Abnormity prediction method, device and equipment of indoor distribution passive device, medium and product

Abstract

The invention provides an anomaly prediction method, device, equipment, medium and product for an indoor distribution passive device, and the method comprises the steps: obtaining the anomaly information, network parameter adjustment features and alarm features of an index anomaly cell; calculating abnormal index correlation between the index abnormal cell and adjacent cells in different ranges, and generating abnormal index characteristics according to the abnormal index correlation; inputting the abnormal information, the network parameter adjustment feature, the alarm feature and the abnormal index feature into a preset abnormal prediction model, and predicting whether there is an abnormal indoor distribution passive device in the index abnormal cell; wherein the anomaly prediction model is obtained based on index anomaly cell sample training. According to the technical scheme provided by the invention, whether the indoor distribution passive device is abnormal or not is predicted in advance through the abnormality prediction model, and the troubleshooting efficiency and accuracy of the indoor distribution passive device are improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to an abnormality prediction method, device, equipment, medium and product for indoor passive devices. Background Art

[0002] With the continuous advancement of technology and the continuous improvement of the network, the widespread application of 5G technology has greatly increased the requirements for data transmission speed and stability. With the rapid increase of mobile users in cities and the increasing number of high-rise buildings, the traffic density and coverage requirements are also rising. Professional indoor distribution systems are used to enhance signal coverage capabilities to solve the impact of building structures and materials on indoor communication quality. It can solve indoor blind spot problems, balance network traffic, strengthen indoor coverage signals, improve service quality and expand network capacity, and provide users with more stable, reliable and high-quality communication services.

[0003] The existing maintenance methods for indoor distribution systems are mostly passive responses after an alarm has occurred or a complaint has been received. The passive response processing method can no longer meet the high-quality requirements of operators for the network. The passive components of the indoor distribution system cannot generate alarms themselves. At present, the problem of weak indoor coverage mainly relies on on-site positioning and troubleshooting by maintenance personnel, which is inefficient and has poor accuracy. There is a lack of standardized and effective identification and positioning methods for the hidden dangers of indoor passive components. If indoor passive components are damaged or aged, the troubleshooting time is long and the positioning is difficult, which seriously affects the perception and experience of indoor users. Summary of the invention

[0004] The present invention provides a method, device, equipment, medium and product for predicting abnormalities of indoor passive components, so as to solve the defects of low efficiency and poor accuracy of indoor passive component troubleshooting in the prior art, and to improve the efficiency and accuracy of indoor passive component troubleshooting.

[0005] The present invention provides a method for predicting abnormality of indoor passive devices, comprising: Obtain abnormal information of cells with abnormal indicators, network parameter adjustment characteristics, and alarm characteristics; Calculating the abnormal index correlation between the abnormal index cell and neighboring cells in different ranges, and generating abnormal index features according to the abnormal index correlations; The abnormal information, the network parameter adjustment characteristics, the alarm characteristics and the abnormal index characteristics are input into a preset abnormal prediction model to predict whether there is an abnormal indoor passive device in the abnormal index cell; wherein the abnormal prediction model is trained based on the abnormal index cell samples.

[0006] According to a method for predicting abnormalities of indoor passive devices provided by the present invention, the calculation of the abnormal index correlation between the abnormal index cell and adjacent cells within different ranges, and the generation of abnormal index features according to the abnormal index correlations, includes: Based on the location information of the cell with abnormal indicators and the location information of the neighboring cells, distance information between the cell with abnormal indicators and each of the neighboring cells is calculated by dynamic weighting; Determine a plurality of neighboring cell sets based on the distance information, and calculate the correlation between the abnormal indicator cell and the abnormal indicator of each of the neighboring cell sets; The abnormal indicator feature is generated according to the correlation of each abnormal indicator.

[0007] According to a method for predicting abnormalities of indoor passive devices provided by the present invention, the method of determining a plurality of neighboring cell sets based on the distance information and calculating the correlation between the abnormal indicator cell and the abnormal indicator of each of the neighboring cell sets includes: Based on the distance information, calculating the Pearson correlation coefficient between the indicator abnormal cell and each of the neighboring cell sets; The abnormal indicator correlation is determined according to the Pearson correlation coefficient.

[0008] According to a method for predicting abnormality of a room-divided passive device provided by the present invention, the process of constructing the abnormality prediction model includes: Acquire training data of multiple samples of abnormal indicator cells; wherein the training data includes training samples and sample labels; The abnormal cell samples of each indicator are input into the basic abnormal prediction model for iterative training until the training is completed to obtain the abnormal prediction model; wherein the basic abnormal prediction model is obtained based on a combined long short-term memory network and a Transformer model.

[0009] According to a method for predicting abnormality of indoor passive devices provided by the present invention, before obtaining abnormal information of cells with abnormal indicators, network parameter adjustment characteristics, and alarm characteristics, the method further includes: Obtaining first alarm information of the cell with abnormal indicators, second alarm information of a neighboring cell co-located with the cell with abnormal indicators, and third alarm information of a neighboring cell not co-located with the cell with abnormal indicators; Calculate the number of alarms at a multi-dimensional alarm level based on the first alarm information, the second alarm information, and the third alarm information; wherein the multi-dimensional alarm level includes a radio remote unit out-of-service alarm, a transmission alarm, a base station out-of-service alarm, a cell out-of-service alarm, a performance alarm, a master device alarm, and a standing wave alarm; The alarm feature is obtained based on the alarm number.

[0010] According to a method for predicting abnormalities of indoor passive devices provided by the present invention, the number of alarms at a multi-dimensional alarm level is calculated based on the first alarm information, the second alarm information, and the third alarm information, including: At each alarm level, obtain first alarm weights corresponding to the first alarm information, the second alarm information, and the third alarm information, respectively, and calculate corresponding second alarm weights based on the first alarm information, the second alarm information, and the third alarm information, respectively; Based on the first alarm weight, the second alarm weight, the first alarm information, the second alarm information and the third alarm information, the number of alarms at the corresponding alarm level is weightedly calculated.

[0011] According to a method for predicting abnormality of indoor passive devices provided by the present invention, before obtaining abnormal information of cells with abnormal indicators, network parameter adjustment characteristics, and alarm characteristics, the method further includes: Obtain performance data of multiple indoor cells; Processing the performance data by using the interquartile range method, and obtaining target performance data of each of the indoor cells based on the processing results; An index threshold is calculated based on the target performance data, and the index abnormal cell is determined from the plurality of indoor cells based on the index threshold and the target performance data.

[0012] The present invention also provides an abnormality prediction device for indoor passive devices, comprising: A first acquisition module is configured to acquire abnormal information of cells with abnormal indicators, network parameter adjustment characteristics, and alarm characteristics; A first calculation module is configured to calculate the abnormal index correlation between the abnormal index cell and neighboring cells in different ranges, and generate abnormal index features according to the abnormal index correlation; The prediction module is configured to input the abnormal information, the network parameter adjustment characteristics, the alarm characteristics and the abnormal index characteristics into a preset abnormal prediction model to predict whether there is an abnormal indoor passive device in the abnormal index cell; wherein the abnormal prediction model is trained based on the abnormal index cell sample.

[0013] 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 program, the method for predicting abnormalities of indoor passive devices as described in any one of the above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for predicting abnormalities of indoor passive devices as described in any one of the above is implemented.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the abnormality prediction method of the indoor passive device as described in any one of the above is implemented.

[0016] The abnormal prediction method, device, equipment, medium and product of indoor passive devices provided by the present invention calculate the abnormal index correlation between the abnormal index cell and the neighboring cells within different ranges, and generate abnormal index features according to the correlation of each abnormal index, consider the influence between the abnormal cell and the neighboring cell, and then combine the abnormal information of the abnormal cell, the network parameter adjustment feature and the alarm feature of the abnormal cell, and realize the abnormal prediction of indoor passive devices through the abnormal prediction model, which can effectively improve the troubleshooting efficiency of problems such as weak coverage of indoor cells, improve the accuracy of pre-anomaly prediction of indoor passive devices, and further improve the work efficiency of network optimization and maintenance. Through the technical solution provided by the present invention, it is not necessary to carry out a large number of manual tests on site, and the abnormality of indoor passive devices can be accurately located before the fault occurs, which greatly promotes the improvement of work efficiency and effectively reduces the difficulty of work. By predicting the abnormality of indoor passive devices in advance, the present invention can guide the allocation and scheduling of operation and maintenance resources, realize the abnormality troubleshooting of high-risk indoor sites, and improve the efficiency and quality of network operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] 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.

[0018] Figure 1 It is a flow chart of the abnormality prediction method of indoor passive devices provided by the present invention.

[0019] Figure 2 It is a box plot of the interquartile range method in the embodiment provided by the present invention.

[0020] Figure 3 Schematic diagram of the LSTM neuron structure in the embodiment provided by the present invention.

[0021] Figure 4 It is a schematic diagram of the structure of the self-attention mechanism and the cross-sequence attention mechanism in the embodiment provided by the present invention.

[0022] Figure 5It is a schematic diagram of the structure of the Transformer model in the embodiment provided by the present invention. Figure 6 It is a schematic diagram of the structure of the basic abnormality prediction model in the embodiment provided by the present invention.

[0023] Figure 7 It is a schematic diagram of the structure of the abnormality prediction device for indoor passive devices provided by the present invention.

[0024] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0025] 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.

[0026] Indoor distribution systems use a large number of passive devices. Unlike active devices, passive devices will not generate alarms or other prompt information when a fault occurs. In addition, indoor passive devices are closer to areas where people do daily activities and are easily damaged by human activities, resulting in an increase in communication system failures and hidden dangers. Since passive devices are far away from base station equipment in most cases and have many branches, fault detection data is not obvious. Once a fault or hidden danger occurs, it is difficult to detect, which will affect the user experience for a long time. How to accurately and timely predict sites with abnormal indoor passive devices before indoor passive devices fail, and improve the efficiency of identifying and troubleshooting hidden dangers of indoor passive devices, has become the key to improving network operation quality, improving user experience, and guiding operation and maintenance personnel to conduct on-site troubleshooting.

[0027] At present, the hidden danger identification method of indoor distribution system passive components has two main shortcomings: First, it relies too much on operation and maintenance personnel to conduct on-site testing, and analyzes hidden dangers and fault conditions based on the test results. The time interval between each test of this method is not fixed, and hidden dangers cannot be discovered in time, which affects the normal user experience, and it relies heavily on the experience of on-site operators. Second, at this stage, there is a lack of methods to predict hidden dangers of indoor distribution system passive components in advance. The only way is to go to the site to gradually check the indoor distribution system based on user complaints. In addition, since passive components are uncontrollable, unobservable, and without alarms, in the event of component aging, performance degradation, or damage due to human factors, it is impossible to check and locate the problem in time, which is time-consuming and labor-intensive, inefficient, and has poor identification accuracy.

[0028] The current methods for identifying hidden dangers in indoor distribution systems cannot accurately identify hidden dangers in indoor passive components, such as problems with splitters and combiners. In addition, when identifying hidden dangers in indoor passive components, the impact on neighboring cells is not fully considered. When a passive component in a certain indoor cell has a fault or hidden danger, neighboring cells at different distances will be affected to varying degrees. This impact will be reflected in some business indicators, such as performance management data. The existing methods do not fully utilize the correlation between these indicators and neighboring cells for effective association, and also use the distance between the neighboring cell and the cell with abnormal indicators as a feature for judging abnormal passive components in indoor cells when constructing features.

[0029] In view of the above problems, the present invention provides a method, device, equipment, medium and product for predicting abnormality of indoor passive devices. Figure 1-Figure 8 The present invention describes the abnormality prediction method, device, equipment, medium and product of indoor passive devices.

[0030] Figure 1 FIG. 1 is a flow chart of a method for predicting abnormality of a passive device in a room according to an exemplary embodiment. Figure 1 As shown, in an exemplary embodiment, the abnormality prediction method of a room-divided passive device includes steps 110 to 130, which are described in detail as follows.

[0031] Step 110, obtaining abnormal information, network parameter adjustment characteristics, and alarm characteristics of the cell with abnormal indicators.

[0032] In the embodiment of the present invention, the abnormal information includes the type and duration of abnormal indicators of the indicator-abnormal cell in the past period of time.

[0033] The network parameter adjustment features include whether there is a parameter modification record for the cell with abnormal indicators, the type of parameter that has changed, the amount of change in the parameter modification value, the parameter modification date, the parameter modification record of the neighboring cell, the type of parameter modification of the neighboring cell, the change value of the parameter modification of the neighboring cell, the parameter modification date of the neighboring cell, etc.

[0034] The alarm characteristics are obtained based on the alarm data of the cell with abnormal indicators, the alarm data of the co-sited neighboring cells, and the alarm data of the non-co-sited neighboring cells.

[0035] Step 120, calculating the abnormal index correlation between the abnormal index cell and neighboring cells in different ranges, and generating abnormal index features according to the abnormal index correlations.

[0036] In the embodiment of the present invention, when there are hidden dangers or failures in the passive components of the indoor cell, the service indicators of the cell and the neighboring cells often fluctuate and are correlated, and this indicator correlation is often related to the distance between the cell with abnormal indicators and the neighboring cell. The embodiment of the present invention fully takes into account that neighboring cells at different distances are affected to different degrees by the abnormal cell, and constructs abnormal indicator features starting from the correlation between abnormal indicators of the cell with abnormal indicators and neighboring cells within different ranges, solving the problem that conventional modeling does not consider the correlation between the degree of influence of neighboring cells and the distance and the uneven distribution of neighboring cells along each component. In the embodiment of the present invention, the neighboring cell is an indoor cell that is relatively close to the cell with abnormal indicators.

[0037] Step 130, input the abnormal information, the network parameter adjustment characteristics, the alarm characteristics and the abnormal index characteristics into a preset abnormal prediction model to predict whether there is an abnormal indoor passive device in the abnormal index cell; wherein the abnormal prediction model is trained based on the abnormal index cell sample.

[0038] In an embodiment of the present invention, an abnormality prediction model is pre-trained, and network parameter adjustment characteristics, alarm characteristics and abnormal indicator characteristics are input into the preset abnormality prediction model for processing, so as to predict whether there will be abnormal indoor passive devices in the indicator abnormal cell in the future. The abnormal indoor passive devices may be a failure of the indoor passive devices, or there may be hidden dangers in the indoor passive devices.

[0039] In an exemplary embodiment of the present invention, before obtaining the abnormal information of the abnormal cell, the network parameter adjustment characteristics, and the alarm characteristics, the method further includes the following steps, which are described in detail as follows.

[0040] Obtain performance data of multiple indoor cells; Processing the performance data by using the interquartile range method, and obtaining target performance data of each of the indoor cells based on the processing results; An index threshold is calculated based on the target performance data, and the index abnormal cell is determined from the plurality of indoor cells based on the index threshold and the target performance data.

[0041] In the embodiment of the present invention, the performance data of the indoor cell with daily granularity in quarters is obtained, and the performance data includes indicators such as the average number of RRC (Radio Resource Control) connections, the average number of RRC effective connections, and the uplink and downlink traffic.

[0042] Specifically, the performance data of the indoor sub-district in the past month was extracted, and the interquartile range method was used to identify abnormal indicators for each indicator in the performance data. The box plot corresponding to the interquartile range method is shown in the figure below. Figure 2As shown. Calculate the upper quartile (Q3) and lower quartile (Q1) of a certain indicator, and calculate the interquartile range (IQR) based on the difference between Q3 and Q1. Remove the outliers of each indicator based on Q3, Q1 and IQR. The calculation method of the outlier is to take the data greater than Q3+1.5IQR or less than Q1-1.5IQR as the outlier, and take the data after the outlier is removed as the target performance data. The subsequent calculation will use the preprocessed target performance data.

[0043] The mean μ and standard deviation σ of the target performance data of various indicators are calculated, and then the indicator threshold Y of the indicator is obtained through the formula Y=μ-1.5σ.

[0044] Mark the target performance data that is less than the indicator threshold Y and regard it as abnormal fluctuation indicator data. Alert the indoor cells that have abnormal fluctuation indicators for more than 3 consecutive days and regard them as abnormal indicator cells. Record the basic information of the abnormal indicator cells, such as: cell number, city, district, county, cell name, equipment manufacturer type, coverage scenario, etc. And based on the box plot and triple standard deviation detection method, record the number of days with abnormal indicators and the start time of the abnormality. Perform indicator abnormality detection on the number of RRC connections, the number of effective RRC connections, and the uplink and downlink traffic data respectively, and merge the results of the three indicator abnormality detections.

[0045] In an exemplary embodiment of the present invention, the calculation of the abnormal index correlation between the abnormal index cell and neighboring cells in different ranges, and the generation of abnormal index features according to each abnormal index correlation, includes the following steps, which are described in detail as follows.

[0046] Based on the location information of the cell with abnormal indicators and the location information of the neighboring cells, distance information between the cell with abnormal indicators and each of the neighboring cells is calculated by dynamic weighting; Determine a plurality of neighboring cell sets based on the distance information, and calculate the correlation between the abnormal indicator cell and the abnormal indicator of each of the neighboring cell sets; The abnormal indicator feature is generated according to the correlation of each abnormal indicator.

[0047] In the embodiment of the present invention, after determining the cell with abnormal indicators, full consideration is given to the different influences of the neighboring cells at different distances on the cell with abnormal indicators, and the correlation of abnormal indicators is calculated based on this. In addition, the distribution of neighboring cells of a cell has a certain regularity in terms of geographical location. For example, most buildings are distributed in a rectangular shape, which means that the neighboring cells of a certain indoor cell may be evenly distributed along the direction of the building (east-west or north-south). Therefore, a dynamic weighting method is adopted in feature design to eliminate the distribution differences caused by the features in the geographical location arrangement.

[0048] Specifically, according to the location information in the engineering parameter data of the abnormal indicator cell, such as longitude and latitude information, the distance information between the abnormal indicator cell and the adjacent cell is calculated using the Euclidean metric (also known as the Euclidean distance). The calculation formula of the Euclidean metric is as follows: ; in, Indicates the latitude and longitude of the area with abnormal indicators. Indicates the longitude and latitude of the neighboring cell for which the Euclidean distance is to be calculated. is the Euclidean distance.

[0049] Ordinary Euclidean distance treats the differences between different attributes of two cells equally, that is, only the straight-line distance between the cell with abnormal indicators and the adjacent cell is considered, and the geographical distribution characteristics of nearby neighboring cells are not considered. When the distribution metrics of neighboring cells in various directions are not uniform, that is, the density is different, the traditional Euclidean distance calculation cannot meet the feature construction requirements. Therefore, the embodiment of the present invention further proposes a Euclidean distance based on dynamic weighting.

[0050] In the embodiment of the present invention, firstly, all the neighboring cells of the cell with abnormal indicators are collected to form a neighboring cell sample set Z. The mean of each neighboring cell sample set is calculated as , the standard deviation is s, then the new neighboring cell sample set is obtained by dynamically weighting the abnormal cell with the neighboring cells in the neighboring cell sample set for: .

[0051] Abnormal indicators With a neighboring community Dynamically weighted Euclidean distance for: ;in, and are the standard deviations of the longitude and latitude in the neighboring cell sample sets.

[0052] Based on the calculated distance information, with the cell with abnormal indicators as the center, obtain the neighboring cells within different ranges of distance information. For example, set the sliding window size to 450m, start at 50m, and slide at an interval of 50m to obtain the neighboring cells within 50m, 100m, 150m, 200m, 250m, 300m, 350m, 400m, 450m, and 500m of the cell with abnormal indicators. The neighboring cells with distance information within the same range are regarded as a neighboring cell set.

[0053] The abnormal index correlation between the abnormal index cell and each adjacent cell set is calculated, and the abnormal index feature is generated according to the abnormal index correlation.

[0054] In an exemplary embodiment of the present invention, determining a plurality of neighboring cell sets based on the distance information and calculating the abnormal index correlation between the abnormal index cell and each of the neighboring cell sets includes the following steps, which are described in detail as follows.

[0055] Based on the distance information, calculating the Pearson correlation coefficient between the indicator abnormal cell and each of the neighboring cell sets; The abnormal indicator correlation is determined according to the Pearson correlation coefficient.

[0056] In the embodiment of the present invention, the Pearson correlation coefficient is used to calculate the abnormality index correlation. The calculation formula of the Pearson correlation coefficient is as follows: in, Indicators for abnormal cells The first average value within the calculation period, Indicators for abnormal cells The first standard deviation within the calculation period, is the second average value of the same indicator of the neighboring cells within the calculation period, It is the second standard deviation of the same indicator of the neighboring cells within the calculation period. is the Pearson correlation coefficient between the two. It is the performance data of a certain indicator in the target performance data, such as the average number of RRC connections, the average number of RRC valid connections, the uplink and downlink traffic, etc.

[0057] The correlation between abnormal indicators of cells with abnormal indicators and neighboring cells is calculated based on the different Euclidean distances based on dynamic weighting. If the correlation coefficient is close to -1, it indicates that there is a strong negative correlation between the two cells. As the indicators of the abnormal cell increase or decrease, the neighboring cells show the opposite indicator fluctuations, indicating that the resident users of the cell may flow to the neighboring cell.

[0058] Taking the abnormal indicator cell as the center, the Euclidean distance based on dynamic weighting is used to calculate the mean of the abnormal indicator correlation between the abnormal indicator cell and the neighboring cells within 50m, 100m, 150m, 200m, 250m, 300m, 350m, 400m, 450m, and 500m, respectively, to obtain the abnormal indicator characteristics. The abnormal indicator characteristics are shown in the following table 1: Table 1 The embodiment of the present invention is based on the abnormal index characteristics of the dynamic weighted Euclidean distance, fully considering that the abnormal index cell has different degrees of influence on the business fluctuations of neighboring cells at different distances, and the distribution density of neighboring cells in different directional components is different, solving the problem that conventional modeling does not consider that the degree of influence of neighboring cells is related to the distance and the neighboring cells are unevenly distributed along each component. It provides reliable data support for subsequent AI models to capture and analyze features, and improves the accuracy of abnormal prediction of indoor passive devices.

[0059] In an exemplary embodiment of the present invention, before obtaining the abnormal information of the abnormal cell, the network parameter adjustment characteristics, and the alarm characteristics, the method further includes the following steps, which are described in detail as follows.

[0060] Obtaining first alarm information of the cell with abnormal indicators, second alarm information of a neighboring cell co-located with the cell with abnormal indicators, and third alarm information of a neighboring cell not co-located with the cell with abnormal indicators; Calculate the number of alarms at the multi-dimensional alarm level based on the first alarm information, the second alarm information, and the third alarm information; wherein the multi-dimensional alarm level includes RRU out-of-service alarm, transmission alarm, base station out-of-service alarm, cell out-of-service alarm, performance alarm, master device alarm, and standing wave alarm; The alarm feature is obtained based on the alarm number.

[0061] In an embodiment of the present invention, the original alarm information in the first alarm information of the cell with abnormal indicators is preprocessed. The original alarm information contains a large amount of redundant information, such as abnormal alarm information and the same alarm information repeated many times. First, according to the alarm occurrence time, alarm name, network element name, base station name, and alarm end time, the original alarm information is deduplicated, and the alarm information with completely consistent fields is eliminated. Secondly, some alarm information is generated by human factors, such as network parameter adjustment, business changes, etc., and the alarm information that does not affect the network quality, the alarm information generated by the base station automatically entering the power saving mode at night, the alarm information generated by the base station during the carrier scheduling period, and the alarm information generated by the artificial test adjustment of the base station are eliminated.

[0062] The first alarm information also includes wireless fault alarm information and dynamic ring alarm information. Wireless fault alarm information includes equipment type, alarm title, alarm occurrence time, equipment manufacturer name, equipment room information, location information, alarm elimination time, and alarm project status. Dynamic ring alarm information includes equipment type, alarm title, alarm occurrence time, equipment manufacturer name, network element name, location information, alarm elimination time, alarm project status, and equipment room.

[0063] Taking the computer room as the aggregation point, the wireless fault alarm information and the dynamic environment alarm information are associated based on the same computer room, and the sliding window is aggregated according to the start time and end time of the indicator abnormal period, and the aggregated alarm data is sorted in chronological order. The associated wireless fault alarm information and dynamic environment alarm information are shown in the following table 2.

[0064] Table 2

[0065] After associating the dynamic ring alarm information with the wireless alarm information, considering that when there are hidden dangers or failures in the indoor passive devices, the impact on network performance may interfere with the neighboring cells, the embodiment of the present invention aggregates the alarm information of the neighboring cells when constructing the alarm features of the cell with abnormal indicators.

[0066] In the embodiment of the present invention, the cell with abnormal indicators is taken as the core cell, and the alarm information occurring in the core cell and the neighboring cells (co-sited neighboring cells and non-co-sited neighboring cells) is traversed periodically based on the time period of abnormal indicators, and an alarm feature is constructed based on this, and these alarm information are recorded and encoded.

[0067] The indicator abnormal cell is taken as the core cell. The core cell sets the prefix cc (core cell) before the alarm code, and the alarm information occurring in the indicator abnormal period is recorded as cc-alarm name1 and cc-alarm name2; the neighboring cells with co-site are set with the prefix co (co-site neighbor cell) before the alarm code, and the alarms occurring in the indicator abnormal period are recorded as co-alarm name3 and co-alarm name4; the neighboring cells with non-co-site are set with the prefix nco (non co-site neighbor cell) before the alarm code, and the alarms occurring in the indicator abnormal period are recorded as nco-alarm name5 and nco-alarm name6. The same alarm occurring in the same period of the neighboring cells of the same type is recorded only once.

[0068] Taking into account that different types of alarms have different effects on the abnormal judgment of indoor passive devices, an embodiment of the present invention proposes a feature construction scheme based on dynamic aggregation of alarm data of neighboring cells, which considers whether an alarm occurs, the total number of alarms, the number of remote radio unit (RRU) decommissioning alarms, the number of transmission alarms, the number of base station decommissioning alarms, the number of cell decommissioning alarms, the number of performance alarms, the number of main equipment alarms, the number of standing wave alarms, etc.

[0069] In an exemplary embodiment of the present invention, the calculation of the number of alarms at the multi-dimensional alarm level based on the first alarm information, the second alarm information and the third alarm information includes the following steps, which are described in detail as follows.

[0070] At each alarm level, obtain first alarm weights corresponding to the first alarm information, the second alarm information, and the third alarm information, respectively, and calculate corresponding second alarm weights based on the first alarm information, the second alarm information, and the third alarm information, respectively; Based on the first alarm weight, the second alarm weight, the first alarm information, the second alarm information and the third alarm information, the number of alarms at the corresponding alarm level is weightedly calculated.

[0071] In the embodiment of the present invention, when constructing the alarm feature, the first alarm information, the second alarm information and the third alarm information have different proportions when counting the number of times. The specific calculation method is as follows: in, The number of times the core cell alarm cc-alarm name is reported. The number of co-alarmname alarms for the co-site neighboring cell. The number of times the neighboring cell that does not share a site is alarmed is nco-alarm name. Dynamically weight the number of three types of alarms and aggregate them according to the percentage set by experience, that is, the first alarm weight. The second alarm weight is calculated based on the first alarm information, the second alarm information and the third alarm information, and then the weighted calculated alarm characteristics are used. , , Even in the case of uneven distribution, dynamic weights can be used to highlight the key points. The constructed alarm features are shown in Table 3 below: Table 3

[0072] The embodiment of the present invention solves the problem of insufficient utilization of alarm information of indoor cells and neighboring cells in conventional modeling by dynamically aggregating alarm information.

[0073] In an exemplary embodiment of the present invention, the construction process of the abnormality prediction model includes the following steps, which are described in detail as follows.

[0074] Acquire training data of multiple samples of abnormal indicator cells; wherein the training data includes training samples and sample labels; The abnormal cell samples of each indicator are input into the basic abnormal prediction model for iterative training until the training is completed to obtain the abnormal prediction model; wherein the basic abnormal prediction model is obtained based on a combined long short-term memory network and a Transformer model.

[0075] In an embodiment of the present invention, the indoor cell is used as a grouping unit, and the cell indicator abnormal time period is used as a period to generate an indicator abnormal cell sample. The basic information of the indicator abnormal cell sample mainly includes: cell number, city, cell name, equipment manufacturer type, coverage scenario, indicator abnormal days, abnormal start time, RRC connection number abnormal days, RRC valid connection number abnormal days, traffic abnormal days, detection date and other information. Extract the abnormal information related to the indoor passive device abnormality from the basic information. At the same time, the alarm data for each neighboring area mainly includes: co-station neighboring area alarm data and non-co-station neighboring area alarm data. Generate corresponding alarm features and network parameter adjustment features for the indicator abnormal cell sample. The abnormal information, network parameter adjustment features, and alarm features of each indicator abnormal cell sample are used as its corresponding sample features.

[0076] According to the period of n days to be predicted, a sample label is added to each sample of cell with abnormal indicators. Taking the abnormal indicator period as a sliding window, according to the feedback information from the operation and maintenance personnel, the reasons for the abnormal indicators of each sample of cell with abnormal indicators in the next n days are counted. The corresponding sample labels caused by abnormal passive components of the indoor distribution system are marked as category a, the corresponding sample labels caused by normal personnel flow are marked as category b, the corresponding sample labels caused by network parameter adjustment are marked as category c, the corresponding sample labels caused by active component alarms are marked as category d, and the corresponding sample labels caused by other reasons are marked as category e.

[0077] The embodiment of the present invention makes full use of the correlation between indoor cells and neighboring cells, and takes into account that the service fluctuations of neighboring cells may be related to the distance and that the distribution of neighboring cells along different components is different. Sample features are constructed from multiple angles such as abnormal indicator correlation, dynamic alarm aggregation, and network parameter adjustment records, which provides a solid data foundation for subsequent prediction of passive device abnormalities in indoor cells and greatly improves the accuracy of prediction.

[0078] When constructing sample features, the embodiment of the present invention integrates multiple influencing factors, such as abnormal information, alarm features after dynamic aggregation, network parameter adjustment features, abnormal index features based on dynamic weighted Euclidean distance, etc. These features are not isolated and unrelated, so it is necessary to combine multiple features for comprehensive analysis. Therefore, when constructing the abnormal prediction model, the improved recurrent neural network (RNN), that is, the long short-term memory network (LSTM), is selected to capture multiple features at the same time. At the same time, considering that the influence factor of each feature on the abnormality of the room-divided passive device is different, the embodiment of the present invention selects the attention mechanism to amplify the features with large influence factors, and amplifies the influence factors that affect the abnormal prediction. Therefore, the embodiment of the present invention not only takes the overall situation into consideration, comprehensively considers the influencing factors of most of the features that may characterize the abnormality of the room-divided passive device, but also highlights the key features, and locally amplifies the features with large influence factors. Therefore, the accuracy and efficiency of the embodiment of the present invention in predicting the abnormality of the room-divided passive device are greatly improved.

[0079] A recurrent neural network is a type of recurrent neural network that takes sequence data as input, recursively in the direction of sequence evolution, and all neurons are connected in a chain. Its main function is to process and predict sequence data, and to characterize the relationship between the current output of a sequence and previous information. A recurrent neural network has memory and has certain advantages in learning the nonlinear characteristics of a sequence. Therefore, the embodiment of the present invention selects a recurrent neural network to receive input features, and can perform correlation analysis on various factors that affect the abnormality of indoor passive devices, and store memories.

[0080] The embodiment of the present invention selects the long short-term memory network (LSTM) as the neural network closest to the input layer. It improves the long-term dependency problem of the conventional RNN. Its neuron structure is as follows: Figure 3 As shown. Among them, It represents the output of the previous neuron after capturing and processing the characteristic factors affecting the passive components of the room. Represents the output of its memory unit, that is, the feature information stored at the previous moment; It represents the output of this neuron after receiving and processing the feature vector. It represents the storage content of this neuron memory unit; It is the input of the network, namely the characteristic vector of factors affecting hidden dangers of indoor passive devices.

[0081] LSTM consists of three gate structures, among which the forget gate is responsible for controlling the amount of information stored in the memory unit that affects the prediction of abnormalities of indoor passive devices. It forgets information that is irrelevant to the prediction of abnormalities of indoor passive devices and retains information that affects the prediction of abnormalities of indoor passive devices. Its output is: ; in, is the weight matrix of the forget gate, is the bias term, is the sigmoid function.

[0082] The input gate determines that among the large number of influencing factors obtained, only the characteristic information related to the abnormal prediction of the indoor passive device is stored in the memory unit, while irrelevant information is ignored. The output is as follows: ; in, is the weight matrix of the input gate, is the bias term.

[0083] The memory unit is responsible for storing the key influencing factors that can determine whether there is an abnormal passive device in the indoor distribution cell in the previously received information and memorizing it for a long time until the prediction is completed. Its output is: ; in, is the weight matrix of the memory unit, is the bias term.

[0084] The output gate is responsible for controlling the output content and only outputs information that is helpful for subsequent prediction of abnormalities of passive components in the room. The output is as follows: ; in, is the weight matrix of the output gate, is the bias term.

[0085] LSTM adds a long-term memory module to the traditional RNN short-term memory model, which is conducive to learning and capturing the abnormalities of indoor passive devices and the correlation between various feature parameters. It associates and stores and analyzes information such as abnormal information, network parameter adjustment features, alarm features, and abnormal indicator features, so as to more comprehensively and accurately combine these feature information with abnormal predictions of indoor passive devices. LSTM further processes the information obtained in the early feature construction and sample generation stages to avoid ignoring the correlation between these information. This paves the way for subsequent abnormal prediction.

[0086] The Attention Model (AM) is a technology used in neural networks to solve sequence processing problems. It allows the model to focus on key parts when processing long sequences. The embodiments of the present invention use it to focus on features that have a greater impact on the abnormality of indoor passive devices, which helps to improve the overall accuracy and efficiency of the abnormality prediction model for indoor passive devices. The Transformer model is a new type of neural network architecture that uses an attention mechanism to process sequence data. Its core components are the self-attention mechanism and the cross-sequence attention mechanism. These two attention mechanisms allow the abnormality prediction model to better capture the long-distance dependencies between various influencing factors during feature processing, thereby improving the accuracy of abnormality prediction.

[0087] The attention mechanism is the basis of the Transformer model. It can dynamically adjust the weights for different features in the input sequence, such as abnormal information, network parameter adjustment features, alarm features, and abnormal indicator features, so that the model can focus on elements related to the abnormal prediction of indoor passive devices, thereby enabling the model to automatically focus on different positions in the feature vector.

[0088] The schematic diagram of the implementation process of the self-attention mechanism and the cross-sequence attention mechanism is as follows Figure 4 As shown. The core of the self-attention mechanism is to calculate the attention score assigned to each influencing factor and then multiply these scores by the representation of its previous and next relationships. This process can be expressed by the following formula: ; Among them, Q, K, and V represent query vector, key vector, and value vector respectively. is the dimension of the key vector.

[0089] The core of the cross-sequence attention mechanism is to calculate the attention score assigned to each feature vector, which allows the anomaly prediction model to pay attention to different types of feature vectors. This process can be expressed by the following formula: ; Where h is the number of attention heads, is the output weight matrix. Each attention head is calculated as follows: ; Transformer consists of an encoder and decoder structure (Encoder-Decode), and both the encoder and decoder are superimposed based on the attention mechanism module. The encoder is composed of n identical layers, each layer has two sub-layers, the first is multi-head self-attention, and the second sub-layer is a position-based feed-forward network; the decoder is also composed of n identical layers, and compared with the two sub-layers of the encoder, the decoder has a third sub-layer encoder-decoder attention layer (encoder-decoderattention).

[0090] The architecture of the Transformer model is as follows Figure 5 As shown. The embodiment of the present invention uses the Transformer model as the network for feature analysis, and its attention mechanism can effectively allocate more attention to the key factors that affect the abnormality of the indoor passive device, and allocate less attention to the features with small influencing factors. Therefore, the abnormality prediction model constructed by the embodiment of the present invention focuses on the key feature vectors, providing a better basis for the subsequent prediction of abnormalities of indoor passive devices.

[0091] The basic anomaly prediction model constructed is as follows Figure 6 As shown in the figure, the input layer is a fully connected neural network, which is responsible for receiving feature vectors generated based on abnormal information, network parameter adjustment features, alarm features and abnormal indicator features; the LSTM network is responsible for associating and extracting the features of the input layer; the Transformer allocates the attention of the overall model to the features that have a great impact on the prediction of abnormalities of indoor passive devices, thereby ignoring the features with less impact, and improving the accuracy of the indoor passive device abnormality prediction network; the Softmax layer is responsible for outputting the prediction results of indoor passive device abnormalities.

[0092] The constructed sample of abnormal indicator cells is input into the basic anomaly prediction model for iterative training. The weight of the model is calculated and adjusted according to its accuracy on the validation set until the training is completed to obtain the anomaly prediction model. The trained anomaly prediction model can be used to identify whether the abnormal indicator cell has abnormal indoor passive devices in the future, and output a list of cells with abnormal indoor passive devices.

[0093] The embodiment of the present invention first performs anomaly detection on the service indicators of the indoor cell, detects the indicator abnormal cell with abnormal fluctuation, and makes a preliminary positioning for the subsequent prediction of abnormal passive devices in the indoor cell. Secondly, the features that may characterize the abnormality of the indoor passive device are constructed based on the dynamic weighted Euclidean distance, with the indicator abnormal cell as the core, and the correlation between its abnormal index and the neighboring cell is calculated, and the dynamic alarm aggregation is performed on the co-station and non-co-station neighboring cells, and the network parameter adjustment record is added to the feature vector, and finally the sample feature is generated. Finally, a basic abnormality prediction model based on Transformer-LSTM is designed, which can not only analyze the correlation of the above features, but also focus on the features with large influencing factors for key analysis, which greatly improves the accuracy and efficiency of the prediction of abnormal passive devices in the indoor cell. The technical solution provided by the present invention combines the correlation of neighboring cells, is more in line with the actual situation and more scientific, and can not only identify whether there is an abnormal indoor passive device in the indicator abnormal cell in the future, so as to intervene in advance, but also use the attention mechanism to capture the features with the largest influencing factors, thereby greatly improving the accuracy and efficiency of the prediction of abnormal passive devices in the indoor cell.

[0094] In the technical solution provided by the present invention, based on the attention mechanism, feature construction is performed around performance management data (PM), cell parameter modification record data, indoor cell alarm data, cell engineering parameter data, etc., and an abnormal prediction model for indoor passive devices is built by constructing and learning samples of abnormal indicator cells. Through automatic data collection and sample recycling, the accuracy of the model is continuously optimized by learning and iterating big data samples, and the abnormal prediction of indoor passive devices is realized, which can effectively improve the efficiency of troubleshooting problems such as weak coverage of indoor cells, improve the accuracy of pre-anomaly prediction of indoor passive devices, and further improve the work efficiency of network optimization and maintenance. Through the technical solution provided by the present invention, there is no need to perform a large number of manual tests on site, and the abnormality of indoor passive devices can be accurately located before the fault occurs, which greatly promotes the improvement of work efficiency and effectively reduces the difficulty of work. By predicting the abnormality of indoor passive devices in advance, the present invention can guide the allocation and scheduling of operation and maintenance resources, realize the abnormality troubleshooting of high-risk indoor sites, and improve the efficiency and quality of network operation and maintenance.

[0095] The abnormality prediction device for room-divided passive devices provided by the present invention is described below. The abnormality prediction device for room-divided passive devices described below and the abnormality prediction method for room-divided passive devices described above can be referred to each other. It should be noted that the device provided in the following embodiment and the method provided in the above embodiment belong to the same concept, and the specific manner in which each module and unit performs the operation has been described in detail in the method embodiment, which will not be repeated here.

[0096] In an exemplary embodiment of the present invention, see Figure 7 , Figure 7 According to an exemplary embodiment, a device for predicting abnormality of a passive device in a room is shown, comprising: The first acquisition module 710 is configured to obtain abnormal information of the cell with abnormal indicators, network parameter adjustment characteristics, and alarm characteristics; The first calculation module 720 is configured to calculate the abnormal index correlation between the abnormal index cell and neighboring cells in different ranges, and generate abnormal index features according to the abnormal index correlations; The prediction module 730 is configured to input the abnormal information, the network parameter adjustment characteristics, the alarm characteristics and the abnormal index characteristics into a preset abnormal prediction model to predict whether there is an abnormal indoor passive device in the abnormal index cell; wherein the abnormal prediction model is trained based on the abnormal index cell sample.

[0097] In an exemplary embodiment of the present invention, the first calculation module 720 includes: A first calculation submodule is configured to calculate the distance information between the indicator abnormal cell and each of the neighboring cells by dynamic weighting based on the location information of the indicator abnormal cell and the location information of the neighboring cells; A second calculation submodule is configured to determine a plurality of neighboring cell sets based on the distance information, and calculate the correlation between the abnormal indicator cell and the abnormal indicator of each of the neighboring cell sets; The generating submodule is configured to generate the abnormal indicator feature according to the correlation of each abnormal indicator.

[0098] In an exemplary embodiment of the present invention, the second computing submodule includes: A calculation unit, configured to calculate the Pearson correlation coefficient between the indicator abnormal cell and each of the neighboring cell sets based on the distance information; A determination unit is configured to determine the abnormal indicator correlation according to the Pearson correlation coefficient.

[0099] In an exemplary embodiment of the present invention, the process of constructing the abnormality prediction model includes: A second acquisition module is configured to acquire training data of a plurality of abnormal indicator cell samples; wherein the training data includes training samples and sample labels; The training module is configured to input each of the abnormal indicator cell samples into the basic abnormal prediction model for iterative training until the training is completed to obtain the abnormal prediction model; wherein the basic abnormal prediction model is obtained based on a joint long short-term memory network and a Transformer model.

[0100] In an exemplary embodiment of the present invention, the abnormality prediction device for indoor passive devices further includes: A third acquisition module is configured to acquire first alarm information of the abnormal indicator cell, second alarm information of a neighboring cell co-located with the abnormal indicator cell, and third alarm information of a neighboring cell not co-located with the abnormal indicator cell; A second calculation module is configured to calculate the number of alarms at a multi-dimensional alarm level based on the first alarm information, the second alarm information, and the third alarm information; wherein the multi-dimensional alarm level includes a radio remote unit out-of-service alarm, a transmission alarm, a base station out-of-service alarm, a cell out-of-service alarm, a performance alarm, a master device alarm, and a standing wave alarm; The alarm characteristic module is configured to obtain the alarm characteristic based on the alarm number.

[0101] In an exemplary embodiment of the present invention, the second calculation module includes: an acquisition submodule, configured to acquire, at each alarm level, first alarm weights corresponding to the first alarm information, the second alarm information, and the third alarm information, respectively, and calculate corresponding second alarm weights based on the first alarm information, the second alarm information, and the third alarm information, respectively; The weighted calculation submodule is configured to weightedly calculate the number of alarms at the corresponding alarm level based on the first alarm weight, the second alarm weight, the first alarm information, the second alarm information and the third alarm information.

[0102] In an exemplary embodiment of the present invention, the abnormality prediction device for indoor passive devices further includes: A fourth acquisition module is configured to acquire performance data of a plurality of indoor cells; A processing module, configured to process the performance data by using an interquartile range method, and obtain target performance data of each of the indoor cells based on the processing result; The third calculation module is configured to calculate an indicator threshold based on the target performance data, and determine the indicator abnormal cell from the multiple indoor cells based on the indicator threshold and the target performance data.

[0103] Figure 8 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 8As 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 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the abnormal prediction method of the indoor passive device, the method comprising: obtaining abnormal information of the indicator abnormal cell, the network parameter adjustment characteristics and the alarm characteristics; Calculating the abnormal index correlation between the abnormal index cell and neighboring cells in different ranges, and generating abnormal index features according to the abnormal index correlations; The abnormal information, the network parameter adjustment characteristics, the alarm characteristics and the abnormal index characteristics are input into a preset abnormal prediction model to predict whether there is an abnormal indoor passive device in the abnormal index cell; wherein the abnormal prediction model is trained based on the abnormal index cell samples.

[0104] 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 this understanding, the technical solution of the present invention is essentially 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 to enable 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.

[0105] On the other hand, the present invention also provides a computer program product, the computer program product includes a computer program, the computer program can be stored on a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the abnormality prediction method of the indoor passive device provided by the above methods, the method includes: obtaining abnormal information of the indicator abnormal cell, network parameter adjustment characteristics, and alarm characteristics; Calculating the abnormal index correlation between the abnormal index cell and neighboring cells in different ranges, and generating abnormal index features according to the abnormal index correlations; The abnormal information, the network parameter adjustment characteristics, the alarm characteristics and the abnormal index characteristics are input into a preset abnormal prediction model to predict whether there is an abnormal indoor passive device in the abnormal index cell; wherein the abnormal prediction model is trained based on the abnormal index cell samples.

[0106] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the abnormality prediction method of the indoor passive device provided by the above methods, the method comprising: obtaining abnormal information of the indicator abnormal cell, network parameter adjustment characteristics, and alarm characteristics; Calculating the abnormal index correlation between the abnormal index cell and neighboring cells in different ranges, and generating abnormal index features according to the abnormal index correlations; The abnormal information, the network parameter adjustment characteristics, the alarm characteristics and the abnormal index characteristics are input into a preset abnormal prediction model to predict whether there is an abnormal indoor passive device in the abnormal index cell; wherein the abnormal prediction model is trained based on the abnormal index cell samples.

[0107] 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.

[0108] 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, 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.

[0109] 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 predicting abnormality of indoor passive devices, characterized in that: include: Obtain abnormal information of cells with abnormal indicators, network parameter adjustment characteristics, and alarm characteristics; Calculating the abnormal index correlation between the abnormal index cell and neighboring cells in different ranges, and generating abnormal index features according to the abnormal index correlations; The abnormal information, the network parameter adjustment characteristics, the alarm characteristics and the abnormal index characteristics are input into a preset abnormal prediction model to predict whether there is an abnormal indoor passive device in the abnormal index cell; wherein the abnormal prediction model is trained based on the abnormal index cell samples.

2. The abnormality prediction method of indoor passive devices according to claim 1 is characterized in that: The calculating of the abnormal index correlation between the abnormal index cell and neighboring cells within different ranges, and generating abnormal index features according to the abnormal index correlations, includes: Based on the location information of the cell with abnormal indicators and the location information of the neighboring cells, distance information between the cell with abnormal indicators and each of the neighboring cells is calculated by dynamic weighting; Determine a plurality of neighboring cell sets based on the distance information, and calculate the correlation between the abnormal indicator cell and the abnormal indicator of each of the neighboring cell sets; The abnormal indicator feature is generated according to the correlation of each abnormal indicator.

3. The abnormality prediction method of indoor passive devices according to claim 2 is characterized in that: The determining of a plurality of neighboring cell sets based on the distance information, and calculating the correlation between the abnormal indicator cell and each of the neighboring cell sets includes: Based on the distance information, calculating the Pearson correlation coefficient between the indicator abnormal cell and each of the neighboring cell sets; The abnormal indicator correlation is determined according to the Pearson correlation coefficient.

4. The abnormality prediction method of indoor passive devices according to claim 1 is characterized in that: The construction process of the abnormal prediction model includes: Acquire training data of multiple samples of abnormal indicator cells; wherein the training data includes training samples and sample labels; The abnormal cell samples of each indicator are input into the basic abnormal prediction model for iterative training until the training is completed to obtain the abnormal prediction model; wherein the basic abnormal prediction model is obtained based on a combined long short-term memory network and a Transformer model.

5. The abnormality prediction method of indoor passive devices according to claim 1 is characterized in that: Before obtaining abnormal information of the abnormal cell, network parameter adjustment characteristics, and alarm characteristics, the method further includes: Obtaining first alarm information of the cell with abnormal indicators, second alarm information of a neighboring cell co-located with the cell with abnormal indicators, and third alarm information of a neighboring cell not co-located with the cell with abnormal indicators; Calculate the number of alarms at a multi-dimensional alarm level based on the first alarm information, the second alarm information, and the third alarm information; wherein the multi-dimensional alarm level includes a radio remote unit out-of-service alarm, a transmission alarm, a base station out-of-service alarm, a cell out-of-service alarm, a performance alarm, a master device alarm, and a standing wave alarm; The alarm feature is obtained based on the alarm number.

6. The abnormality prediction method of indoor passive devices according to claim 5 is characterized in that: The calculating the number of alarms at the multi-dimensional alarm level based on the first alarm information, the second alarm information, and the third alarm information includes: At each alarm level, obtain first alarm weights corresponding to the first alarm information, the second alarm information, and the third alarm information, respectively, and calculate corresponding second alarm weights based on the first alarm information, the second alarm information, and the third alarm information, respectively; Based on the first alarm weight, the second alarm weight, the first alarm information, the second alarm information and the third alarm information, the number of alarms at the corresponding alarm level is weightedly calculated.

7. The abnormality prediction method of indoor passive device according to any one of claims 1 to 6, characterized in that: Before obtaining abnormal information of the abnormal cell, network parameter adjustment characteristics, and alarm characteristics, the method further includes: Obtain performance data of multiple indoor cells; Processing the performance data by using the interquartile range method, and obtaining target performance data of each of the indoor cells based on the processing results; An index threshold is calculated based on the target performance data, and the index abnormal cell is determined from the plurality of indoor cells based on the index threshold and the target performance data.

8. An abnormality prediction device for indoor passive devices, characterized in that: include: A first acquisition module is configured to acquire abnormal information of cells with abnormal indicators, network parameter adjustment characteristics, and alarm characteristics; A first calculation module is configured to calculate the abnormal index correlation between the abnormal index cell and neighboring cells in different ranges, and generate abnormal index features according to the abnormal index correlation; The prediction module is configured to input the abnormal information, the network parameter adjustment characteristics, the alarm characteristics and the abnormal index characteristics into a preset abnormal prediction model to predict whether there is an abnormal indoor passive device in the abnormal index cell; wherein the abnormal prediction model is trained based on the abnormal index cell sample.

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 program, the abnormality prediction method of the indoor passive device as described in 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 abnormality prediction method of the indoor passive device as described in 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 abnormality prediction method of the indoor passive device as described in any one of claims 1 to 7 is implemented.

Images