Method for detecting sleep cell based on AI modelTechnical Field
The invention relates to the technical field of mobile communication, in particular to a method for detecting a sleep cell based on an AI model.
Background
In a long term evolution (Long Term Evolution, LTE) and NR systems of a communication technology, a cell is a basic unit for providing services, and as the number of users increases, the capacity of a network is continuously increased, and the number of cells is also increased in series, so that the intelligent operation and maintenance of the cell are critical, the cell is often difficult to access to the user of the cell due to radio frequency and software faults in the operation process, no traffic is generated for a long time and no alarm is generated, and the cell becomes a sleeping cell, and the sleeping cell cannot provide services for a long time and is difficult to be found by maintenance personnel through a conventional operation and maintenance method, so that the sleeping cell can be intelligently and rapidly detected to effectively reduce the complexity and cost of the network operation and maintenance.
The existing sleep cell detection mechanism mainly has two ideas, one is to judge through configuring fixed detection parameters or thresholds, the other is to detect whether UE is accessed or not through a certain self-detection mode at the base station side, for example, the base station automatically sends paging, the former detection mechanism has the defects that the detection parameters and the thresholds are fixed inflexibly, different cell traffic conditions are different, even if the same cell presents different traffic characteristics in different time periods, such as tidal effect of working days and holidays, the fixed parameters are used for judging to easily cause missed detection and false detection, and the latter mode needs to configure paging ID of the UE at the network management side, so that the UE ID information of the accessed cell needs to be known in advance before detection, and is difficult to implement in daily operation and maintenance.
Based on the technical problems in the prior art, the invention provides a method for detecting a sleeping cell based on an AI model.
Disclosure of Invention
The invention provides a method for detecting a sleep cell based on an AI model.
In order to achieve the above purpose, the invention adopts the following technical scheme:
a method of detecting a sleeping cell based on an AI model, comprising:
step 1, network management periodically collects key performance indexes KPIs related to cell telephone traffic;
Step 2, constructing a cell telephone traffic model according to the key performance index KPI;
Step 3, detecting the cell traffic according to the cell traffic model, and adding the cell with abnormal cell traffic into a real-time monitoring list for the cell with abnormal cell traffic;
Step 4, monitoring the cell with abnormal cell traffic according to the real-time monitoring list, and judging whether the cell is a sleeping cell according to the cell traffic model;
Step 5, reporting the judging result of the sleeping cell to the network manager;
and 6, updating the telephone traffic model.
Further, in step 2, the key performance indicators KPIs are classified according to the collected time characteristics.
Further, in step 2, the traffic model can reflect the traffic distribution corresponding to the cell time period and the counted KPIs.
Further, in step 3, whether the traffic of the cell is abnormal is represented by the traffic curve, when the fitting degree of the traffic curve is lower than a preset monitoring threshold, the traffic of the cell is considered to be abnormal, and the cell with abnormal traffic is added into the real-time monitoring list.
Further, in step 1, the key performance indicators KPI include RRC connection establishment attempt number, RRC average user number, and hand-in user number.
Further, in step 1, the time period of the key performance indicator KPI is identified during traffic collection, and the holiday time period is marked.
Further, in step 1, when the network manager collects the key performance index KPI, the network manager collects the key performance index KPI according to the granularity of reporting the key performance index KPI by the base station.
In step 1, the key performance index KPI is directly imported from the historical performance data or is acquired online in real time by the network management side.
In step 2, the time slot label of each day of the cell, the RRC connection establishment attempt times of the corresponding time granularity, the average RRC user number, and the number of hand-in users are used as the characteristic inputs, and during model training, the collected data are distinguished into a working day and a holiday, and a traffic curve is respectively constructed according to the working day and the holiday.
Further, in step 2, training is performed through an AI algorithm during model training, traffic curves with different time granularities of the cells are fitted, and traffic peak time periods and corresponding traffic amounts of the cells are output according to the fitted traffic curves.
Further, in step 2, training is performed by a linear regression algorithm during model training.
Further, in step 2, during model training, the data is divided into first data and second data, wherein the first data is used for training, the second data is used for testing the fitting degree of the constructed traffic model, and when the fitting degree of the tested traffic model reaches a preset threshold, the model can be directly used, otherwise, the step 1 is returned to for data collection again.
Further, in step 4, for the cell added into the monitoring list, the network manager monitors the traffic in real time in the peak traffic time according to the traffic peak time predicted by the model, when the traffic in the peak traffic time continuously deviates from the traffic predicted by the model by more than 90%, the cell can be judged to be a sleeping cell, and when the traffic in the monitored cell does not meet the deviation threshold, the monitored cell continues for 3 monitoring periods, and the cell is deleted from the monitoring cell list.
Further, in step 6, the network manager triggers the collection of the key performance index KPI of the cell traffic according to the configured update period, and after the data collection is completed, the network manager triggers the reconstruction of the traffic model according to the collected data again, so as to complete the model update to avoid the error detection caused by the fact that the prediction model is not updated due to the change of the cell traffic.
Compared with the prior art, the invention has the advantages that:
the method for detecting the sleeping cell based on the AI model can accurately predict the characteristics of the traffic curve, the peak time period, the busy hour traffic and the like of the traffic model of the cell in different time periods (workdays and holidays) according to the traffic conditions of different cells and the same cell, and detect the sleeping cell according to the characteristics of the traffic model, thereby solving the problems of false detection and missed detection caused by the fact that the detection parameters are fixed and cannot be adapted to the traffic characteristics of different cells in the prior art and improving the detection accuracy and detection rate of the sleeping cell.
Drawings
Fig. 1 is a flowchart of a method for detecting a sleeping cell based on an AI model in an embodiment of the present invention;
Fig. 2 is a schematic diagram of traffic distribution corresponding to KPIs and time slots in an embodiment of the invention.
Detailed Description
In order that the above-recited objects, features and advantages of the present application can be more clearly understood, a more particular description of the application will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings, it being understood that embodiments of the application and features of the embodiments may be combined with each other without departing from the scope of the appended claims.
Examples
A method of detecting a sleeping cell based on an AI model, comprising:
Step 1, periodically collecting key performance index KPI related to cell telephone traffic by a network manager, wherein the collecting duration is 4 weeks;
Step 2, constructing a cell telephone traffic model according to the key performance index KPI;
Step 3, detecting the cell traffic according to the cell traffic model, and adding the cell with abnormal cell traffic into a real-time monitoring list for the cell with abnormal cell traffic;
Step 4, monitoring the cell with abnormal cell traffic according to the real-time monitoring list, and judging whether the cell is a sleeping cell according to the cell traffic model;
Step 5, reporting the judging result of the sleeping cell to the network manager;
and 6, updating the telephone traffic model.
In the above embodiments, the periodicity can be characterized as a quarterly or half year update period.
In step 3, whether the cell traffic is abnormal or not is represented by the traffic curve, if the fitting degree of the traffic curve is lower than a preset monitoring threshold, the cell traffic is considered to be abnormal, and the cell traffic is added into a real-time monitoring list, wherein the monitoring threshold is 0.5.
In step 1, the key performance indicators KPI include RRC connection establishment attempt number, RRC average user number, and cut-in user number.
In the step 1, the time period of the key performance index KPI is marked during telephone traffic collection, and the holiday time period is marked.
In step 1, when the network manager collects key performance index KPIs, the network manager collects the key performance index KPIs according to granularity of reporting the key performance index KPIs by the base station.
In the embodiment, the granularity of reporting the key performance indicator KPI by the base station is 5min.
In step 2 of the above embodiment, different traffic models of holidays, workdays, and the like are classified and constructed to predict traffic model characteristics for different times.
In step 2 of the above embodiment, the traffic model uses the KPI collected in step 1 to perform data classification to distinguish holidays and working days as input of AI algorithm, and uses algorithm to fit traffic models of holidays and working days, for example, the AI algorithm is a linear algorithm.
In step 2 of the above embodiment, as shown in fig. 2, the number of RRC connection establishment attempts with an hour granularity in one day is taken as an example, and the abscissa in the figure represents a traffic start period and the ordinate represents the number of RRC connection establishment attempts corresponding to a statistical period.
In the step 1, key performance indexes KPIs are directly imported from historical performance data or are started to be collected on line in real time by a network management side.
The time for constructing the model can be shortened by directly importing the historical performance data.
In step 2, the time period label of each day of the cell, the RRC connection establishment attempt times of corresponding time granularity, the RRC average user number and the cut-in user number are used as characteristic inputs, during model training, the collected data are distinguished into working days and holidays, and a telephone traffic curve is respectively constructed according to the working days and the holidays.
In step 2, training is performed through AI algorithm during model training, traffic curves of different time granularities (such as multiple granularities of 30min, 15min, etc.) of the cell are fitted, and the traffic peak time period and corresponding traffic volume of the cell are output according to the fitted traffic curves.
In step2, training is performed by a linear regression algorithm during model training.
In step 2, during model training, the data is divided into first data and second data, wherein the first data is used for training, the second data is used for testing the fitting degree of the constructed telephone traffic model, the model can be directly used when the tested telephone traffic model fitting degree reaches a preset threshold, and otherwise, the step 1 is returned to for data collection again.
As the dataset included 30 days of data, 24 days (80%) of data were taken for training and the remaining 6 days (20%) were used to test the fit of the constructed traffic model.
In step 4, for the cell added into the monitoring list, the network manager monitors the traffic in real time in the peak traffic time according to the traffic peak time predicted by the model, when the traffic in the peak traffic time continuously deviates greatly from the traffic predicted by the model, the cell can be judged to be a sleeping cell, and when the traffic in the monitored cell does not meet the deviation threshold, the cell is deleted from the monitoring cell list after a plurality of monitoring periods.
In step 6, the network manager triggers the collection of the key performance indexes KPI (RRC connection establishment attempt number, RRC average user number, cut-in user number) of the cell traffic according to the configured update period (e.g. 4 weeks), and after the data collection is completed, the network manager triggers the reconstruction of the traffic model according to the collected data again, thereby completing the model update to avoid the occurrence of error detection caused by the fact that the prediction model is not updated due to the change of the cell traffic.
The present invention is not limited to the above-described embodiments, and the above-described embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made therein without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims.