CN110891283A

Movatterモバイル変換

Info

Publication number: CN110891283A
Application number: CN201911151942.9A
Authority: CN
Inventors: 钟巍
Original assignee: Super Communications Co ltd
Current assignee: Super Communications Co ltd
Priority date: 2019-11-22
Filing date: 2019-11-22
Publication date: 2020-03-17

Abstract

The invention provides a small base station monitoring device and a method based on an edge calculation model, wherein the monitoring device comprises an original data classification module and a machine learning model, the original data classification module classifies data generated by a small base station and sends the data to the machine learning model, the machine learning model analyzes the received classified data by using different models, and an analysis result is reported to a network manager. The invention adopts the machine learning technology, can identify accurate fault results and predict the change of the service volume in a large amount of invalid data according to the model, effectively reduces the load of the network management of the small base station, provides more accurate state monitoring and fault positioning, and can take automatic recovery measures according to established strategies.

Description

Small base station monitoring device and method based on edge calculation model

Technical Field

The invention relates to the technical field of base station monitoring, in particular to a small base station monitoring device and method based on an edge calculation model.

Background

In the 5G era, as the deployment frequency band is higher and higher, the coverage area of a single base station is smaller, so that dense base station coverage is required, and meanwhile, the penetration capability of high-frequency signals is weakened, so that indoor coverage becomes a difficult problem to be solved urgently. The small base station becomes a new choice for enhancing the network coverage capability in the 5G era.

With the release of technologies such as a commercial-oriented 4G/5G O-RAN dual-mode open design clouded small station scheme and the like, a small base station is arranged in each family or becomes a trend in the future, and the network coverage mode is expected to really get through the last kilometer of the 5G landing.

With the joint efforts of the market and operators, it is seen that the number of small base stations will increase greatly in the near future, and this brings about an increase in the difficulty of monitoring and normal operation and maintenance of the small base station equipment.

The deployment environment of the small base station is different from that of the macro station, is not controlled by an operator, and is influenced by unpredictable power, network and user autonomous behaviors.

The number of the small base stations is greatly increased, and the online mode of the small base stations makes the traditional network management mode not suitable any more, the physical and logical relevance of the managed network elements is more and more, a single fault often generates a large amount of alarm information in the related network elements, so that the identification and the positioning of the fault become difficult, and when a plurality of faults are concurrent, the situation becomes more complicated. When facing massive alarm information, an administrator often has difficulty in finding out the true cause of a fault, so that fault repair and fault removal cannot be rapidly implemented.

In such a situation, it is necessary to provide an intelligent monitoring mode, which improves the accuracy of the alarm and the automatic recovery function.

Disclosure of Invention

In view of the above, to solve the problems in the prior art, the present invention provides a small cell monitoring apparatus and method based on an edge calculation model, which can effectively reduce the load of the network manager of the small cell and provide more accurate state monitoring and fault location.

The technical scheme of the invention is as follows:

in a first aspect, the present invention provides a small base station monitoring apparatus based on an edge calculation model, which includes an original data classification module and a machine learning model, wherein the original data classification module classifies data generated by a small base station and sends the classified data to the machine learning model, and the machine learning model analyzes the received classified data by using different models and reports an analysis result to a network manager.

Further, the original data classification module classifies data generated by the small base station into alarm data, a control plane, a user plane and monitoring data.

Further, the machine learning model comprises an automatic alarm association and root cause identification model, a telephone traffic prediction model, a user off-network prediction model and a data caching module;

the data cache module is used for receiving the classified data sent by the original data and forwarding the classified data to the automatic alarm association and root cause identification model, the telephone traffic prediction model or the user off-network prediction model;

automatic alarm association and root cause identification model: receiving alarm data, filtering, merging and root cause judging the alarm data, giving out fault judgment with highest probability from the alarm data, and reporting to a network manager;

traffic prediction model: receiving data of a control plane, presuming whether the traffic has an outbreak phenomenon or not through historical control plane data, and reporting the outbreak phenomenon to a network manager;

and the user off-network prediction model is used for receiving user plane data, judging whether the user is off-line normally from the use condition of the user equipment and reporting the user off-line to the network manager.

In a second aspect, the present invention provides a small cell monitoring method based on an edge calculation model, including the following steps:

step S1, sending the data generated by the small base station to an original data classification module, and classifying the data generated by the small base station according to a data source and a processing flow by the original data classification module;

step S2, the original data classification module sends the classified data to the machine learning model;

and step S3, the machine learning model analyzes the classified data by adopting different models and reports the analysis result to the network manager.

Further, the original data classification module classifies data generated by the small cell into alarm data, a control plane, a user plane and monitoring data.

the data cache module receives the classified data, stores and forwards the classified data to a corresponding automatic alarm association and root cause identification model, a telephone traffic prediction model or a user off-network prediction model;

the automatic alarm correlation and root cause identification model analyzes alarm data, the telephone traffic prediction model analyzes data of a control plane, and the user off-network prediction model analyzes data of a user plane.

Further, the step of generating the automatic alarm correlation and root cause identification model specifically includes:

step S101, collecting historical alarm data;

step S102, processing alarm data by using a TF-IDF algorithm;

s103, performing model training by using an XGboost algorithm to generate an automatic alarm correlation and root cause identification model;

and step S104, deploying the automatic alarm correlation and root cause identification model to the edge computing node.

Further, the step of generating the traffic prediction model specifically includes:

step S201, collecting historical control surface data;

step S202, data of a control surface are arranged;

step S203, carrying out normalization processing on each feature data;

step S204, carrying out model training, and generating a telephone traffic prediction model by using an XGboost algorithm and a Ridge Regression algorithm;

and S205, deploying the traffic prediction model to the edge computing node.

Further, the step of generating the user off-grid prediction model specifically includes:

s301, collecting historical user plane data;

s302, integrating offline time characteristics according to time granularity to form different data sets;

step S303, model training, namely training by using a random forest and an XGboost algorithm on a data set, and performing model fusion by using a Stacking algorithm to obtain a user off-network prediction model;

and S304, deploying the user off-network prediction model to the edge computing node.

Further, the step of analyzing the data is:

after the automatic alarm correlation and root cause identification model receives the alarm data, firstly, the TF-IDF algorithm is used for filtering the alarm information, and the alarm data is forwarded or filtered, aggregated and judged according to the preset strategy and then sent to the network manager.

The invention has the beneficial effects that:

the machine learning technology is adopted at the base station side, so that accurate fault results and changes of predicted business volume can be identified in a large amount of invalid data according to a model, the load of a network manager of the small base station is effectively reduced, more accurate state monitoring and fault positioning are provided, and automatic recovery measures can be taken according to a set strategy.

Drawings

FIG. 1 is a schematic diagram of a service architecture based on edge computing;

FIG. 2 is a schematic structural diagram of a small cell monitoring apparatus based on an edge calculation model according to the present invention;

FIG. 3 illustrates steps for analyzing data generated by a small cell using a machine learning model according to the present invention.

DETAILED DESCRIPTION OF EMBODIMENT (S) OF INVENTION

The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

The embodiments of the present disclosure are described below with specific examples, and other advantages and effects of the present disclosure will be readily apparent to those skilled in the art from the disclosure in the specification. It is to be understood that the described embodiments are merely illustrative of some, and not restrictive, of the embodiments of the disclosure. The disclosure may be embodied or carried out in various other specific embodiments, and various modifications and changes may be made in the details within the description without departing from the spirit of the disclosure. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict. All other embodiments, which can be derived by a person skilled in the art from the embodiments disclosed herein without making any creative effort, shall fall within the protection scope of the present disclosure.

Example one

Fig. 1 is a schematic diagram of a service architecture based on edge computing for data aggregation, forwarding and content provision. SDN/MEC SERVER is a monitoring apparatus of a small cell based on an edge computing model of the present invention, the monitoring apparatus is an edge computing node deployed between a small cell and a core network, and a schematic structural diagram of the monitoring apparatus is shown in fig. 2.

The node uses SDN based switching technology in actual deployment, providing the following functions:

the edge gateway is used for providing data forwarding of a user plane and a control plane for the small base station; gathering and filtering data of related data such as signaling and alarm; other traffic related services such as data caching.

The invention discloses a small base station monitoring device based on an edge calculation model.

The original data classification module classifies data generated by the small base station into alarm data, a control plane, a user plane and monitoring data according to a data source and a processing flow.

The benefits of classifying the raw data are: and transmitting the different types of original data to different models in the machine learning model, and processing and managing the original data in a targeted manner.

The machine learning model comprises an automatic alarm correlation and root cause identification model, a telephone traffic prediction model, a user off-network prediction model and a data caching module.

The data cache module is used for receiving the classification data sent by the original data and forwarding the classification data to the automatic alarm association and root cause identification model, the telephone traffic prediction model or the user off-network prediction model.

The automatic alarm correlation and root cause identification model receives alarm data, filters, combines and judges the root causes of the alarm data, gives out fault judgment with highest probability from a large amount of alarm data, and then sends the fault judgment to a network manager.

The alarm data of the small base station is generally classified into the following types:

and (5) instant interruption warning: the life cycle is short, and the reference value is not too large;

and (3) frequent alarm: when a certain number of specific alarms and specific events occur within a certain time, certain correlation exists between the alarms and the events;

and (3) faults in the same network element: the alarm generated on a physical object (single board, topology) in the same network element can cause other physical objects and logical objects on the network element to generate related alarms;

fault early warning: a fault early warning can be sent out before a fault occurs;

and (4) fault warning: the method can predict and detect the periodic variation indexes and has alarm classification.

The automatic alarm association and root cause identification model needs to filter the alarm data, and combines the alarm data according to proper dimensionality to show general information;

and analyzing the fault root after the alarm data are combined, giving the most possible reason, and assisting people in making decisions.

The automatic alarm association and root cause identification module can also select and execute a proper fault self-healing strategy according to the fault cause, and automatically solve the fault.

The machine learning and deep learning technology is applied to the edge computing node to process the data, a more accurate and more reference result is generated and reported to the network manager, and an operator can visually judge and know the operation condition of the equipment.

Traffic prediction model of cell: and receiving the control plane information, and estimating whether the traffic has an explosion phenomenon or not according to historical control plane information data so as to take necessary measures in advance.

And the user off-network prediction model is used for receiving user plane data and judging whether the equipment is normally off-line or not according to the use condition of the user equipment so as to avoid unnecessary alarm.

Under the condition that the base station is off-line, the edge computing node judges the probability of abnormal off-line of the base station through the model and reports the probability to the network manager.

Example two

The embodiment provides a small base station monitoring method based on an edge calculation model, which comprises the following steps:

the original data classification module classifies data generated by the small base station into alarm data, a control plane, a user plane and monitoring data.

Alarm data, alarm information generated on hardware and software of base station equipment;

the control plane is used for controlling the establishment, maintenance and release of a call flow by signaling transmission. The information includes data that cannot be sensed by the application layer, for example, interaction of a wireless air interface core network, and more detailed information can be acquired from the bottom layer through signaling.

Service data, such as voice data or packet service data.

And monitoring data, wherein the monitoring data is the basis of alarm data and comprises all available data on the base station equipment. Such as operating parameters and status of the system, debugging and alarm information for software, kpi data for services, etc.

the machine learning model comprises an automatic alarm correlation and root cause identification model, a telephone traffic prediction model, a user off-network prediction model and a data cache module;

and the data caching module receives the classified data, stores and forwards the classified data to a corresponding automatic alarm correlation and root cause identification model, a telephone traffic prediction model or a user off-network prediction model.

The automatic alarm correlation and root cause identification model analyzes alarm data, the telephone traffic prediction model analyzes data of a control plane, and the user off-network prediction model analyzes data of a user plane

The step of generating the automatic alarm correlation and root cause identification model specifically comprises the following steps:

step S101, collecting historical alarm data;

the underlying data of the historical alarm data must include fields for time, equipment number, alarm information, fault classification.

Step S102, processing alarm data by using a TF-IDF algorithm;

TF-IDF (term frequency-inverse document frequency) is a commonly used weighting technique for information retrieval and data mining. TF means Term Frequency (Term Frequency), and IDF means Inverse text Frequency index (Inverse Document Frequency).

And S103, performing model training by using an XGboost algorithm to generate an automatic alarm correlation and root cause identification model.

In actual operation, the single algorithm XGBoost has the strongest prediction capability and is generally used as a main algorithm, but in order to obtain a better effect, a plurality of algorithms are fused to obtain a better result, the fusion algorithm is a weighted Voting method, and other related main algorithms include Voting/Voting, Decisiontree/decision tree, random _ forest/random forest, SVM/support vector machine. Firstly, an XGboost/Decision tree/random _ forest/SVM algorithm model is used for simultaneously predicting data needing prediction, then the classification votes of the learner are multiplied by the weight of each algorithm (the XGboost can be set as double weight), and finally the weighted votes of each category are summed to obtain a final result. Whether a single algorithm or a fusion algorithm is used is decided according to the final prediction result.

The step of generating the traffic prediction model specifically includes:

step S201, collecting historical control surface data;

the control plane information comprises call statistical data of the base station equipment and related kpi data;

step S202, data of a control surface are arranged;

counting data characteristics in a specific time period, wherein the data characteristics comprise a maximum value, a minimum value, a mean value, a median value, a variance and a standard deviation;

step S203, carrying out normalization processing on each feature data;

step S204, carrying out model training, and generating a telephone traffic prediction model by using an XGboost algorithm and a Ridge Regression as main algorithms;

and S205, deploying the traffic prediction model to the edge computing node.

The step of generating the user off-grid prediction model specifically comprises the following steps:

s301, collecting historical user plane data;

the user plane data comprises equipment information, offline time, offline duration and cell KPI statistical information.

Step S302, integrating the offline time characteristics on time, by day, or at other time granularity to form different data sets;

step S303, model training, namely training a data set by using a random forest and XGboost, and performing model fusion by using a Stacking algorithm to obtain a user off-network prediction model;

random forest refers to a classifier that trains and predicts a sample using multiple trees. The XGboost (extreme Gradient boosting) full name is the king card of the integrated learning method.

Further, the specific steps for data analysis are as follows:

after the automatic alarm correlation and root cause identification model receives the alarm data, the alarm data is firstly subjected to data cleaning, the TF-IDF algorithm is utilized to filter the alarm information, data caching is carried out, the alarm data is subjected to processing such as forwarding, filtering, aggregation, root cause judgment and the like according to a preset strategy and then is sent to a network manager.

Because of the existence of the edge computing service node of the small base station, all managed base station equipment under the node does not directly interact with a network manager, but performs data forwarding through the node.

And forwarding, namely transparently transmitting some key data without processing, such as key alarms.

And filtering, namely, data which is not required to be sent to a network manager, such as a transient interruption alarm.

And aggregation, namely merging the alarms of the same type to reduce network data traffic, such as frequent alarms.

And analyzing the root cause, namely, the sent alarm data are all original data, although part of the data is filtered, whether the alarm generated at the same time corresponds to the actual fault or not cannot be judged for the network manager, and analyzing the fault by a root cause judging module to judge the fault probability corresponding to the alarm chain generated at the time point, wherein the fault probability belongs to the fault categories of hardware/software/electric power/transmission/dynamic loop/fault alarm and the like.

The running environment of the automatic alarm correlation and root cause identification model is python, and a sklern function library is used for realizing an algorithm. Python is a cross-platform computer programming language, an object-oriented dynamic type language. Scikit-leann (skleann) is a commonly used third party module in machine learning.

The telephone traffic prediction model receives the control plane information, and estimates whether the traffic has an outbreak phenomenon or not according to the data of the telephone traffic prediction model, thereby taking necessary measures in advance. For example, in some areas where traffic causes tidal effect, such as shopping malls and schools, the number of small base stations can be increased appropriately, and the problems are solved through the dense networking of the small base stations and the macro-micro cooperative networking. On the basis, the intelligent antenna is utilized to carry out space division multiplexing, so that the bandwidth utilization rate can be improved, and the wireless resource utilization rate is improved.

And the edge computing node returns the telephone traffic prediction data of the subordinate base station equipment in a specific time period according to the call of the network manager.

And the user off-network prediction model is used for receiving user plane data and predicting whether the equipment is normally off-line from the use condition of the user equipment to avoid unnecessary alarm. Unnecessary alarms comprise transient alarm and partial frequent alarm, and the main characteristic is that the alarms can be cancelled in a very short time after the alarms are generated, and the partial alarm data accounts for a large proportion and has no reference value.

The above description is for the purpose of illustrating embodiments of the invention and is not intended to limit the invention, and it will be apparent to those skilled in the art that any modification, equivalent replacement, or improvement made without departing from the spirit and principle of the invention shall fall within the protection scope of the invention.

Claims

1. A small base station monitoring device based on an edge calculation model is characterized in that: the small base station analysis system comprises an original data classification module and a machine learning model, wherein the original data classification module classifies data generated by the small base station and sends the data to the machine learning model, the machine learning model analyzes the received classified data by using different models, and an analysis result is reported to a network manager.

2. The small cell monitoring apparatus based on edge calculation model according to claim 1, wherein: the original data classification module classifies data generated by the small base station into alarm data, a control plane, a user plane and monitoring data.

3. The small cell monitoring apparatus based on edge calculation model according to claim 1, wherein: the machine learning model comprises an automatic alarm correlation and root cause identification model, a telephone traffic prediction model, a user off-network prediction model and a data cache module;

4. A small base station monitoring method based on an edge calculation model is characterized in that: the method comprises the following steps:

5. The method for monitoring the small cell base station based on the edge calculation model according to claim 4, wherein: the original data classification module classifies data generated by the small base station into alarm data, a control plane, a user plane and monitoring data.

6. The small cell monitoring method based on the edge calculation model according to claim 5, wherein: the machine learning model comprises an automatic alarm correlation and root cause identification model, a telephone traffic prediction model, a user off-network prediction model and a data cache module;

7. The small cell monitoring method based on the edge calculation model according to claim 6, wherein:

step S101, collecting historical alarm data;

step S102, processing alarm data by using a TF-IDF algorithm;

s103, performing model training by using an XGboost algorithm or a fusion algorithm to generate an automatic alarm correlation and root cause identification model;

8. The small cell monitoring method based on the edge calculation model according to claim 6, wherein:

the step of generating the traffic prediction model specifically includes:

step S201, collecting historical control surface data;

step S202, data of a control surface are arranged;

step S203, carrying out normalization processing on each feature data;

and S205, deploying the traffic prediction model to the edge computing node.

9. The small cell monitoring method based on the edge calculation model according to claim 6, wherein:

s301, collecting historical user plane data;

10. The small cell monitoring method based on the edge calculation model according to claim 6, wherein: the method for analyzing the alarm data by the automatic alarm correlation and root cause identification model comprises the following steps: