third quality difference feature; determining the quality difference object based on the quality difference prediction result, wherein the quality difference object includes quality difference users and quality difference cells.

[0114] In some embodiments of the present application, after obtaining the quality difference prediction result of the quality difference prediction model, the method also includes: sorting each third quality difference feature in order from large to small according to the corresponding quality difference probability, and determining a preset number of third quality difference features ranked first as the root cause of the quality difference.

[0115] In some embodiments of the present application, abnormal communication scenarios include at least one of the following: inability to access the network, the access time consumed to access the network is greater than a preset time, and the signal strength within the coverage area of ​​the base station is lower than a preset strength.

[0116] It should be noted that the various modules in the above-mentioned network quality difference prediction device can be program modules (for example, a set of program instructions that implement a certain specific function) or hardware modules. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above-mentioned modules is a processor, or the functions of each of the above-mentioned modules are implemented by a processor.

[0117] An embodiment of the present application provides a non-volatile storage medium, in which a program is stored, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the following network quality difference prediction method: obtaining a first quality difference feature from original signaling data, wherein the first quality difference feature is a feature that characterizes the existence of an abnormal communication scenario in the communication network; determining a second quality difference feature based on the first quality difference feature, wherein the second quality difference feature is a statistical feature of the first quality difference feature, and the statistical type of the statistical feature includes at least one of the following: mean, range, variance, median, quantile; screening the first quality difference feature and the second quality difference feature, and training a quality difference prediction model based on the target quality difference feature obtained after screening, wherein the quality difference prediction model is used to predict a quality difference object, and the quality difference object is an object whose probability of encountering an abnormal communication scenario meets preset requirements.

[0118] An embodiment of the present application provides an electronic device, including: a memory and a processor, the processor is used to run a program stored in the memory, wherein the following network quality difference prediction method is executed when the program is running: a first quality difference feature is obtained from original signaling data, wherein the first quality difference feature is a feature that characterizes the existence of an abnormal communication scenario in the communication network; a second quality difference feature is determined based on the first quality difference feature, wherein the second quality difference feature is a statistical feature of the first quality difference feature, and the statistical type of the statistical feature includes at least one of the following: mean, range, variance, median, quantile; the first quality difference feature and the second quality difference feature are screened, and a quality difference prediction model is trained based on the target quality difference feature obtained after the screening, wherein the quality difference prediction model is used to predict a quality difference object, and the quality difference object is an object whose probability of encountering an abnormal communication scenario meets preset requirements.

[0119] An embodiment of the present application provides a computer program product, including a computer program, which implements the following network quality difference prediction method when executed by a processor: obtaining a first quality difference feature from original signaling data, wherein the first quality difference feature is a feature that characterizes the existence of an abnormal communication scenario in the communication network; determining a second quality difference feature based on the first quality difference feature, wherein the second quality difference feature is a statistical feature of the first quality difference feature, and the statistical type of the statistical feature includes at least one of the following: mean, range, variance, median, quantile; screening the first quality difference feature and the second quality difference feature, and training a quality difference prediction model based on the target quality difference feature obtained after screening, wherein the quality difference prediction model is used to predict a quality difference object, and the quality difference object is an object whose probability of encountering an abnormal communication scenario meets preset requirements.

[0120] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0121] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0122] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0123] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0124] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0125] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A network quality prediction method, characterized in that: include: Acquire a first quality difference feature from the original signaling data, wherein the first quality difference feature is a feature characterizing the existence of an abnormal communication scenario in the communication network; Determine a second quality difference feature according to the first quality difference feature, wherein the second quality difference feature is a statistical feature of the first quality difference feature, and the statistical type of the statistical feature includes at least one of the following: mean, range, variance, median, and quantile; The first quality difference feature and the second quality difference feature are screened, and a quality difference prediction model is trained based on the target quality difference feature obtained after screening, wherein the quality difference prediction model is used to predict a quality difference object, and the quality difference object is an object whose probability of encountering an abnormal communication scenario meets preset requirements.

2. The network quality difference prediction method according to claim 1, characterized in that: Screening the first quality difference feature and the second quality difference feature includes: The first quality difference feature is screened according to the missing rate of the first quality difference feature, and the second quality difference feature is screened according to the missing rate of the second quality difference feature to obtain a first candidate feature set; According to the feature type of each candidate feature in the first candidate feature set, a difference index determination method matching the feature type is determined, and the difference index of the candidate feature is determined according to the difference index determination method, wherein the difference index is used to reflect the degree of discrimination of the quality difference feature for the quality difference target; Screening the candidate features in the first candidate feature set according to the difference index to obtain a second candidate feature set; Determine a correlation index between each candidate feature in the second candidate feature set and other candidate features, and screen the candidate features in the second candidate feature set according to the correlation index to obtain a third candidate feature set, wherein the correlation index includes covariance; Determine the importance index of each candidate feature in the third candidate feature set, and determine the target quality difference feature from the third candidate feature set based on the importance index, wherein the importance index includes at least one of the following: information gain value, information value.

3. The network quality difference prediction method according to claim 2, characterized in that: The missing rate includes a positive sample missing rate and a negative sample missing rate; the first quality difference feature and the second quality difference feature are screened according to the missing rate, and the first candidate feature set is obtained, including: When the positive sample missing rate of the first quality difference feature or the second quality difference feature is higher than a preset missing rate threshold, removing the quality difference feature; Determine the positive sample missing rate of each of the first quality difference feature and the second quality difference feature, and remove the first quality difference feature whose positive sample missing rate is higher than a preset missing rate threshold, and remove the second quality difference feature whose positive sample missing rate is higher than a preset missing rate threshold; When the positive sample missing rate of the first quality difference feature is lower than the preset missing rate threshold and the negative sample missing rate of the quality difference feature is higher than the preset missing rate threshold, data filling processing is performed on the first quality difference feature; and when the positive sample missing rate of the second quality difference feature is lower than the preset missing rate threshold and the negative sample missing rate of the quality difference feature is higher than the preset missing rate threshold, data filling processing is performed on the second quality difference feature.

4. The network quality difference prediction method according to claim 1, characterized in that: The quality difference prediction model is trained based on the target quality difference features obtained after screening, including: Determine a training sample set, wherein each sample in the training sample set includes the target quality difference feature and a feature value corresponding to the target quality difference feature; Adjusting the weights of positive samples and negative samples in the sample set by an extreme gradient boosting model; The quality difference prediction model is trained by using the following loss function and the samples in the sample set after weight adjustment; FL(p t )=-a t (1-p t ) γ log(p t ) Among them, p t is the probability predicted by the quality difference prediction model, α t is the balance factor, γ is the focus parameter, α t and γ are used to adjust the weights of positive and negative samples again.

5. The network quality difference prediction method according to claim 1, characterized in that: After training the quality difference prediction model, the method further includes: Acquire original signaling detailed data to be predicted, and determine a third quality difference feature according to the original signaling detailed data to be predicted; Inputting the third quality difference feature into the quality difference prediction model, and obtaining a quality difference prediction result output by the quality difference prediction model, wherein the quality difference prediction result includes the quality difference probability indicated by each of the third quality difference features; The poor quality object is determined according to the poor quality prediction result, wherein the poor quality object includes poor quality users and poor quality cells.

6. The network quality difference prediction method according to claim 5, characterized in that: After obtaining the quality difference prediction result of the quality difference prediction model, the method further includes: The third quality difference features are sorted in descending order according to the corresponding quality difference probabilities, and a preset number of the third quality difference features in the front order are determined as the quality difference root causes.

7. The network quality difference prediction method according to claim 1, characterized in that: The abnormal communication scenario includes at least one of the following: being unable to access the network, the access time consumed to access the network being greater than a preset time, and the signal strength within the coverage area of ​​the base station being lower than a preset strength.

8. A network quality prediction device, characterized in that: include: An acquisition module, configured to acquire a first quality difference feature from the original signaling data, wherein the first quality difference feature is a feature characterizing the existence of an abnormal communication scenario in the communication network; A determination module, configured to determine a second quality difference feature based on the first quality difference feature, wherein the second quality difference feature is a statistical feature of the first quality difference feature, and the statistical type of the statistical feature includes at least one of the following: mean, range, variance, median, and quantile; A screening module is used to screen the first quality difference feature and the second quality difference feature, and train a quality difference prediction model based on the target quality difference feature obtained after screening, wherein the quality difference prediction model is used to predict a quality difference object, and the quality difference object is an object whose probability of encountering an abnormal communication scenario meets preset requirements.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores a program, wherein when the program is executed, the device where the non-volatile storage medium is located is controlled to execute the network quality difference prediction method described in any one of claims 1 to 7.

10. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is used to run a program stored in the memory, wherein the network quality difference prediction method according to any one of claims 1 to 7 is executed when the program is run.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the network quality difference prediction method described in any one of claims 1 to 7 are implemented.

Images