Wireless perception poor quality identification method, device, equipment, medium and program product

Abstract

The invention provides a wireless perception poor quality identification method, device and equipment, a medium and a program product, and relates to the technical field of communication, and the method comprises the steps: determining a dynamic threshold value influence factor of a perception index of each cell; the dynamic threshold influence factors comprise at least one of an index trend trend factor, a volatility influence factor, a city influence factor and a scene level influence factor; determining a dynamic threshold of the sensing index according to the initial degradation judgment threshold of the sensing index and the dynamic threshold influence factor; and triggering the work order under the condition that the index value of the sensing index is greater than the dynamic threshold value. According to the method, the work order is triggered through the dynamic threshold value, unnecessary order dispatching is avoided, the problem of false report or missing report is reduced, and therefore the network optimization efficiency is improved.

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 identifying wireless perceptual quality differences. Background Art

[0002] At present, 5G network optimization is mainly centralized optimization, analyzing and processing network performance problems. The specific method is to issue degraded cells as work orders, process work orders and solve problems. Among them, problem discovery is particularly important as the source and object of network optimization work.

[0003] The common method of discovering poor perception quality is to analyze the end-to-end signaling, select the main perception degradation measurement indicators, configure thresholds and time degradation rules, monitor the problems of the entire network, and trigger and dispatch work orders when they meet the rules. However, the above solution uses fixed thresholds to trigger work orders, which cannot adapt to changes in network environment and service quality, resulting in false alarms or missed alarms, thereby reducing network optimization efficiency. Summary of the invention

[0004] The present invention provides a wireless perception quality difference identification method, device, equipment, medium and program product, which are used to solve the defect of low network optimization efficiency caused by using fixed threshold value to trigger work orders in the prior art, and realize triggering work orders through dynamic threshold value, avoid unnecessary dispatching of orders, reduce false alarms or missed alarms, and thus improve network optimization efficiency.

[0005] The present invention provides a method for identifying poor wireless perceptual quality, comprising the following steps:

[0006] Determine a dynamic threshold influencing factor of the perception index of each cell; the dynamic threshold influencing factor includes at least one of an index trend factor, a volatility influencing factor, a city influencing factor, and a scene level influencing factor;

[0007] Determining a dynamic threshold of the perception indicator according to the initial degradation judgment threshold of the perception indicator and the dynamic threshold influencing factor;

[0008] When the indicator value of the perception indicator is greater than the dynamic threshold, a work order is triggered.

[0009] According to a method for identifying poor wireless perceptual quality provided by the present invention, the indicator trend factor is determined based on the following method:

[0010] Obtaining predicted values ​​of the perception indicator in the next multiple days and an average value of the perception indicator within N days;

[0011] The indicator trend factor is determined according to the indicator average value, the indicator prediction value and the first preset adjustment amplitude value.

[0012] According to a method for identifying poor wireless perceptual quality provided by the present invention, the volatility influencing factor is determined based on the following method:

[0013] Determine a first normalized value of a target perception indicator; the target perception indicator is any one of the plurality of perception indicators;

[0014] Determining, according to the first normalized value, second normalized values ​​of a plurality of the perception indicators in a same period as the target perception indicator;

[0015] constructing a sequence set according to the plurality of the second normalized values;

[0016] The volatility influencing factor is determined according to the covariance between the perception indicator and the sequence set, the first variance of the perception indicator, the second variance of the sequence set, and a second preset adjustment amplitude value.

[0017] According to a method for identifying poor wireless perception quality provided by the present invention, obtaining predicted values ​​of the perception indicators in the next multiple days includes:

[0018] The perception index is normalized:

[0019] The normalized perception index of the same network element is input into the prediction model to obtain the index prediction value of the perception index in the next multiple days output by the prediction model; wherein the prediction model is obtained based on Transformer network training.

[0020] According to a method for identifying poor wireless perceptual quality provided by the present invention, the normalization processing of the perceptual index comprises:

[0021] determining a degradation of each of the perception indicators;

[0022] If the degradation condition is no degradation, the perception index is normalized to 0;

[0023] If the degradation condition is that the perception indicator reaches a degradation judgment threshold, normalizing the perception indicator to 1;

[0024] If the degradation condition is that the perception indicator exceeds the degradation judgment threshold, a normalized value of the perception indicator is determined according to the indicator value of the perception indicator.

[0025] According to a method for identifying poor wireless perceptual quality provided by the present invention, when the value of the perceptual indicator is greater than the dynamic threshold, after triggering a work order, the method further includes:

[0026] Determine a deviation value of the degree of degradation of the perception indicator according to an average value of the perception indicator within N days and an indicator value of the perception indicator within a set time period;

[0027] According to the number of degraded cells and the number of degradation problems, the number of degraded cell categories that have triggered work orders is determined;

[0028] Determining the priority of the work order according to the degradation degree deviation value of the perception indicator and the number of the degraded cell categories;

[0029] The work order is dispatched according to the priority of the work order.

[0030] According to a method for identifying poor wireless perceptual quality provided by the present invention, when the value of the perceptual indicator is greater than the dynamic threshold, after triggering a work order, the method further includes:

[0031] Determine the main degradation problem type of the work order;

[0032] Merging the work orders according to the main degradation problem types of the work orders;

[0033] Generate a core issue work order based on the merged work order.

[0034] The present invention also provides a wireless perceptual quality difference identification device, comprising the following modules:

[0035] A dynamic threshold influencing factor determination module, used to determine the dynamic threshold influencing factor of the perception index of each cell; the dynamic threshold influencing factor includes at least one of an index trend factor, a volatility influencing factor, a city influencing factor and a scene level influencing factor;

[0036] A dynamic threshold determination module, used to determine the dynamic threshold of the perception indicator according to the initial degradation judgment threshold of the perception indicator and the dynamic threshold influencing factor;

[0037] The perceptual quality difference recognition module is used to trigger a work order when the indicator value of the perceptual indicator is greater than the dynamic threshold.

[0038] The present invention also provides 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 computer program, any of the wireless perceptual quality difference identification methods described above is implemented.

[0039] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the wireless perceptual quality difference identification method described in any one of the above is implemented.

[0040] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the wireless perceptual quality difference identification method described above is implemented.

[0041] The wireless perception quality difference identification method, device, equipment, medium and program product provided by the present invention determine the dynamic threshold influencing factors of the perception index of each cell; the dynamic threshold influencing factors include at least one of the index trend factor, volatility influencing factor, city influencing factor and scene level influencing factor; according to the initial degradation judgment threshold of the perception index and the dynamic threshold influencing factor, the dynamic threshold of the perception index is determined; when the index value of the perception index is greater than the dynamic threshold, a work order is triggered. The present invention triggers the work order through the dynamic threshold, avoids unnecessary dispatching, reduces the problem of false alarm or missed alarm, and thus improves the efficiency of network optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces 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 creative work.

[0043] Figure 1 It is a flow chart of the wireless perception quality difference identification method provided by the present invention.

[0044] Figure 2 It is a schematic diagram of the process of perceptual quality difference discovery provided by the present invention.

[0045] Figure 3 It is a flow chart of the overall idea of ​​the 5G network problem discovery and mining method provided by the present invention.

[0046] Figure 4 It is a flowchart of the perceptual quality difference prediction and identification method provided by the present invention.

[0047] Figure 5 It is a schematic diagram of data association mining provided by the present invention.

[0048] Figure 6 It is a schematic diagram of indicator prediction of the Transformer model provided by the present invention.

[0049] Figure 7 It is a structural schematic diagram of the wireless perception quality difference identification device provided by the present invention.

[0050] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] At present, the perceptual quality optimization recognition has the following disadvantages:

[0053] (1) Identification of unnecessary problems wastes manpower and material resources: Based on unified automatic identification rules, it is impossible to accurately identify perception degradation problems, resulting in unnecessary problem identification and waste of production resources.

[0054] (2) Poor perceived quality risks cannot be accurately identified: Based on fixed rules, it is impossible to explore poor perceived quality risks, discover potential hidden dangers and solve them in advance.

[0055] (3) Problem self-recovery affects work progress: In the production of poor quality and excellent network, there are a large number of self-recovery problems, which are automatically recovered during the processing due to factors such as the improvement of indicator trends, resulting in a waste of resources.

[0056] (4) The identification scope is highly limited: In terms of identification objects, it mainly conducts mining at the base station cell network element level; in terms of identification scope, it mainly monitors the degradation of perception indicators of end-to-end signaling and dispatches orders.

[0057] Based on the above problems, the present invention proposes a method for identifying poor wireless perceptual quality.

[0058] Combine the following Figure 1-Figure 8 The present invention describes the wireless perceptual quality difference identification method, device, equipment, medium and program product.

[0059] Figure 1 is a flow chart of the wireless perceptual quality difference identification method provided by the present invention, such as Figure 1 As shown, the method includes the following:

[0060] Step 1001: determine the dynamic threshold influencing factors of the perception index of each cell.

[0061] In an embodiment of the present invention, the dynamic threshold influencing factor includes at least one of an indicator trend factor, a volatility influencing factor, a city influencing factor, and a scene level influencing factor. The perception index includes an uplink packet loss rate, a downlink packet loss rate, an average processing delay, a low channel ratio, and a low user downlink rate.

[0062] The factors affecting each dynamic threshold are determined as follows:

[0063] (1) Indicator trend factor Z(k): Obtain the indicator prediction value of the perception indicator in the next multiple days and the indicator average value of the perception indicator within N days, and then determine the indicator trend factor based on the indicator average value, the indicator prediction value and the first preset adjustment amplitude value.

[0064] For example, the Transformer model is used to predict the predicted value of the perception index in the next three days, such as (k n+1 , k n+2 , k n+3 ), and then combine the N-day average indicator Calculate indicator trends and deterioration levels.

[0065] The average value of the perception index K is calculated as:

[0066]

[0067] Among them, k i represents the i-th index, n represents the number of indicators, Indicates the average value of the indicator.

[0068] The indicator trend factor Z(k) is judged to be deteriorating or improving, and the maximum adjustment range is set to z∈(0,1), that is, the first preset adjustment range value, linear transition, the improvement decreases, and the trend increases sensitivity. Among them, the calculation method of Z(k) is:

[0069]

[0070] (2) Indicator volatility influencing factor R(k): Determine a first normalized value of a target perception indicator, where the target perception indicator is any one of a plurality of perception indicators; then, based on the first normalized value, determine a second normalized value of a plurality of perception indicators in the same period as the target perception indicator; further, based on the plurality of second normalized values, construct a sequence set; and finally, determine the volatility influencing factor based on the covariance between the perception indicator and the sequence set, the first variance of the perception indicator, the second variance of the sequence set, and the second preset adjustment amplitude value.

[0071] For example, suppose the input perception index K (k1, k2, ...k n , continuous n-day indicators); based on the normalization degree of the target perception indicator X new (i.e. the first normalized value), calculate the average normalized x of multiple indicators in the same period n =Ave(X new ) (i.e., the second normalized value), and obtain the sequence set X (x1, x2, ..., x) at time T. n );

[0072] Calculate the first variance of the perceptual index K:

[0073] Compute the sequence set X(x1, x2, ..., x n )’s second variance:

[0074]

[0075] Calculate the perceptual index K and the sequence set X(x1, x2, ..., x n )’s covariance:

[0076]

[0077] The index volatility influencing factor R(k) with the Pearson correlation coefficient added is:

[0078]

[0079] Among them, r∈(0,1) represents setting a maximum adjustment amplitude reference, that is, a second preset adjustment amplitude value.

[0080] (3) Key city influencing factors L(k): 49 key cities increase threshold sensitivity, and the index fluctuates sharply to reduce threshold sensitivity. The maximum adjustment range is set as 1∈(0,1), then:

[0081] The first-class cities are 4+8 cities, the second-class cities are 49-class cities (excluding 4+8 cities), and the others are general cities. Based on this, the influencing factor L(k) of key cities is:

[0082]

[0083] (4) Scene level influencing factor J(k): The key scene increases the threshold sensitivity, and the drastic fluctuation of the index reduces the threshold sensitivity. The maximum adjustment range is set as j∈(0,1), then:

[0084] VVIP is a preset important area with the highest priority; VIP level is residential buildings, commercial streets, universities, high-speed railways, highways and other scenes with medium priority, and other rural scenes are low scene levels.

[0085] The coefficient is adjusted to

[0086] Step 1002: Determine the dynamic threshold of the perception indicator according to the initial degradation judgment threshold of the perception indicator and the dynamic threshold influencing factor.

[0087] In the embodiment of the present invention, the overall idea of ​​dynamic threshold includes initial threshold setting, indicator trend factor Z(k), volatility influencing factor R(k), key city influencing factor L(k), and scene level influencing factor J(k). The calculation method of dynamic threshold is:

[0088] Q mn =Δ mn *Z(k)*R(k)*L(k)*J(k);

[0089] Among them, Δ mn represents the initial degradation judgment threshold, Z(k) represents the indicator trend factor, R(k) represents the volatility influencing factor, L(k) represents the key city influencing factor, and J(k) represents the scenario level influencing factor.

[0090] Step 1003: trigger a work order when the indicator value of the perception indicator is greater than the dynamic threshold.

[0091] After determining the dynamic threshold of the perception indicator, the indicator value of the perception indicator is compared with the dynamic threshold. If the indicator value of the perception indicator is greater than the dynamic threshold, it means that the service quality has been reduced to a level that the user may perceive as unacceptable, and a work order is triggered.

[0092] Furthermore, based on the average value of the perception indicator within N days and the indicator value of the perception indicator within a set time period, the degradation degree deviation value of the perception indicator is determined; based on the number of degraded cells and the number of degradation problems, the number of degraded cell categories that have triggered work orders is determined; based on the degradation degree deviation value of the perception indicator and the number of degraded cell categories, the priority of the work order is determined; and based on the priority of the work order, the work order is dispatched.

[0093] For example, after the dynamic threshold calculation is completed, the indicators of the last 3 days (k n-2 , k n-1 , k n ) are compared with the dynamic thresholds respectively. If the degradation degree reaches the dynamic threshold judgment threshold, it is judged as a degraded cell e and a work order is triggered.

[0094] The calculation method of the work order priority p is: p=x+y; where x represents the deviation of the degradation degree of indicator K in the last three days; y represents the number of degradation indicator categories, that is, the number of degraded cell categories that trigger the work order.

[0095] The calculation method of deviation x is:

[0096] The number of problem types to be judged is c, then the number of degradation indicator categories is: y = num(e) / c;

[0097] The overall priority P of the work order is:

[0098]

[0099] The priority judgment (threshold is adjustable) is:

[0100] Among them, a,b∈(0,1), a is the low priority judgment threshold, b is the high priority judgment threshold, b>a.

[0101] Furthermore, the main degradation problem type of the work order is determined, and then the work orders are merged according to the main degradation problem type of the work order. Finally, the core problem work order is generated according to the merged work orders. For example, according to the degradation cycle, the sample point set of the work order to be generated is determined, and the influence value of each involved element on other elements is calculated, that is, o(K ij ), and calculate as follows:

[0102]

[0103] Among them, P represents the duration of the combined degradation cycle, o represents K ij Duration of appearance, K ij Indicates the degree of degradation, F(K ij ) represents the sum of the degrees of degradation.

[0104] By comparing the elements o(K ij ) size and the value of correlation, we can determine the merging and correlation, and determine the main degradation problem type. We can optimize the judgment of the main degradation problem of the work order, merge and reduce the work order, use the merged items as auxiliary judgment, and realize the generation and output of the core problem work order.

[0105] By classifying and grouping various types of perception work orders, we can achieve resource integration and collaborative optimization of similar problems from the source of network optimization, thereby improving the efficiency of network optimization work. For example, first, classify various types of perception work orders, and then group similar perception work orders, that is, group work orders with similar properties and solutions together, so as to avoid repeatedly solving the same or similar problems during the processing process and save manpower and material resources. Organizing and integrating appropriate resources, including personnel, tools, and time, for each work order helps to effectively schedule and allocate resources to ensure that each problem is solved in a timely and appropriate manner. Collaborative optimization of similar problems means adopting an overall optimization strategy for a class of problems instead of processing each work order separately. This can be achieved through batch processing, intelligent algorithms, or experience sharing to improve the efficiency and quality of solving problems.

[0106] The wireless perception quality difference identification method provided by the embodiment of the present invention determines the dynamic threshold influencing factors of the perception indicators of each cell; the dynamic threshold influencing factors include at least one of the indicator trend factor, volatility influencing factor, city influencing factor and scene level influencing factor; according to the initial degradation judgment threshold of the perception indicator and the dynamic threshold influencing factor, the dynamic threshold of the perception indicator is determined; when the indicator value of the perception indicator is greater than the dynamic threshold, a work order is triggered. The present invention triggers the work order through the dynamic threshold, avoids unnecessary dispatching, reduces the problem of false alarm or missed alarm, and thus improves the efficiency of network optimization.

[0107] Based on the above embodiment, obtaining the predicted index value of the perception index in the next multiple days includes:

[0108] Step 1010, normalizing the perception index:

[0109] Step 1011, input the normalized perception index of the same network element into the prediction model to obtain the index prediction value of the perception index in the next multiple days output by the prediction model; wherein the prediction model is obtained based on Transformer network training.

[0110] For the indicator values ​​of 4G KQI perception problems and 5G KQI perception problems, standard deviation normalization is selected according to data distribution and business requirements.

[0111] Step 3011, standardization is performed using the standard deviation as a unit, and the formula is as follows:

[0112]

[0113] Among them, X new represents the normalized data value, X s represents the original data value, μ represents the mean of the original data, and σ represents the standard deviation of the original data.

[0114] Based on the above normalization, the indicators are normalized, but the importance and sensitivity of different types of indicators are different. In order to solve the problem that different types of indicators cannot distinguish the severity in conventional normalization, the conventional normalization algorithm is improved to determine the degradation of each perception indicator; if the degradation is no degradation, the perception indicator is normalized to 0; if the degradation is that the perception indicator reaches the degradation judgment threshold, the perception indicator is normalized to 1; if the degradation is that the perception indicator exceeds the degradation judgment threshold, the normalized value of the perception indicator is determined according to the indicator value of the perception indicator.

[0115] For example, completely no problem is set to 0, indicating the value of the indicator in the best state; just degraded to the threshold is set to 1, indicating the value when the indicator just reaches the acceptable minimum threshold; no problem to degradation threshold is set as 1 interval, and the value becomes larger and larger when the degradation exceeds the threshold, indicating that the more the indicator develops towards the degraded state, the value will gradually increase.

[0116] For each indicator in the 4G KQI perception problem and the 5G KQI perception problem, its value is mapped to a normal interval [0, 1] or a degraded interval [1, +∞] to ensure that all indicators are compared within the same value range.

[0117] The prediction model is pre-trained based on the Transformer network. The normalized perception indicators of the same network element are input into the prediction model to obtain the predicted values ​​of the perception indicators output by the prediction model for the next several days. For example, the historical KQI indicators are trained through Encoder encoding and Decoder decoding to generate a model to predict the trend of the subsequent KQI indicators and obtain the predicted values ​​for the next three days (k n+1 , k n+2 , k n+3 ).

[0118] The embodiment of the present invention uses the Transformer algorithm to predict, mine indicators with positive trends or large volatility, and perform dynamic threshold adjustments to increase the difficulty of discovering problems that are easy to self-recover, thereby reducing the number of problems that are easy to self-recover, reducing the dispatch of unnecessary processing work orders, saving manpower and material resources, and achieving cost reduction and efficiency improvement.

[0119] In order to further analyze and illustrate the wireless perceptual quality difference identification method proposed in the present invention, refer to the following embodiments.

[0120] The purpose of the embodiment of the present invention is to provide a wireless perceptual quality AI prediction and identification method to address the defects of the current wireless network perceptual quality discovery, so as to achieve the optimal high-value problem discovery for network perception improvement work, realize accurate perceptual quality identification, and conduct potential mining for user perception improvement. Among them, the position of perceptual quality discovery in network optimization work is Figure 2 The content marked in the solid box (that is, the box where "Auto Discovery" is located).

[0121] 1. The overall idea of ​​5G network problem discovery and mining methods.

[0122] like Figure 3As shown in the figure, the base station cell network element, user level and scenario level are used as identification objects, and application types such as instant messaging, short video, web browsing, payment services, game services, VONR (VoNR Voice over New Radio, new air interface carrying voice) are introduced. Historical data such as packet loss, delay, access, retention, channel quality, and rate are input. Single network element modeling is performed based on the Transformer large model AI algorithm to achieve the change trend prediction of the indicator sequence. The problem triggering benchmark threshold rules are formulated, and factors such as prediction trends, key cities, key scenarios, long and short cycles, and volatility are introduced to adjust the dynamic threshold of problem triggering, and to determine whether to trigger the work order to identify the perception quality problem. Through the above method, user-level, scenario and problem cell prediction and identification are realized, which can effectively and accurately identify potential perception quality problems and avoid occasional self-recovery problems that lead to resource waste.

[0123] 1.1) Perception data mining.

[0124] Data input: Input user, base station cell, scenario and other network element information, mainly including the city, region (urban, rural), equipment type (indoor, macro), coverage scenario (residential area, university, high-speed rail), scenario level (VVIP, VIP, general scenario) and other information.

[0125] Input N consecutive days of data (N>=90) of performance perception signaling data and calculate the index value as the basic data source for sequence prediction.

[0126] Degradation classification and association: Degradation association and classification are performed based on the focus indicators of different application services. The horizontal service application types are divided into performance type, voice type (VONR, VINR), instant messaging, short video, web browsing, payment service, game service and other types. The vertical degradation types mainly include packet loss type, delay type, channel quality type, access type, retention type, rate type and other types. A two-dimensional association matrix is ​​formed according to the focus of different service application types.

[0127] 1.2) Transformer modeling prediction.

[0128] The wireless perceptual quality AI prediction and identification method aims to accurately predict the future value of the cell KQI (Key Quality Indicators) indicator through historical data analysis and AI algorithms based on the Transformer model, so as to help network operators and related technical teams better understand and respond to changes and trends in cell performance. Specific purposes include:

[0129] Analyze the periodic and trend characteristics of historical KQI indicator data in order to accurately grasp the fluctuation law and long-term change trend of cell performance.

[0130] By using the AI ​​algorithm of the Transformer model, a future value prediction model of the 5G cell KQI indicator for a single network element is established to provide accurate future KQI indicator value prediction results for optimal resource allocation.

[0131] 1.3) Accurate identification of dynamic thresholds.

[0132] Establish benchmark threshold rules for problem triggering, introduce factors such as predicted trends, key cities, key scenarios, long and short cycles, volatility, etc., adjust the dynamic threshold for problem triggering, and determine whether to trigger a work order to identify perceived poor quality issues.

[0133] Based on expert experience, benchmark threshold rules are formulated, including degradation indicator judgment thresholds, trigger limit thresholds, and time rule thresholds. Based on the prediction results, trend characteristics (Transformer prediction characteristics), long-short cycle characteristics, and fluctuation characteristics (Pearson algorithm) are introduced to adjust the dynamic threshold of the benchmark threshold. Based on factors such as the city characteristics (key cities) and scene characteristics (VIP, VVIP, general scenes) of the network element, the dynamic threshold is adjusted again; based on the correlation between the dynamic threshold and recent indicators, it is judged whether the network is degraded and whether a work order is triggered. The work order priority is judged based on the degradation scale, scene level, and degradation degree (degree normalization algorithm), and the work order information is output.

[0134] 1.4) System design and effects.

[0135] The system function design is mainly divided into four parts: problem perception rule configuration, AI training module, dynamic threshold judgment, and work order identification module.

[0136] Problem mining and configuration mainly involves discovering the underlying data flow and problems of the system architecture, and performing algorithm configuration and function configuration. The generative large model Transformer algorithm is introduced for modeling. The training module performs AI modeling training and forecasts the trend for the next three days based on the mined sample points. The dynamic threshold correction algorithm is introduced to correct the fixed threshold based on the trend, volatility, and periodicity characteristics of the AI ​​prediction training results, combined with urban characteristics and scene characteristics, and perform identification and judgment of perceived quality problems. The work order generation module mainly generates work orders, as well as calculates and judges the primary and secondary degradations, merges them, and presents them in GIS.

[0137] 2. The steps for implementing the perceptual quality difference prediction and recognition method are as follows: Figure 4 shown.

[0138] Based on the application of the Transformer algorithm model in perceptual quality difference prediction, the implementation steps are explained:

[0139] Step 100, data input: input network element information such as users, base station cells, scenarios, and performance perception signaling data of N consecutive days (N>=90) as the basic data source for sequence prediction and recognition;

[0140] Step 200, data processing: mining the key indicators that affect different perception scenarios, grouping and classifying them, calculating the indicators for multiple consecutive days, and studying the correlation between their mutual influences;

[0141] Step 300, AI training: Modeling is performed based on the AI ​​algorithm of the Transformer model. Through the self-attention mechanism and feedforward neural network of the continuous sequence (indicator), cyclic training revision is performed, and algorithms are introduced for correlation and long and short cycles to achieve the goal of future indicator prediction.

[0142] Step 400, dynamic threshold calculation: formulate problem triggering benchmark threshold rules, introduce factors such as forecast trends, key cities, key scenarios, long and short cycles, volatility, etc., adjust the problem triggering dynamic threshold, and determine whether to trigger a work order to identify perceived poor quality problems;

[0143] Step 500, problem work order identification: Use the model trained by AI modeling to calculate the newly added problem degradation elements, obtain the primary and secondary degradation problem judgments of the work order, and apply them in production.

[0144] Furthermore, the above step 100 further includes step 101 and step 102:

[0145] Step 101, PM (Persistent Memory) performance data, end-to-end perception data and other underlying data are collected and accessed as data sources for degradation problem mining, and data access, cleaning and storage are performed. The format of the data fields after aggregation is shown in the following table:

[0146] Field Name Business Classification Standard Time (yyyy-mm-ddhh:mm:ss) 5G cgi 5G eci 5G Community name 5G Number of successful IM logins Instant Messaging 5G Playback smoothness denominator (Tencent Video) Video Services 5G Download rate < 550kbps times (instant messaging) Instant Messaging 5G Video download effective time Video Services 5G Number of service delays > 4s (web browsing) Web browsing 5G The number of uplink and downlink traffic <= 500kB (web browsing) Web browsing 5G Number of service delays > 4s (paid) Payment Services 5G Uplink and downlink traffic <= 500kB times (paid) Payment Services 5G Number of bad quality calls Video Services 5G Denominator of poor quality call records Video Services 5G Payment business payment success rate numerator Payment Services 5G Payment business payment success rate denominator (number of times) Payment Services 5G Video traffic (KB) Video Services 5G Page traffic (KB) Web browsing 5G Total VoNR users End-to-end signaling 5G Number of users with a connection rate below 70% End-to-end signaling 5G 5. Number of legacy flows of QI1 (QI1) voice 5G 5QI1 switch-in flow number (QI1) voice 5G

[0147] Step 102: data cleaning.

[0148] Clean and organize the data, count the same network element and the same indicator K in the time sequence H continuously Counter (C1, C2, C3, ..., C h ), handle missing values ​​to ensure the accuracy and completeness of the data. If the missing ratio (valid data fields / total data fields) is less than 80%, the data will be discarded.

[0149] The interpolation process is performed using linear interpolation. The steps are as follows:

[0150] (1) Determine the location of missing values: First, determine the location of missing values ​​in the data, that is, the data points to be linearly interpolated;

[0151] (2) Find adjacent data points: For the missing value location, these two data points are the data points before and after the missing value location;

[0152] (3) Calculate the slope: Use the linear relationship between known data points to calculate the slope of linear interpolation. Assume that the missing value position (C n , H n )The previous data point is (C n-1 , H n-1 ), the next data point is (C n+1 , H n+1 ), then the slope k is: k = (C n+1 -C n-1 ) / (H n+1 -H n-1 ).

[0153] (4) Filling missing values: Use the linear interpolation formula to interpolate and fill the missing value position based on the known data points and the calculated slope. Assuming that the missing value position to be filled is (x, y), and the known data points are (x1, y1) and (x2, y2), the linear interpolation formula can be used for filling, that is:

[0154] C n =C n-1 +(H n -H n-1 )×(C n+1 -C n-1 ) / (H n+1 -H n-1 );

[0155] (5) Verify the results: After filling in the missing values, check whether the data is consistent with the actual situation and verify the accuracy and rationality of the interpolation;

[0156] (6) Data reconstruction: After completing the linear interpolation filling of missing values, the data can be organized into a complete data set to ensure the continuity and integrity of the data.

[0157] Furthermore, the above step 200 further includes step 201 and step 202:

[0158] Step 201, data calculation:

[0159] Based on the performance perception signaling data source, the impact perception data indicators are mined, a large number of network element indicators are summarized and calculated, and the database is organized according to the time series. For example, the impact perception indicator items (examples) are shown in the following table:

[0160] Indicator name Standard Business Classification 5G page quality is poor 5G Web browsing IM login success rate 5G Instant Messaging Playback success rate (Tik Tok) 5G Video Services Playback success rate (Tencent) 5G Video Services Proportion of high latency in small instant messaging packets 5G Instant Messaging The proportion of small packets with high latency when browsing web pages 5G Web browsing Page display success rate 5G Web browsing Average page display time 5G Web browsing Percentage of small payment packets with high latency 5G Payment Services Poor perceived quality - VR video playback delay is too long 5G Video Services Poor perceived quality - poor payment success rate for payment services 5G Payment Services Poor perceived quality - poor video quality 5G Video Services Game login success rate (%) 5G Gaming Business Residential area with high proportion of low-connected users 5G End-to-end signaling Poor video service quality area 5G End-to-end signaling VoNR service flow establishment success rate (5QI1) 5G voice ViNR service flow establishment success rate (5QI2) 5G voice

[0161] Step 202: perform correlation matching based on the characteristics of the work order problem items:

[0162] like Figure 5 As shown in the figure, from the perspective of data and work order types, it is mainly divided into continuous numerical data and discrete data. Different algorithms need to be matched for data mining. Horizontally, KQI, voice, and data (instant messaging, short video, web browsing, game business, and payment business) are grouped and classified. Vertically, uplink packet loss rate, downlink packet loss rate, average processing delay, low channel ratio, low user downlink rate, switching type, connection type, and disconnection type are grouped and classified to form a two-dimensional indicator matrix associated with perceived degradation.

[0163] For the horizontal m application types and vertical n indicator types, the data group features are refined to form a two-dimensional indicator matrix, as shown below:

[0164] Perception

[0165] The n indicators of the mth perception application type form a continuous sequence in time for t consecutive hours, forming a prediction sequence matrix:

[0166]

[0167] Furthermore, the above step 300 further includes step 301, step 302 and step 303:

[0168] Step 301: data preprocessing.

[0169] For the indicator values ​​of 4G KQI perception problems and 5G KQI perception problems, standard deviation normalization is selected according to data distribution and business requirements.

[0170] Step 3011, standardization is performed using the standard deviation as a unit, and the formula is as follows:

[0171]

[0172] Among them, X new represents the normalized data value, X s represents the original data value, μ represents the mean of the original data, and σ represents the standard deviation of the original data.

[0173] Based on the above normalization, the indicators are normalized, but the importance and sensitivity of different types of indicators are different. In order to solve the problem that different types of indicators cannot distinguish the severity in conventional normalization, the conventional normalization algorithm is improved, and completely no problem is set to 0, indicating the value of the indicator in the best state; just deteriorating to the threshold is set to 1, indicating the value when the indicator just reaches the acceptable minimum threshold; no problem to deterioration threshold is set as 1 interval, and the value becomes larger and larger when the deterioration exceeds the threshold, indicating that the indicator will gradually increase as it develops towards the deteriorated state.

[0174] Step 3012, normalize each indicator separately: for each indicator in the 4G KQI perception problem and the 5G KQI perception problem, map its value to the normal interval [0, 1] or the degraded interval [1, +∞] to ensure that all indicators are compared within the same value range. For example, as shown in the following table:

[0175]

[0176] Based on the above indicator normalization algorithm, there are three main categories: the larger the indicator percentage, the better (100% no problem), the smaller the indicator percentage, the better (0% no problem), and the threshold category (delay category, number of users, interference threshold, etc.). Summarize and sort out different indicators to obtain the degradation degree assignment K n , forming the index K assignment matrix, the specific method is as follows:

[0177]

[0178] Among them, k jn It represents the indicator value in the time period of hour n, k is the problem threshold value, and k0 is the value without any problem.

[0179] The degradation degree assignment matrix of K is obtained as follows:

[0180]

[0181] Step 302: Model building.

[0182] The model to be used is an encoder-decoder Transformer, where the encoder part takes the history of the KQI indicator data time series as input, and the decoder part predicts future values ​​in an autoregressive manner. The decoder is connected to the encoder using an attention mechanism.

[0183] like Figure 6 As shown in the figure, the historical KQI indicators are trained by encoder encoding and decoder decoding, and the model is generated to predict the trend of the subsequent KQI indicators, and the forecast value for the next three days (k n+1 , kn+2 , k n+3 ) for result output and subsequent calculations.

[0184] Encoder: The structure of the Transformer encoder consists of the following main components:

[0185] (1) Input Embedding: The KQI indicators or tags in the input sequence are converted into their corresponding vector representations and finally into QKV vectors.

[0186] X n =(x1, x2, x3, ..., x i , ..., x n );

[0187] Q=X n *w1;

[0188] K=X n *w2;

[0189] V=X n *w3;

[0190] Among them, x i represents the corresponding i-th indicator information, and n represents the number of all predicted indicators. n Represents the vector corresponding to all indicators predicted at one time. w1, w2, and w3 all represent weights, and Q, K, and V represent query, key, and value, respectively, as three linear transformation matrices in the self-attention mechanism.

[0191] (2) Self-Attention Mechanism: The self-attention mechanism allows the model to integrate information from different positions into the representation by calculating the attention weight for each position in the input sequence.

[0192]

[0193] Among them, d k The number of columns of the matrix representing QKV, i.e., the vector dimension. First, the similarity between the query vector Q and the key vector K is calculated, the similarity is divided by the square root of the scaling factor, the Softmax function is applied to the scaled similarity, and finally, the value vector V is weighted summed using the attention weight.

[0194] (3) Multi-Head Attention: The multi-head attention mechanism allows the model to learn and capture different semantic information in different representation subspaces, improving the model’s ability to model sequence information.

[0195] MultiHead(Q,K,V)=Concat(head1,...,head h )W o ;

[0196] Among them, W o Representing the weight information of multi-head attention, the outputs of all attention heads are concatenated in the channel dimension to form the final output of multi-head attention. The concatenated vector is then projected through another linear transformation layer (usually a fully connected layer) to produce the final multi-head attention representation.

[0197] (4) Feed-Forward Neural Network: Feed-forward neural network is usually followed by attention mechanism, which is used to perform nonlinear transformation and mapping of attention representation, thereby enhancing the model's ability to abstractly represent sequence data.

[0198] FFN(x)=max(0,xW1+b1)W2+b2;

[0199] Among them, the input variables include:

[0200] x: input vector, representing the representation or features of the previous layer, which captures the dependencies between the KQI indicators of each position in the input sequence;

[0201] W1: The weight matrix of the first linear transformation, used to perform a linear transformation on the input vector;

[0202] b1: The bias term of the first linear transformation, used for the offset of the linear transformation;

[0203] W2: The weight matrix of the second linear transformation, used to perform a linear transformation on the result of the first linear transformation;

[0204] b2: Bias term of the second linear transformation, used for the offset of the second linear transformation.

[0205] Intermediate Variables include:

[0206] xW1+b1: The intermediate representation after the first linear transformation, usually called the hidden layer representation or feature representation;

[0207] max(0,xW1+b1): The intermediate representation after the ReLU (rectified linear unit) activation function, which sets negative values ​​to zero and retains positive values;

[0208] (xW1+b1)W2+b2: The final output after the second linear transformation, usually a new feature representation or representation vector.

[0209] The input variables are the input features or representations of the model, while the intermediate variables are the hidden layer representations generated between the two linear transformations, and the final output is the new feature representation after the linear transformation and activation function processing.

[0210] Decoder: The decoder is an important part of the model and is used to handle sequence-to-sequence learning tasks. The decoder usually has the following key components:

[0211] (1) Self-Attention Mechanism: Similar to the encoder part, the decoder uses the self-attention mechanism to understand the dependencies between different positions in the input sequence. Through the self-attention mechanism, the decoder can better understand the previously generated sequence part.

[0212]

[0213] Among them, the query vector (Q) represents the hidden state of the current position, which is used to measure the degree of association between the current position and each position in the encoder output sequence; the key (K) key vector is used to calculate the attention weight to measure the correlation between the current position and each position in the encoder output sequence; the value vector (V) is used to perform weighted summation of the encoder output sequence according to the attention weight to obtain information related to the current position. k The number of columns of the QKV matrix, i.e. the vector dimension. The similarity calculation is similar to the encoder method.

[0214] (2) Multi-Head Attention: The multi-head attention mechanism enables the decoder to learn semantic information in different representation subspaces in parallel, which helps the model better capture semantic information at different levels.

[0215] MultiHead(Q,K,V)=Concat(head1,...,head h )W o ;

[0216] Among them, W o Represents the weight information of multi-head attention, concatenates the outputs of all attention heads in the channel dimension to form the final output of multi-head attention.

[0217] (3) Feed-Forward Neural Network: Feed-forward neural network is usually followed by attention mechanism, which is used to perform nonlinear transformation and mapping of attention representation, thereby enhancing the model's ability to abstractly represent sequence data.

[0218] FFN(x)=max(0,xW1+b1)W2+b2;

[0219] Among them, the input variables include:

[0220] x: input vector, representing the representation or features of the previous layer;

[0221] W1 is the weight matrix of the first linear transformation.

[0222] b1 is the bias term of the first linear transformation, which is used for the offset of the linear transformation;

[0223] W2 is the weight matrix of the second linear transformation, which is used to perform a linear transformation on the result of the first linear transformation;

[0224] b2 is the bias term of the second linear transformation, which is used for the offset of the second linear transformation.

[0225] Intermediate Variables include:

[0226] xW1+b1: The intermediate representation after the first linear transformation, usually called the hidden layer representation or feature representation;

[0227] max(0,xW1+b1): The intermediate representation after the ReLU (rectified linear unit) activation function, which sets negative values ​​to zero and retains positive values;

[0228] (xW1+b1)W2+b2: The final output after the second linear transformation, usually a new feature representation or representation vector.

[0229] (4) Output header information.

[0230] At the end of the decoder, the output header involves input and output. Based on this, the output header will perform the following operations:

[0231] Linear Transformation: First, the output of the last self-attention layer of the decoder is mapped to the target tag space through a linear transformation. The linear transformation usually consists of a weight matrix and a bias vector, and its formula can be expressed as:

[0232] Linear_Transformation=Decoder_Output×W+b;

[0233] The output representation of the last self-attention layer of the decoder, W represents the learnable weight matrix, and b represents the learnable bias vector. Then, the Softmax function is applied to the result of the linear transformation to convert it into a probability distribution for predicting the probability of the target tag. The formula of the Softmax function is:

[0234]

[0235] The calculation of the output head Y allows the model to predict the probability distribution of the target token KQI based on the final representation of the decoder. By sampling this probability distribution or selecting the token with the highest probability as the next prediction, the decoder can gradually generate the target sequence.

[0236] Step 303: Output and verify model training results

[0237] Use the trained model to predict the 5G KQI index in the future time period and provide the prediction results of the future value of the KQI index (k n+1 , k n+2 , k n+3 ).

[0238] Furthermore, the above step 400 further includes step 401, step 402 and step 403:

[0239] Through the historical KQI indicator data of the community, the periodic characteristics and trend characteristics of the historical performance data are detected. Combined with basic information, the AI ​​algorithm of the intelligent Transformer model is used to perform unsupervised learning and predictive analysis on the indicators of the degraded community in the first 90 days, and calculate the abnormal fluctuations and degradation trends of the performance indicators.

[0240] Combined with the trend of problem indicators such as high latency, high packet loss, high quality difference, and low rate in the community, it is determined whether there is a risk of perception deterioration. Combined with the city to which the community belongs (such as whether it is one of the 49 key cities), the importance of the scenario to which it belongs, the indicator trend, indicator fluctuations, and the indicators affected by the application scenario, the initial threshold is corrected to achieve dynamic threshold judgment and work order prediction and identification.

[0241] Step 401, dynamic threshold calculation model.

[0242] The overall idea of ​​dynamic threshold includes the factors of indicator trend, volatility, key cities, and scenario level. The overall idea is to increase the sensitivity of dynamic threshold in terms of downward trend, key cities, key scenarios, and low volatility, so that it is easier to trigger work orders, and vice versa. Assume that the indicator configuration initial degradation judgment threshold Δ mn , dynamic threshold Q mn The judgment method is:

[0243] Q mn =Δ mn *Z(k)*R(k)*L(k)*J(k);

[0244] Among them, Z(k) represents the indicator trend factor, R(k) represents the volatility influencing factor, L(k) represents the key city influencing factor, and J(k) represents the scenario level influencing factor.

[0245] Step 402, detailed calculation of dynamic threshold influencing factors.

[0246] The overall idea of ​​dynamic threshold includes initial threshold setting, indicator trend factor Z(k), volatility influencing factor R(k), key city influencing factor L(k), scenario level influencing factor J(k), and specific implementation methods.

[0247] The implementation steps and calculation logic are as follows:

[0248] Step 4021, for the kth mn Set the initial degradation judgment threshold of the indicator Δ mn .

[0249]

[0250] Based on the above grouping and classification, a threshold matrix B is formed:

[0251]

[0252] Step 4022, indicator trend factor Z(k).

[0253] Based on the indicator value (k n+1 , k n+2 , k n+3 ), combined with the average value of the N-day indicator Calculate indicator trends and deterioration levels.

[0254] The average value of K is calculated as:

[0255]

[0256] Among them, k i represents the i-th index, and n represents the number of indicators.

[0257] The indicator trend factor Z(k) is judged to be deteriorating or improving, and the maximum adjustment range is set to z∈(0,1). The linear transition reduces the sensitivity when the trend is good, and increases the sensitivity when the trend is bad. The calculation method of Z(k) is:

[0258]

[0259] Step 4023, index volatility influencing factor R(k).

[0260] The index fluctuates dramatically to reduce the threshold sensitivity, and the maximum adjustment range benchmark is set to r∈(0,1). The volatility of R(k) Pearson calculation is -1 to 1. The volatility of the correlation between the fluctuation of index K and the average normalized degree X of the cell in the same period is studied.

[0261] Input index K (k1, k2, ...k n , continuous n-day index); based on the normalization degree of an indicator X new , calculate the average normalized x of multiple indicators in the same period n =Ave(X new ), and obtain the sequence set X(x1, x2, ..., x) of time T n );

[0262] Calculate the variance of indicator K:

[0263] Calculate the normalized sequence set X(x1, x2, ..., x n )’s variance:

[0264]

[0265] Calculate the index K and the normalized sequence set X(x1, x2, ..., x n )’s covariance:

[0266]

[0267] The index volatility influencing factor R(k) with the Pearson correlation coefficient added is:

[0268]

[0269] Step 4024, key city influencing factors L(k).

[0270] The threshold sensitivity of 49 key cities is increased, and the threshold sensitivity is reduced when the index fluctuates sharply. The maximum adjustment range is set as 1∈(0,1), then:

[0271] The first-class cities are 4+8 cities, the second-class cities are 49-class cities (excluding 4+8 cities), and the others are general cities. Based on this, the influencing factor L(k) of key cities is:

[0272]

[0273] Step 4025, focus on scene level influencing factors J(k).

[0274] The threshold sensitivity is increased in key scenarios, and the threshold sensitivity is reduced when the index fluctuates violently. The maximum adjustment range is set as j∈(0,1), then:

[0275] VVIP is a preset important area with the highest priority; VIP level is residential buildings, commercial streets, universities, high-speed railways, highways and other scenes with medium priority, and other rural scenes are low scene levels.

[0276] The coefficient is adjusted to

[0277] Step 403: dynamic threshold recognition and judgment.

[0278] The overall idea of ​​dynamic threshold includes indicator trend factors, volatility factors, key city factors and scenario level factors. The dynamic threshold calculation result is:

[0279]

[0280] Among them, Z(k) is the indicator trend ratio, R(k) is the degree of Pearson correlation coefficient, L(k) is based on the city classification; J(k) is based on the scene classification.

[0281] After the dynamic threshold calculation is completed, the indicators of the last 3 days (k n-2 , k n-1 , k n ) are compared with the dynamic thresholds respectively. If the degradation degree reaches the dynamic threshold judgment threshold, it is judged as a degraded cell e and a work order is triggered.

[0282] The calculation method of the work order priority p is: p=x+y; where x represents the deviation of the degradation degree of indicator K in the last three days; y represents the number of degradation indicator categories, that is, the number of degraded cell categories that trigger the work order.

[0283] The calculation method of deviation x is:

[0284] The number of problem types to be judged is c, then the number of degradation indicator categories is: y = num(e) / c;

[0285] The overall priority P of the work order is:

[0286]

[0287] The priority judgment (threshold is adjustable) is:

[0288] Among them, a,b∈(0,1), a is the low priority judgment threshold, b is the high priority judgment threshold, b>a.

[0289] Furthermore, the above step 300 further includes step 501 and step 502:

[0290] Step 501, problem work order identification and judgment.

[0291] According to the degradation cycle, determine the sample point set of the work order to be generated, and calculate the impact value of each involved element on other elements, that is, o(K ij ), and calculate as follows:

[0292]

[0293] Among them, P represents the duration of the combined degradation cycle, o represents K ij Duration of appearance, K ij Indicates the degree of degradation, F(K ij ) represents the sum of the degrees of degradation.

[0294] By comparing the elements o(K ij ) sizes and their interrelated values ​​can be used to determine the merging and correlation situations, identify the main degradation problem type, and thus generate a work order.

[0295] Output information: cell ECI (E-UTRAN Cell Identifier), whether it is a degraded cell, whether it is a work order, k n+1 , k n+2 , k n+3 , work order priority.

[0296] Step 502, perceive poor quality production application.

[0297] Determine the main degradation issues of the optimization work order, merge and reduce the work orders, use the merged items as auxiliary judgments, and realize the generation and output of core problem work orders, as well as GIS rendering presentation.

[0298] The embodiment of the present invention realizes the application of 5G perception quality difference algorithm in network centralized optimization. Through perception data mining, a two-dimensional matrix of horizontal business scenario classification and vertical problem degradation type affecting perception indicators is generated. As the source of centralized optimization work, it has higher advancement than the problem discovery link in the traditional network optimization link. Further, by studying the data cleaning method, the slope algorithm of adjacent data points is introduced based on the associated data and the previous and next effective values. Based on the previous and next effective values ​​and the associated effective values, the default value is supplemented and the abnormal value is corrected, thereby improving the data accuracy. Through the data cleaning and sorting matched by the algorithm, the data accuracy and completeness of the discovery of network problems can be effectively achieved, and the accuracy of user perception recognition can be improved. Further, the Transformer algorithm is used for indicator prediction, and the sequence of indicator K at time T is used to predict the future indicator situation and is applied to dynamic threshold correction. Based on the KQI indicator of continuous time period, the KQI indicator or mark in the sequence is converted into its corresponding vector, and the self-attention mechanism is introduced to capture the dependency between different indicators. Nonlinear transformation and mapping are performed based on the feedforward neural network, encoder encoding and decoder decoding training are realized, and the generation model predicts the trend of the subsequent KQI indicator to obtain the future prediction value. Furthermore, based on the abnormal fluctuation and degradation prediction results of the AI ​​algorithm of the Transformer model, combined with the trend of problem indicators such as high latency, high packet loss, high quality difference, and low rate in the community, it is judged whether there is a risk of perception deterioration. In combination with the city to which the community belongs (whether it is one of the 49 key cities), the importance of the scenario to which it belongs, the indicator trend, the indicator fluctuation, and the indicators affected by the application scenario, the initial threshold correction is performed to achieve dynamic threshold judgment and work order prediction and identification.

[0299] Compared with the related art, the present invention has the following technical advantages:

[0300] 1. Identification of unnecessary problems consumes manpower and material resources: Based on dynamic threshold identification, unnecessary dispatching is reduced, and various types of perceptions are grouped, optimized and managed during the network optimization process. Resource integration and collaborative optimization of similar problems are achieved from the source of network optimization, thereby improving the efficiency of network optimization work.

[0301] 2. Accurate identification of hidden dangers of poor perceptual quality: Based on the dynamic threshold recognition and judgment method, factors such as indicator trend, indicator volatility, core cities, and key scenarios are introduced to predict hidden dangers of poor perceptual quality and identify them in advance, so as to achieve comprehensive exploration of core and key perceptual quality differences.

[0302] 3. Reduction of self-recovery problems: Through Transformer algorithm prediction, we can mine indicators with positive trends or large volatility, make dynamic threshold adjustments, increase the difficulty of discovering problems that are easy to self-recover, reduce the number of problems that are easy to self-recover, reduce the dispatch of unnecessary work orders, save manpower and material resources, and achieve cost reduction and efficiency improvement.

[0303] 4. Collaborative optimization of multiple perception quality sources: Various types of perception are grouped, optimized and managed during the network optimization process, and there are duplications in the same type of work order cells, which requires a lot of manpower and material resources. The embodiment of the present invention classifies and groups various types of perception work orders, realizes resource integration and collaborative optimization of similar problems from the source of network optimization, and improves the efficiency of network optimization work.

[0304] The following is a description of a wireless perceptual quality difference identification device provided by the present invention. The wireless perceptual quality difference identification device described below and the wireless perceptual quality difference identification method described above can be referred to each other.

[0305] refer to Figure 7 The wireless perceptual quality difference identification device provided by the present invention includes a dynamic threshold influencing factor determination module 701, a dynamic threshold determination module 702 and a perceptual quality difference identification module 703.

[0306] The dynamic threshold influencing factor determination module 701 is used to determine the dynamic threshold influencing factor of the perception index of each cell; the dynamic threshold influencing factor includes at least one of the index trend factor, the volatility influencing factor, the city influencing factor and the scene level influencing factor;

[0307] A dynamic threshold determination module 702, configured to determine a dynamic threshold of the perception indicator according to an initial degradation judgment threshold of the perception indicator and the dynamic threshold influencing factor;

[0308] The perceptual quality difference identification module 703 is used to trigger a work order when the indicator value of the perceptual indicator is greater than the dynamic threshold.

[0309] The wireless perception quality difference identification device provided by the embodiment of the present invention determines the dynamic threshold influencing factors of the perception index of each cell; the dynamic threshold influencing factors include at least one of the index trend factor, volatility influencing factor, city influencing factor and scene level influencing factor; according to the initial degradation judgment threshold of the perception index and the dynamic threshold influencing factor, the dynamic threshold of the perception index is determined; when the index value of the perception index is greater than the dynamic threshold, a work order is triggered. The present invention triggers the work order through the dynamic threshold, avoids unnecessary dispatching, reduces the problem of false alarm or missed alarm, and thus improves the efficiency of network optimization.

[0310] In one embodiment, the dynamic threshold influencing factor determination module 701 is further used to:

[0311] Obtaining predicted values ​​of the perception indicator in the next multiple days and an average value of the perception indicator within N days;

[0312] The indicator trend factor is determined according to the indicator average value, the indicator prediction value and the first preset adjustment amplitude value.

[0313] In one embodiment, the dynamic threshold influencing factor determination module 701 is further used to:

[0314] Determine a first normalized value of a target perception indicator; the target perception indicator is any one of the plurality of perception indicators;

[0315] Determining, according to the first normalized value, second normalized values ​​of a plurality of the perception indicators in a same period as the target perception indicator;

[0316] constructing a sequence set according to the plurality of the second normalized values;

[0317] The volatility influencing factor is determined according to the covariance between the perception indicator and the sequence set, the first variance of the perception indicator, the second variance of the sequence set, and a second preset adjustment amplitude value.

[0318] In one embodiment, the dynamic threshold influencing factor determination module 701 is further used to:

[0319] The perception index is normalized:

[0320] The normalized perception index of the same network element is input into the prediction model to obtain the index prediction value of the perception index in the next multiple days output by the prediction model; wherein the prediction model is obtained based on Transformer network training.

[0321] In one embodiment, the dynamic threshold influencing factor determination module 701 is further used to:

[0322] determining a degradation of each of the perception indicators;

[0323] If the degradation condition is no degradation, the perception index is normalized to 0;

[0324] If the degradation condition is that the perception indicator reaches a degradation judgment threshold, normalizing the perception indicator to 1;

[0325] If the degradation condition is that the perception indicator exceeds the degradation judgment threshold, a normalized value of the perception indicator is determined according to the indicator value of the perception indicator.

[0326] In one embodiment, the perceptual quality difference recognition module 703 is further configured to:

[0327] Determine a deviation value of the degree of degradation of the perception indicator according to an average value of the perception indicator within N days and an indicator value of the perception indicator within a set time period;

[0328] According to the number of degraded cells and the number of degradation problems, the number of degraded cell categories that have triggered work orders is determined;

[0329] Determining the priority of the work order according to the degradation degree deviation value of the perception indicator and the number of the degraded cell categories;

[0330] The work order is dispatched according to the priority of the work order.

[0331] In one embodiment, the perceptual quality difference recognition module 703 is further configured to:

[0332] Determine the main degradation problem type of the work order;

[0333] Merging the work orders according to the main degradation problem types of the work orders;

[0334] Generate a core issue work order based on the merged work order.

[0335] Figure 8 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 complete mutual communication through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the wireless perception quality difference identification method, the method comprising: determining the dynamic threshold influencing factors of the perception index of each cell; the dynamic threshold influencing factors include at least one of the index trend factor, the volatility influencing factor, the city influencing factor and the scene level influencing factor; determining the dynamic threshold of the perception index according to the initial degradation judgment threshold of the perception index and the dynamic threshold influencing factor; triggering a work order when the index value of the perception index is greater than the dynamic threshold.

[0336] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, 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, and the computer software product is stored in a storage medium, including a number of 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 invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0337] On the other hand, the present invention also provides a computer program product, which includes a computer program, and 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 execute the wireless perception quality difference identification method provided by the above-mentioned methods, and the method includes: determining the dynamic threshold influencing factors of the perception indicators of each cell; the dynamic threshold influencing factors include at least one of the indicator trend factor, volatility influencing factor, city influencing factor and scene level influencing factor; determining the dynamic threshold of the perception indicator according to the initial degradation judgment threshold of the perception indicator and the dynamic threshold influencing factor; triggering a work order when the indicator value of the perception indicator is greater than the dynamic threshold.

[0338] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the wireless perception quality difference identification method provided by the above-mentioned methods, the method comprising: determining a dynamic threshold influencing factor of the perception indicator of each cell; the dynamic threshold influencing factor comprises at least one of an indicator trend factor, a volatility influencing factor, a city influencing factor and a scene level influencing factor; determining the dynamic threshold of the perception indicator based on the initial degradation judgment threshold of the perception indicator and the dynamic threshold influencing factor; triggering a work order when the indicator value of the perception indicator is greater than the dynamic threshold.

[0339] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0340] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for 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 some parts of the embodiments.

[0341] 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 embodiments of the present invention.

Claims

1. A method for identifying poor wireless perceptual quality, characterized in that: include: Determine the dynamic threshold influencing factors of the perception indicators of each cell; The dynamic threshold influencing factor includes at least one of an indicator trend factor, a volatility influencing factor, a city influencing factor and a scene level influencing factor; Determining a dynamic threshold of the perception indicator according to the initial degradation judgment threshold of the perception indicator and the dynamic threshold influencing factor; When the indicator value of the perception indicator is greater than the dynamic threshold, a work order is triggered.

2. The method for identifying poor wireless perceptual quality according to claim 1, characterized in that: The indicator trend factors are determined based on the following method: Obtaining predicted values ​​of the perception indicator in the next multiple days and an average value of the perception indicator within N days; The indicator trend factor is determined according to the indicator average value, the indicator prediction value and the first preset adjustment amplitude value.

3. The method for identifying poor wireless perceptual quality according to claim 1, characterized in that: The volatility influencing factors are determined based on the following method: Determine a first normalized value of a target perception indicator; the target perception indicator is any one of the plurality of perception indicators; Determining, according to the first normalized value, second normalized values ​​of a plurality of the perception indicators in a same period as the target perception indicator; constructing a sequence set according to the plurality of the second normalized values; The volatility influencing factor is determined according to the covariance between the perception indicator and the sequence set, the first variance of the perception indicator, the second variance of the sequence set, and a second preset adjustment amplitude value.

4. The method for identifying wireless perceptual quality difference according to claim 2, characterized in that: Obtain the predicted values ​​of the perception indicators for multiple days in the future, including: The perception index is normalized: The normalized perception index of the same network element is input into the prediction model to obtain the index prediction value of the perception index in the next multiple days output by the prediction model; wherein the prediction model is obtained based on Transformer network training.

5. The method for identifying poor wireless perceptual quality according to claim 4, characterized in that: The normalizing the perception index includes: determining a degradation of each of the perception indicators; If the degradation condition is no degradation, the perception index is normalized to 0; If the degradation condition is that the perception indicator reaches a degradation judgment threshold, normalizing the perception indicator to 1; If the degradation condition is that the perception indicator exceeds the degradation judgment threshold, a normalized value of the perception indicator is determined according to the indicator value of the perception indicator.

6. The method for identifying poor wireless perceptual quality according to claim 1, characterized in that: When the indicator value of the perception indicator is greater than the dynamic threshold, after the work order is triggered, the method further includes: Determine a deviation value of the degree of degradation of the perception indicator according to an average value of the perception indicator within N days and an indicator value of the perception indicator within a set time period; According to the number of degraded cells and the number of degradation problems, the number of degraded cell categories that have triggered work orders is determined; Determining the priority of the work order according to the degradation degree deviation value of the perception indicator and the number of the degraded cell categories; The work order is dispatched according to the priority of the work order.

7. The method for identifying poor wireless perceptual quality according to claim 1, characterized in that: When the indicator value of the perception indicator is greater than the dynamic threshold, after the work order is triggered, the method further includes: Determine the main degradation problem type of the work order; Merging the work orders according to the main degradation problem types of the work orders; Generate a core issue work order based on the merged work order.

8. A wireless perceptual quality difference identification device, characterized in that: include: A dynamic threshold influencing factor determination module, used to determine the dynamic threshold influencing factor of the perception indicator of each cell; The dynamic threshold influencing factor includes at least one of an indicator trend factor, a volatility influencing factor, a city influencing factor and a scene level influencing factor; A dynamic threshold determination module, used to determine the dynamic threshold of the perception indicator according to the initial degradation judgment threshold of the perception indicator and the dynamic threshold influencing factor; The perceptual quality difference recognition module is used to trigger a work order when the indicator value of the perceptual indicator is greater than the dynamic threshold.

9. 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 computer program, the method for identifying wireless perceptual quality difference according to any one of claims 1 to 7 is implemented.

10. 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 method for identifying wireless perceptual quality difference according to any one of claims 1 to 7 is implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for identifying wireless perceptual quality difference according to any one of claims 1 to 7 is implemented.

Images