5G-R multi-monitoring data fusion detection network situation awareness analysis method and system

Abstract

The invention discloses an analysis method and system for 5G-R multi-monitoring-data fusion detection network situation awareness, and the method comprises the steps: carrying out the fusion of multivariate monitoring data through the analysis scheme based on 5G-R multi-monitoring-data fusion detection network situation awareness around the communication process element information (i.e., time, position and event) with the 5G-R service performance as the center; establishing a normalized label system, and dividing the normalized label system into a real-time parallel analysis route and a non-real-time parallel analysis route according to actual requirements in network operation and maintenance; meanwhile, in order to deal with different requirements of real-time and non-real-time processes for realizing a target, a processing flow, a key technology and resources, a series of methods such as label normalization, deviation maintenance, multi-dimensional label association, feature storage and degradation protection are also provided to ensure comprehensiveness, precision and efficiency of analysis. According to the technical scheme, the convenience of programming is considered at the beginning of design, rapid deployment is facilitated, and the data interface, label adaptation and model optimization process during AI introduction are facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of railway 5G-R wireless network communication analysis, and particularly to an analysis method and system for 5G-R multi-monitoring data fusion detection network situation awareness. Background Art

[0002] With the rapid development of high-speed railways, high-speed trains and bullet trains, the mileage of the railway network has increased rapidly. In order to improve the wireless network communication experience of passengers, mobile operators have been constantly striving to optimize the network services along the way. In the future, with the popularization of autonomous driving technology, the requirements for the network will be even higher, especially for the dedicated communication network (5G-R) of high-speed trains and bullet trains, which plays an increasingly important role in train dispatching and control information transmission and is directly related to the safety and reliability of railway operation.

[0003] At present, when evaluating wireless network coverage, quality and events of linear routes such as highways and railways at home and abroad, several statistical indicators are mainly relied on to generally describe the wireless situation of the entire test route. These statistical indicators can provide a macroscopic network overview, but for specific abnormal events and quality defect problems among them, manual analysis is still used for processing. This method has the following problems: the capabilities and work attitudes of analysts are uncertain, lacking objective and quantitative standards, resulting in insufficient analysis accuracy and even possible misjudgments. In addition, the current industry assessment system is also difficult to meet the needs of refined management. Although the industry assessment system provides basic norms, there are still deficiencies in refined management in actual applications.

[0004] In view of the above problems, it is necessary to conduct in-depth research on existing network analysis methods and develop a solution with high accuracy and high efficiency to achieve comprehensive perception and real-time analysis of the 5G-R network situation. This not only helps to improve the quality and stability of network services, but also provides a more powerful guarantee for the safety and efficiency of railway operation. Summary of the Invention

[0005] The purpose of the present invention is to provide an analysis method and system for 5G-R multi-monitoring data fusion detection network situation awareness, which can accurately and efficiently achieve comprehensive perception and real-time analysis of the 5G-R network situation.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] An analysis method for 5G-R multi-monitoring data fusion detection network situation awareness includes:

[0008] Step 1: Obtain multi-source data of the 5G-R communication network and divide it into real-time monitoring data and non-real-time monitoring data;

[0009] Step 2: Store the real-time monitoring data and non-real-time monitoring data as raw data, manage them throughout their life cycles, and use them for AI model training;

[0010] Step 3: Real-time monitoring data processing: Extract the real-time monitoring data in multiple dimensions, classify and label it, and provide it to Step 5 for use. Then, perform data analysis based on the normalized multi-dimensional data labels feedback from Step 5 to obtain the analysis results of the real-time monitoring data;

[0011] Step 4: Non-real-time monitoring data processing: Preprocess the non-real-time monitoring data, provide the corresponding multi-dimensional data to Step 5 for use, perform data segmentation and storage in combination with the normalized multi-dimensional data labels feedback from Step 5, establish associations between different data sources based on the normalized multi-dimensional data labels, divide the data on different time scales, and perform data analysis on the data at each time scale respectively by combining empirical models and AI models to obtain the analysis results of the non-real-time monitoring data;

[0012] Step 5: Build a normalized label system centered on 5G-R services around multi-dimensional data, and transfer labels to Step 3 and Step 4 respectively;

[0013] Step 6: Integrate the analysis results of the real-time monitoring data and the analysis results of the non-real-time monitoring data, and output the analysis results of network situation awareness.

[0014] A 5G-R multi-monitoring data fusion detection network situation awareness analysis system for implementing the foregoing method, including:

[0015] A multi-source data acquisition and division unit for acquiring multi-source data of a 5G-R communication network and dividing it into real-time monitoring data and non-real-time monitoring data;

[0016] A raw data full life cycle management unit for storing the real-time monitoring data and non-real-time monitoring data as raw data, managing them throughout their life cycles, and using them for AI model training;

[0017] A real-time monitoring data processing unit for real-time monitoring data processing, including: Extracting the real-time monitoring data in multiple dimensions, classifying and labeling it, and providing it to Step 5 for use. Then, perform data analysis based on the normalized multi-dimensional data labels feedback from Step 5 to obtain the analysis results of the real-time monitoring data;

[0018] A non-real-time monitoring data processing unit for processing non-real-time monitoring data, including: preprocessing the non-real-time monitoring data and providing the corresponding multi-dimensional data for step 5 to use, and combining the normalized multi-dimensional data labels fed back by step 5 to perform data segmentation and storage, establishing associations between different data sources based on the normalized multi-dimensional data labels, then dividing the data at different time scales, and respectively performing data analysis on the data at each time scale by combining an empirical model and an AI model to obtain the analysis results of the non-real-time monitoring data;

[0019] A label adaptation unit for establishing a normalized label system around multi-dimensional data centered on 5G-R services, and respectively performing label transfer with step 3 and step 4;

[0020] A conclusion report output unit for comprehensively combining the analysis results of the real-time monitoring data and the analysis results of the non-real-time monitoring data, and outputting the analysis results of network situation awareness.

[0021] It can be seen from the technical solutions provided by the present invention above that through an analysis solution for network situation awareness detection by fusing multi-monitoring data of 5G-R, multi-source monitoring data is fused around the communication process element information centered on 5G-R service performance (i.e., the three dimensions of time, location, and event), a normalized label system is established, and according to the actual requirements in network operation and maintenance, it is divided into two parallel analysis routes of real-time and non-real-time; at the same time, in order to cope with the differences in the requirements for achieving goals, processing flows, key technologies, and resource requirements in real-time and non-real-time processes, a series of methods such as label normalization, deviation maintenance, multi-dimensional label association, feature storage, and degradation protection are also provided to ensure the comprehensiveness, accuracy, and efficiency of the analysis. When designing this technical solution, the convenience of programming, convenient and rapid deployment, as well as the data interface, label adaptation, and model optimization process when introducing AI are taken into account. Description of the Drawings

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0023] Figure 1 It is a flowchart of an analysis method for network situation awareness detection by fusing multi-monitoring data of 5G-R provided by an embodiment of the present invention;

[0024] Figure 2 It is a schematic diagram of the overall framework of an analysis method for network situation awareness detection by fusing multi-monitoring data of 5G-R provided by an embodiment of the present invention;

[0025] Figure 3 Schematic diagram of data fusion of the 5G-R communication multi-monitoring system provided by an embodiment of the present invention;

[0026] Figure 4 Schematic diagram of the normalized coordinate relationship of labels of the 5G-R communication multi-monitoring system provided by an embodiment of the present invention;

[0027] Figure 5 Schematic diagram of multi-level situation awareness of the 5G-R communication network provided by an embodiment of the present invention;

[0028] Figure 6 Schematic diagram of three-dimensional guarantee of 5G-R services in numerous scenarios provided by an embodiment of the present invention. Detailed implementation manners

[0029] Next, in combination with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0030] First, the terms that may be used in this article are described as follows:

[0031] The description of terms such as "include", "comprise", "contain", "have" or other similar semantics should be interpreted as non-exclusive inclusion. For example: including a certain technical feature element (such as raw materials, components, ingredients, carriers, dosage forms, materials, dimensions, parts, components, mechanisms, devices, steps, processes, methods, reaction conditions, processing conditions, parameters, algorithms, signals, data, products or articles, etc.) should be interpreted as not only including the clearly listed certain technical feature element, but also including other technical feature elements well known in the art that are not clearly listed.

[0032] The term "consisting of" means excluding any technical feature element that is not clearly listed. If this term is used in a claim, this term will make the claim a closed type, so that it does not include technical feature elements other than the clearly listed technical feature elements, except for the related conventional impurities. If this term only appears in a certain clause of the claim, then it only limits the elements clearly listed in that clause, and the elements recorded in other clauses are not excluded from the overall claim.

[0033] The following provides a detailed description of an analysis method and system for 5G-R multi-monitoring data fusion detection network situation awareness. The content not described in detail in the embodiments of the present invention belongs to the prior art well-known to those skilled in the art. In the embodiments of the present invention, conditions not specified are carried out according to the conventional conditions in the art or the conditions recommended by the manufacturer. Reagents or instruments not specified in the embodiments of the present invention are all conventional products that can be obtained through commercial purchase.

[0034] Embodiment 1

[0035] The embodiments of the present invention provide an analysis method and system for 5G-R multi-monitoring data fusion detection network situation awareness, as Figure 1 shown, which mainly includes the following steps:

[0036] Step 1: Obtain multi-source data of the 5G-R communication network and divide it into real-time monitoring data and non-real-time monitoring data.

[0037] In the embodiments of the present invention, the multi-source data of the 5G-R communication network includes: 5G network northbound alarm data, 5G interface monitoring system data, 5G wireless probe data, 5G transmission network management system data, 5G network dynamic environment monitoring data, antenna and tower monitoring data, 5G-R network northbound performance data, 5G-R road test equipment data, 5G-R network configuration parameters, 5G-R equipment ledger, MDT minimum road test terminal measurement and acquisition system data, CIR system vehicle-mounted equipment monitoring system data, and MCX railway service dedicated processing service data. After obtaining the above multi-source data of the 5G-R communication network, it can be divided into real-time monitoring data and non-real-time monitoring data according to data characteristics.

[0038] Preferably, considering that there will be certain deviations in different measurement schemes. For example, different terminals can be used to measure the signal strength RSRP. Due to different terminal sensitivities, antenna interface attenuations, and antenna gains, deviations will occur, and this deviation amount Δp will change due to equipment aging or failure. Therefore, after obtaining the multi-source data of the 5G-R communication network, the deviation amount of each type of data is corrected. For example, at the same location, there are multiple measurement terminals collecting wireless signals, and they will form comparisons. Since the train control terminals on the railway line are limited, all of them will form comparisons, thereby tracking and correcting the deviations.

[0039] Step 2: Store the real-time monitoring data and non-real-time monitoring data as original data and perform full-life cycle management, as well as use it for AI model training.

[0040] In the embodiments of the present invention, the data in Step 1 is managed in terms of the life cycle, and at the same time, it can also be used for AI model training.

[0041] Step 3, Real-time Monitoring Data Processing: Extract the real-time monitoring data in multiple dimensions, classify and label it, and then provide it for use in Step 5. Then, perform data analysis based on the normalized multi-dimensional data labels fed back by Step 5 to obtain the analysis results of the real-time monitoring data.

[0042] Preferably, the real-time monitoring data processing includes: multi-dimensional data label extraction, anomaly and its weight identification, multi-source data adaptation and association, real-time event analysis, and real-time event experience library maintenance.

[0043] (1) The multi-dimensional data extraction includes: extracting the three dimensions of time, location, and event from various real-time monitoring data respectively.

[0044] (2) The anomaly and its weight identification includes: determining the weight according to the importance of 5G-R communication and the carried services, classifying and labeling accordingly, performing label adaptation in combination with the classification and labeling results, and judging whether there is an abnormal situation.

[0045] In the embodiments of the present invention, the classification and labeling information may include measurement type, signaling type, service type, etc. of the information. Each type of information may be different in time reference, granularity, and accuracy; and may also be different in location reference, granularity, and accuracy. The classification and labeling results can be used for subsequent label adaptation, so as to unify all different classification and labeling information on the label normalization coordinate system (specifically, see the Figure 4 ) for mutual verification of data from different sources.

[0046] Those skilled in the art can understand that the abnormal situation can be divided according to the actual situation. For example, from light to heavy, they are: weak signal level (duration and affected mileage), poor signal quality (duration and affected mileage), high bit error rate, occurrence of rescue handover, handover failure, communication interruption or disconnection, service interruption or disconnection, etc.

[0047] (3) The multi-source data adaptation and association includes: establishing the association between different data sources based on the normalized multi-dimensional data labels fed back by Step 5.

[0048] (4) The real-time event analysis includes: realizing the analysis of events on the basis of multi-source data adaptation and association. That is, when an abnormal event appears in the monitoring data of a certain data source, analyze the data in the same time period and location interval in the monitoring data of other data sources, and combine the analysis results of different monitoring data to obtain the real-time event analysis results, that is, the analysis results of the real-time monitoring data.

[0049] (5) The real-time event experience library maintenance includes: recording the analysis results of each real-time event and classifying them.

[0050] Step 4, non-real-time monitoring data processing: After pre-processing the non-real-time monitoring data, the corresponding multidimensional data is provided to step 5, and the normalized multidimensional data labels fed back in step 5 are combined to perform data segmentation and storage, and the association between different data sources is established based on the normalized multidimensional data labels, and then the data is divided according to different time scales, and the data at each time scale is analyzed separately in combination with the empirical model and the AI ​​model to obtain the analysis results of the non-real-time monitoring data.

[0051] Preferably, the non-real-time monitoring data processing includes: preprocessing and data segmentation, multi-source data adaptation and association, short-term data organization, short-term data analysis, long-term data organization, long-term data analysis, ultra-long-term data organization and ultra-long-term data analysis.

[0052] (1) The preprocessing and data segmentation include: preprocessing the non-real-time monitoring data and providing the corresponding multidimensional data to step 5, and performing data segmentation and storage in combination with the normalized multidimensional data labels fed back from step 5.

[0053] Those skilled in the art can understand that the process of transmitting information in a communication network is a layer-by-layer packaging process from the physical layer to the application layer (usually, a maximum of seven layers). The bottom layer provides services for the upper layer, and each layer has its own identification and encapsulation rules. The preprocessing mainly parses the required data part to ensure the efficiency of the program. The data segmentation method involved is relatively flexible, and users can choose according to actual needs.

[0054] (2) The multi-source data adaptation and association includes: establishing associations between different data sources based on the normalized multi-dimensional data labels fed back in step 5.

[0055] (3) Short-term data organization, long-term data organization, and ultra-long-term data organization, including: organizing data according to the time scale to which they belong; among which, the time scale is divided into three categories, from small to large, namely short-term, long-term, and ultra-long-term.

[0056] (4) Short-term data analysis, long-term data analysis, and ultra-long-term data analysis, including: analyzing the data of the corresponding time scales through empirical models and AI models respectively, and combining the outputs of the two types of models to determine the final analysis results, and integrating the final analysis results of all time scale data to obtain the analysis results of non-real-time monitoring data.

[0057] Step 5: Establish a normalized label system around multi-dimensional data with 5G-R services as the center, and transfer labels with steps 3 and 4 respectively.

[0058] Preferably, the multi-dimensional data includes data in three dimensions: time, location, and event. The data in the three dimensions are respectively normalized, and finally a normalized label system is established.

[0059] (1) The time normalization processing method includes: selecting the time of the monitoring data in the specified data source as the reference time, calculating the time deviation Δt between the time of other data sources and the reference time, and performing time deviation compensation.

[0060] (2) The location normalization processing method includes: the location data label contains various location information. Select the location information of the specified type as the reference location, calculate the distance deviation Δd between the location in each data source and the reference location, and perform distance deviation compensation; wherein, if the current data source does not contain a location, the location in other data sources is shared with the current data source by using a synchronization algorithm.

[0061] In the above time normalization and location normalization processing processes, a relatively reliable benchmark (i.e., reference time, reference location) can usually be determined first. For example, when performing time normalization, the system time of the 5G communication system can be used as the reference time, so that the deviation change between the clock of other systems and the 5G system time can be monitored in real time, and compensation can be made for the cumulative or sudden deviation. When performing location normalization processing, the precise railway track circuit information (kilometer post information) included in the data interface PRI can be used as the reference location. There is a corresponding relationship between the kilometer post and the longitude and latitude. The longitude and latitude obtained by other monitoring systems from Beidou or GPS fluctuate and cannot be directly used. It is necessary to correct them to the railway line in real time in combination with the running speed and the kilometer post; the base station location can also be used as a backup reference location. Although the accuracy is lower than that of the kilometer post, it can be used as a backup to improve the anti-risk ability of the system. Of course, only examples of the reference time and reference location are provided above. In actual applications, users can adjust according to needs or experience.

[0062] (3) The event normalization processing method includes: defining various events, assigning a unique code to each defined event, identifying the event type through the unique code, and establishing the association between each event and time and location.

[0063] Preferably, the steps of event normalization processing include:

[0064] (3.1) Event coding: Assign a unique code to each event.

[0065] (3.2) Event attribute description: Describe the specific attributes of the event.

[0066] (3.3) Event data standardization: Standardize the time of different data sources, including: unifying the data format and standardizing the data fields.

[0067] (3.4) Synchronize events with time.

[0068] (3.5) And associate with location.

[0069] Step 6: Integrate the analysis results of the real-time monitoring data and the non-real-time monitoring data, and output the analysis results of network situation awareness.

[0070] Preferably, the integration of the analysis results of the real-time monitoring data and the non-real-time monitoring data and the output of the analysis results of network situation awareness include: comprehensive scoring and comparison, report generation and push, prediction and alarm, and network optimization operation closed-loop.

[0071] (1) Comprehensive scoring and comparison, including: performing multi-dimensional evaluation and comparison on the analysis results of the real-time monitoring data and the non-real-time monitoring data, that is: establishing a health range benchmark through the analysis results of the non-real-time monitoring data, determining the boundaries of the health status of each operation dimension of the 5G communication network in different scenarios (that is, a set of boundary values corresponding to different health statuses), and comparing the analysis results of the real-time monitoring data with each boundary to obtain the comprehensive scoring result of the 5G communication network; and, also performing multi-dimensional evaluation and comparison on the analysis results of the real-time monitoring data and the non-real-time monitoring data in the 5G-R communication network before and after adjustment and optimization to determine the effect after adjustment.

[0072] Those skilled in the art can understand that the communication system is limited by many factors inside and outside the system. The communication system has a health range based on scenarios, and the health range is defined during the long-term operation process (generally described by a series of operation indicators). When certain performances of the communication system run to the health boundary, the system will compare the deviation amplitude, distribution characteristics, etc. to analyze the cause of the problem and trigger an alarm.

[0073] (2) Report generation and push, including: generating an analysis report and pushing it in combination with the analysis results of the real-time monitoring data, the non-real-time monitoring data, and the comprehensive scoring result.

[0074] (3) Prediction and alarm, including: combining the analysis results of the non-real-time monitoring data for problem prediction and trend analysis, and combining the comprehensive scoring result of the 5G communication network to determine the events in the real-time monitoring data that exceed the health range benchmark and generate corresponding alarm prompts.

[0075] In the embodiments of the present invention, the output of the above analysis report, prediction, and alarm is collectively referred to as the analysis result of network situation awareness; the above alarm prompt is a hierarchical prompt, and the corresponding alarm level is determined according to the situation of crossing the boundary, and then the alarm prompt of the corresponding level is generated.

[0076] (4) The closed-loop of network optimization operations includes: adjusting and optimizing the 5G-R communication network according to the analysis results of network situation awareness.

[0077] Preferably, the method further includes: a step of inputting external environment data, where the external environment data includes: geographical and topographical data, meteorological data, and environmental interference data; the external environment data is provided to steps 3 and 4 for correcting the corresponding analysis results, and further, the external environment data is also used for AI model training.

[0078] Preferably, the method further includes: steps of AI model training and deployment; during training, a training set is constructed using the stored original data to train the selected AI model. During the training process, the AI model is corrected in combination with the external environment data; after training, the performance of the AI model is evaluated, and the AI model is optimized in combination with the evaluation results, and the AI model is deployed to the production environment for processing non-real-time monitoring data.

[0079] The above solution provided by the embodiments of the present invention is applicable to various comprehensive process evaluations from quantitative change to qualitative change, pre-analysis and early warning of abnormal events, and cause analysis of abnormal events (such as: communication events, wireless signal coverage and quality, railway line scenarios, meteorology, and noise, etc.), to solve the automatic correlation and intelligent analysis of multi-data sources, greatly improving the analysis accuracy and work efficiency; especially highlighting the pre-warning from shallow correlation to deep correlation quantitative change until the qualitative change of abnormal events and the correlation backtracking analysis of the cause of abnormal events after the event.

[0080] In order to more clearly show the technical solution provided by the present invention and the technical effects produced, the method provided by the embodiments of the present invention will be described in detail below with specific embodiments.

[0081] As Figure 2 shown, it is the overall framework of an analysis method for 5G-R multi-monitoring data fusion detection network situation awareness provided by the embodiments of the present invention; various 5G-R dedicated communication network monitoring and detection data are used in this method framework, as Figure 3 shown, including: 5G communication network, 5G interface DPI, 5G northbound data, 5G transmission network management, 5G wireless probes, 5G road test equipment, 5G RMS, antenna / tower monitoring, computer room dynamic environment monitoring, etc.; and, the association between data is established through three-dimensional tags (time, location, event), as Figure 4 shown; during the data analysis process, real-time and non-real-time data are processed and combined respectively, as Figure 2 and Figure 5As shown in the figure; to adapt to the AI access to the 5G-R multi-monitoring data fusion detection network situation awareness, design the data and label channels of the AI module, as well as the iterative optimization of the training model. At the beginning of the design of the present invention, the convenience of programming is considered, which is convenient for rapid deployment, greatly improves the accuracy and efficiency of system analysis, and has a wide range of applications. The following will introduce each part of the method framework in detail.

[0082] Part A: Real-time monitoring data interface.

[0083] The real-time monitoring data mainly includes: real-time dynamic data in data sources such as 5G network northbound alarm data, 5G interface monitoring system data, 5G wireless probe data, 5G transmission network management system data, 5G network power environment monitoring data, antenna and tower monitoring data, 5G-R network northbound performance data, 5G-R road test equipment data, 5G-R network configuration parameters, 5G-R equipment inventory, MDT minimum drive test terminal measurement and acquisition system data, CIR system vehicle-mounted equipment monitoring system data, MCX railway service dedicated processing service data, etc. Its main feature is that once an abnormality occurs in the communication network, it will inevitably lead to fluctuations in real-time dynamic data, which is the most sensitive area for situation awareness.

[0084] In this part, the real-time monitoring data is quickly parsed, filtered, and partitioned and buffered. (1) Use fast parsing for the layer-by-layer encapsulated data, and only parse to the layers and information directly related to network performance to ensure parsing efficiency. (2) Filter the layers and information not directly related to network performance to streamline the data. (3) Partitioned caching is the format preparation and label preparation before entering Part E for real-time monitoring data processing. Different data sources pass through their respective independent channels. The channels meet the respective data bandwidth and processing capabilities, and the channels themselves can also check the connection status of the data sources. Generally, it is completed in memory to ensure efficiency.

[0085] In addition, the data stream of the real-time monitoring data will be losslessly retained in Part C in the form of a file for data review or combined with labels as the input for AI training.

[0086] Part B: Non-real-time monitoring data interface.

[0087] Non-real-time monitoring data mainly includes non-real-time dynamic data from data sources such as 5G network northbound alarm data, 5G interface monitoring system data, 5G wireless probe data, 5G transmission network management system data, 5G network power environment monitoring data, antenna and tower monitoring data, 5G-R network northbound performance data, 5G-R road test equipment data, 5G-R network configuration parameters, 5G-R equipment inventory, MDT minimum road test terminal measurement and acquisition system data, CIR system vehicle-mounted equipment monitoring system data, MCX railway service dedicated processing service data, etc. Its main characteristics are: the data is not collected and transmitted immediately during the process of collection, collation and transmission, or the data source is not the equipment participating in the work (indirect monitoring equipment), or the data fluctuation does not necessarily cause extreme events such as communication interruption, but the potential possibility of abnormal events in situation awareness.

[0088] In this part, interface adaptation is performed on non-real-time dynamic data respectively. Non-real-time monitoring data generally undergoes parsing, cleaning and storage in its respective systems. Non-real-time monitoring data usually adopts database sharing, and the data is encapsulated in a shared directory or pushed in a triggered manner according to a fixed time interval or fixed file size. Different data sources pass through independent channels respectively, and the channels meet the respective data bandwidth and reading capabilities.

[0089] In addition, the data stream of non-real-time monitoring data will be losslessly retained in Part C in the form of files for data review or combined with tags as the input for AI training.

[0090] Part C: Full life cycle management of raw data.

[0091] Full life cycle management of raw data is mainly about backing up raw data. Since any storage is not infinite, calling a huge amount of data is very resource-consuming. The present invention uses data source type, scope of action and time segmentation for storage; multiple data can be adapted to various types, such as binary, files, etc.; data value evaluation strategies include: defining the weight of monitoring data for different scenarios, dynamically adjusting the importance weight of multiple data through time sliding window, and downgrading the weight of historical data for different levels of network adjustment.

[0092] Part D: Tag adaptation.

[0093] Tag adaptation needs to establish a three-dimensional normalized tag system centered on 5G-R services around time, location and events (such as Figure 4 ) to analyze and guarantee resources, signaling and quality of services in various scenarios, such as Figure 6As shown. Part E (real-time monitoring data processing) and Part F (non-real-time monitoring data processing) will respectively form an interactive transfer of tags with Part D. Part E and Part F provide time, location, and events in the data. The sources of real-time and non-real-time monitoring data are diverse and involve multiple links of the 5G-R communication network. The starting points of the system clocks, sample granularity, precision deviation, format, etc. of each system are also not unified. Some monitoring data has no location information, and the events belong to different communication protocol layers. Therefore, the three dimensions of 5G-R based on services: time, location, and events need to be normalized, such as Figure 3 As shown.

[0094] (D1) Time normalization processing scheme. Many monitoring systems in the communication network have their own operating system clocks, and most of their clocks have different reference points, transmission delays, precisions, formats, and calibration methods. The first step in joint detection and monitoring is to establish time synchronization and maintain inter-system time deviation compensation. The present invention selects the time of the monitoring data in the specified system data source as the reference time. There is a time deviation Δt between the time of other systems and the reference time. Δt is a variable. Calculate and monitor Δt for deviation compensation. When Δt is too large (i.e., exceeds the set value), an alarm will be triggered to reset the clock of the system with too large Δt deviation.

[0095] Exemplarily, the time of the 5G-R interface monitoring system (because the information of each communication link is configured with complete and high-precision system time) can be selected as the reference time.

[0096] (D2) Location normalization processing scheme. Each communication link has a corresponding location. The location of the wired network in communication is generally very clear and accurate. The difficulty in positioning lies in the wireless network part along the railway line. The types of location information collected along the line include the kilometer post based on the track circuit (high precision, large interval), the longitude and latitude reported by the terminal (the precision is affected by the terrain environment, such as mountain tunnels), and the base station coverage area (the wireless signal fluctuates).

[0097] The present invention selects the location information of a specified type as the reference location. There is a distance deviation Δd between other types of locations and the reference location. Δd is a variable. The vehicle has kilometer posts and speed information passing through the kilometer posts. Calculate and fill in the locations between the kilometer posts. Exemplarily, the kilometer post can be selected as the reference location (because the information of each communication link can be anchored to the relative location of the nearest kilometer post).

[0098] Especially: If the data source has no location information, the location information will be shared with the systems without location information in the system through the inter-system synchronization algorithm (refer to the time reference for supplementation).

[0099] Especially: Figure 4On the horizontal axis shown, the directions included mainly refer to the communication and vehicle operation directions. For example, communication includes uplink and downlink, and the directions mainly include the uplink direction and the downlink direction. Similarly, the vehicle operation directions also mainly include the uplink direction and the downlink direction. The wireless performance in different directions varies greatly.

[0100] (D3) Event normalization processing scheme. There are various types of events in a communication network, including user behavior events, network fault events, equipment maintenance events, etc. To achieve unified management of events, it is first necessary to clearly define and classify various events. For example, user behavior events can be divided into residence, access, hold, movement, end, release, etc.; network fault events can be divided into network faults, wireless network interruption, transmission network interruption, terminal faults, etc. For each defined event, a unique code needs to be assigned to ensure the uniqueness and identifiability of the event in the system, including the following steps:

[0101] (D3.1) Event coding: Assign a unique code to each event for quick retrieval and processing.

[0102] (D3.2) Event attribute description: Describe the specific attributes of the event, including the time, location, trigger conditions, scope of influence, etc. when the event occurs.

[0103] (D3.3) Event data standardization: Standardize the time of different data sources. The event data generated by different systems may have differences in format and data structure. To achieve unified analysis of events, these event data need to be standardized. It mainly includes: (3.31) Data format unification: Convert the event data of different systems into a unified format, such as JSON, XML, etc. (3.32) Data field standardization: Standardize each field in the event data to ensure that each field has the same meaning and data type in different systems.

[0104] (D3.4) Synchronize events with time: To ensure the accuracy of event analysis, events from different data sources need to be synchronized in time. Here, the established normalization time system can be referred to, unify the timestamps of all events to the clock of the 5G-R interface monitoring system, and perform time deviation compensation.

[0105] (D3.5) And associate with the location: The location information where the event occurs is crucial for subsequent analysis and processing. Through the normalization location system, associate the event with specific location information. For example, a base station disconnection event can be associated with the specific location of the base station to help locate and solve network faults.

[0106] Based on the above solution, the following can be achieved: (1) Multi-dimensional parallel adaptation: Parallelly complete the correlation adaptation of multiple monitoring systems in terms of communication events, monitoring time, line location, and railway service dimension, and construct an integrated set based on the 5G-R network railway communication service; (2) Dynamic maintenance of time variables: The monitoring time offset Δt of multiple detection systems is that multiple variables need to have a mechanism to continuously monitor their change ranges and implement compensation for the offset Δt; (3) Dynamic maintenance of location variables along the line: The monitoring railway line location offset Δd of multiple detection systems is that multiple variables need to have a mechanism to continuously monitor their change ranges and implement compensation for the offset Δd; (4) Dynamic maintenance of various measurement deviations: Differences caused by different monitoring system sensitivities, radio frequency connection methods, or usage scenarios require a mechanism to continuously monitor their change ranges and implement correction of the deviation amount Δp.

[0107] Part E: Real-time monitoring data processing.

[0108] This part mainly includes: multi-dimensional data label extraction, anomaly and its weight identification, multi-source data adaptation and association, real-time event analysis, and maintenance of the real-time event experience library.

[0109] (E1) Multi-dimensional data label extraction is to "extract" time, location, and events in the data. If a certain dimension is missing in some data, it can be ignored.

[0110] (E2) Anomaly and its weight identification. This part needs to be classified and identified according to the importance of 5G-R communication and the services it carries. The above two items are both sent to the label adaptation module.

[0111] (E3) Multi-source data adaptation and association is based on the system label normalization coordinate system in Part D (as Figure 3 shown), and calibrates the data of different monitoring systems to form data association for the whole system (as Figure 2 shown).

[0112] (E4) Real-time event analysis is under the architecture of system label normalization coordinate system and multi-monitoring system data fusion as in Figure 2 and 3 , and quickly completes a comprehensive analysis of events. For example: If the 5G wireless probe discovers that the wireless signal strength drops steeply by 7 dB in a specific railway line section during a certain period, then start scanning the relevant measurements and signaling of other systems in the same time period (and the extended time periods before and after), and the location section (the range is appropriately extended): Whether there are alarms for relevant devices, whether the antenna azimuth has shifted, whether there are alarms for the dynamic environment monitoring of the corresponding cell computer room, the signaling (abnormal release and abnormal handover of all terminals) and measurements (abnormal strength and quality) of all terminals at the 5G-R interface during this period and location. Through the mutual verification of the monitoring data of different systems, a more confirmed problem description and analysis result can be obtained.

[0113] (E5) The maintenance of the real-time event experience database can classify and record each event, including historical descriptions such as network device failure files, terminal device performance logs, noise characteristics, coverage and quality, signaling and triggers, etc., as well as effective processing methods.

[0114] Part F: Non-real-time monitoring data processing.

[0115] This part mainly includes: preprocessing and data segmentation, multi-source data adaptation and association, short-term data organization, short-term data analysis, long-term data organization, long-term data analysis, ultra-long-term data organization, and ultra-long-term data analysis.

[0116] (F1) Preprocessing and data segmentation: The sources of non-real-time monitoring data are diverse and involve multiple links of the 5G-R communication network. As Figure 2 shown, each link will be provided with process monitoring data by multiple monitoring systems and classified according to the link; there are network-side monitoring and terminal monitoring, classified according to the reporting object; communication monitoring includes service content, communication events, and channel measurements, classified according to the monitoring content. Finally, label addition and data segmentation storage are completed through interaction with Part D.

[0117] (F2) Multi-source data adaptation and association: According to the communication logic of the 5G-R network, establish cross-data-source connections with tags as the anchor points to complete the reproduction of communication events and channel measurements in the communication process in the time and space dimensions around communication services. 5G-R is divided into wired communication and wireless communication. The wired part of 5G-R is fully IP-based, and the focus is on the rationality of the possible node routing, bandwidth load, and delay of the data passing through. The wireless part of 5G-R needs to cope with the complex scenarios along the railway. While detecting the bandwidth load, it is necessary to continuously monitor the signal strength, quality, interference, and signal overlap relationship.

[0118] (F3) In order to cope with the diversity and complexity of scenarios, the data is divided into multiple time scales and analyzed separately. The changes brought by scenarios to wireless communication are essentially functions on the time axis. Different scenario types correspond to different time ranges. For example, the fading of wireless signals in space is divided into fast fading and slow fading: fast fading is at the millisecond level, and slow fading may last for several seconds or dozens of seconds; the impact of rain, snow, and thick fog on wireless signals depends on the meteorological cycle; the occlusion and attenuation of wireless signals by the vegetation growth cycle process and meteorological changes need to refer to the seasonal cycle; the changes in terrain and landform often accumulate over the years; the device usage cycle also varies due to different working environments and life cycles. Different time analysis periods are set in the non-real-time monitoring data processing to cope with different scenario change cycles.

[0119] Based on this, in the embodiments of the present invention, multiple time periods are allocated for data, which are generally divided into three categories: short-term data organization and analysis (minutes, hours, days), long-term data organization and analysis (weeks, months, quarters), and ultra-long-term data organization and analysis (years, multiple years). The analysis of different periods includes two parts: an empirical model and an AI model. The empirical model is a mathematical description of the distribution of multivariate monitoring data during the stable operation of each link of the communication system under the existing full-cycle sectional scenarios. The AI model is a specially trained "brain" that can learn and predict based on the input data.

[0120] The AI model is data-driven: It is trained with a large amount of data to learn the patterns and rules in the data. For example, a language model will learn many features such as grammar and semantics in the text; Automatic learning: It can automatically extract features from the data without the need for manual design of each feature; Highly complex structure: An AI model such as a deep neural network has many hidden layers and can handle highly complex non-linear problems. The AI model has strong capabilities in handling complex tasks: It performs well in complex tasks such as image recognition, speech recognition, and natural language processing. For example, in the identification of diseases in medical images, it can detect subtle features that are difficult for the human eye to notice; Good scalability: As long as there is enough data and computing resources, its performance can be continuously optimized and expanded; Strong generalization ability: A well-trained model can make reasonable predictions for new data. However, the establishment of an AI model requires a large amount of data: When the amount of data is insufficient, the performance of the model will be greatly reduced; High consumption of computing resources: Training and deploying complex AI models require powerful computing devices and a large amount of energy; Poor interpretability: It is difficult to understand how the model makes decisions, especially for black-box models in deep learning.

[0121] The empirical model is established based on experience: It is a model constructed by people based on past experience, observations, and theoretical knowledge; Parameters are often determined manually: The parameters of the model are usually determined through human experience judgment and statistical methods. The empirical model has strong interpretability: Because it is constructed based on experience and known theories, it is relatively easy to understand its principle and decision-making process; Efficient in familiar fields: In fields with sufficient experience and knowledge, the model can be quickly established and applied; Relatively flexible data requirements: It does not require as much data as the AI model and can be constructed even with less data. However, the empirical model is difficult to handle complex high-dimensional data: It has limited processing capabilities for complex and high-dimensional data (such as a large amount of signal distribution map data); Limited generalization ability: When encountering new situations that are quite different from experience, the accuracy of the model may decrease; Limited by human cognition: The quality of the model highly depends on human experience and knowledge, and some important unknown factors may be ignored.

[0122] Therefore, it is necessary to install and run the empirical model and the AI model in parallel, combining their advantages and disadvantages, for railway communication with extremely high safety requirements.

[0123] Part G: Output of conclusion report.

[0124] This part mainly includes: comprehensive scoring and comparison, report generation and push (routine report, special analysis report), prediction and warning, and closed-loop network optimization operation.

[0125] (G1) Comprehensive scoring and comparison: 5G-R communication consists of multiple links and carries multiple services. Therefore, comprehensive scoring requires a series of evaluation indicators for multi-dimensional evaluation.

[0126] (G2) Report generation and push (routine report, special analysis report): A series of routine reports such as daily reports and weekly reports are required for each specialty and management department in the operation and maintenance of the railway 5G-R communication network. The special analysis report is for specific problem matters.

[0127] (G3) Prediction and warning: Predict and warn of potential problems through a health experience model or an AI model.

[0128] (G4) Closed-loop network optimization operation: During the operation and maintenance of the 5G-R communication network, the network is adjusted and optimized, and the indicators before and after the adjustment are iteratively tracked.

[0129] The data in the conclusion report output part comes from the processing of real-time monitoring data and non-real-time monitoring data, and the report results will optimize the AI model parameters in the AI model training.

[0130] Based on the above solution, multi-level situation awareness can be achieved, such as Figure 5 shown: Based on the multi-level situation awareness of 5G-R services and multi-dimensional evaluation, 5G-R carries many services with different requirements for communication resources and quality. Therefore, each service has different situation awareness requirements.

[0131] In the embodiment of the present invention, an emergency degradation processing mechanism is also introduced: when data is missing, data mismatch occurs, or part of the monitoring system fails. According to the damaged situation, the overall monitoring and analysis will not completely fail, and part of the monitoring tasks can still be completed.

[0132] Part H: Input of external environment data.

[0133] In the embodiment of the present invention, the external environment data mainly includes: geographical and geomorphic data, meteorological data, environmental interference, etc.

[0134] (H1) Geographical and geomorphic data: Generally connected to a three-dimensional geographical information digital map, and the online map has high timeliness of update.

[0135] (H2) Meteorological data: Need to be connected to the national meteorological data release platform to obtain authoritative and accurate data.

[0136] (H3) Environmental interference data: The interference data mainly comes from dedicated wireless monitoring probes, spectrum analyzers, and the subcarrier noise floor of 5G-R communication operation and maintenance.

[0137] In the embodiments of the present invention, the external environmental data participates in the real-time monitoring data processing and the non-real-time monitoring data processing part to correct the analysis results, and the external environmental data enters the AI model training part to enrich the AI training data.

[0138] Part I: AI model training.

[0139] This part mainly includes: AI data interface, AI model selection, AI model training, AI model tuning and deployment, AI model evaluation, and external factor correction.

[0140] (I1) AI data interface: To meet the docking of AI training with massive data, it is necessary to meet lossless data, high-speed transmission, and accurate labeling to feed the AI model, mainly including:

[0141] (I1.1) Data interfaces, including: RESTful API, which interacts with the AI model through HTTP requests. Common operations include data upload, model inference requests, result acquisition, etc. WebSocket, used for real-time data transmission and two-way communication, suitable for scenarios that require low latency and high-frequency interaction. gRPC, a high-performance remote procedure call (RPC) framework that supports multiple languages and is suitable for AI model calls in large-scale distributed systems. GraphQL, a flexible API query language that allows clients to precisely request the required data and reduces unnecessary data transmission.

[0142] (I1.2) Data formats, including: JSON, a lightweight data exchange format widely used for passing data in API calls. XML, a traditional data format that, although less popular than JSON, is still used in some scenarios. ProtocolBuffers (Protobuf), an efficient serialization format developed by Google and commonly used in gRPC communication. CSV / TSV, a simple text format for structured data, commonly used for data import and export.

[0143] (I1.3) Data preprocessing, including: data cleaning, removing noisy data, handling missing values, correcting incorrect data, etc. Data standardization / normalization, converting data to a unified scale for easy model processing. Feature engineering, extracting useful features from the original data to enhance the performance of the model. Data augmentation, expanding the training dataset by generating new data samples (such as image rotation, text synonym replacement, etc.).

[0144] (I1.4) Model inference includes: Online inference, which processes user requests in real time, usually achieved through API calls. Batch inference, which processes a large amount of data at once and is suitable for offline tasks or background processing. Edge computing, which performs model inference locally on the device, reducing dependence on the cloud and suitable for low-latency requirements.

[0145] (I1.5) Security includes: Authentication and authorization to ensure that only authorized users or systems can access the AI model. Data encryption, which uses encryption technology during data transmission and storage to protect sensitive data. Model protection to prevent the model from being maliciously attacked or misused, such as adversarial sample attacks, model reverse engineering, etc.

[0146] (I1.6) Performance optimization includes: Model compression, which reduces the size and computational complexity of the model through techniques such as quantization and pruning to improve inference speed. Distributed computing, which uses multiple machines to process data and model inference in parallel to improve processing power. Caching mechanism, which caches the results of frequently requested data to reduce repeated calculations and improve response speed.

[0147] (I1.7) Development tools and frameworks include: TensorFlow Serving: A tool for deploying and managing TensorFlow models. PyTorch Serve, a model serving framework for PyTorch that supports fast deployment and inference. Kubernetes, which is used for containerized deployment and management of AI models and supports elastic scaling and high availability. Docker, which is used for containerizing AI models to simplify deployment and environment management.

[0148] (I1.8) Data storage includes: Databases, which are used to store training data, model parameters, inference results, etc. Common databases include relational databases (such as MySQL, PostgreSQL) and NoSQL databases (such as MongoDB, Cassandra). Object storage, which is used to store large-scale data files, such as training datasets, model files, etc. Common object storage services include AWS S3, Google Cloud Storage, etc.

[0149] (I1.9) Logging and monitoring includes: Logging, which records key information during API calls and model inference processes for easy problem troubleshooting and performance analysis. Monitoring systems, which monitor the running status of the system in real time, including API request volume, model.

[0150] (I2.1) AI model selection: Select an appropriate AI model based on the requirements of business analysis objectives and data source characteristics. Commonly used models include:

[0151] (I2.2) Classification Models: Classification models are used to assign data to predefined categories. Common tasks include image classification, text classification, spam detection, etc.;

[0152] (I2.3) Regression Models: Regression models are used to predict continuous values. Common tasks include house price prediction, sales prediction, etc.;

[0153] (I2.4) Clustering Models: Clustering models are used to divide data into different groups. Common tasks include customer segmentation, image segmentation, etc.;

[0154] (I2.5) Recommendation Models: Recommendation models are used to provide personalized recommendations for users. Common tasks include product recommendation, video recommendation, etc.;

[0155] (I2.6) Natural Language Processing Models (NLP Models): NLP models are used to process and analyze text data. Common tasks include text classification, sentiment analysis, machine translation, etc.

[0156] (I2.7) Computer Vision Models: Computer vision models are used to process image and video data. Common tasks include image classification, object detection, image segmentation, etc.

[0157] (I2.8) Time Series Forecasting Models: Time series forecasting models are used to process time series data. Common tasks include stock price prediction, weather forecasting, etc.

[0158] (I2.9) Generative Models: Generative models are used to generate new data samples. Common tasks include image generation, text generation, etc.

[0159] (I3) AI Training Process: The full data and the determined model are the inputs for model training, which is a core process in machine learning and deep learning. Through this process, the model learns from the data and optimizes its performance to complete specific tasks. The basic goal of training is to enable the model to make accurate predictions or decisions on new data. The training process includes the following 5 steps:

[0160] (I3.1) Forward Propagation: Pass the input data through the model to calculate the predicted values.

[0161] (I3.2) Calculate the loss: Use the loss function to calculate the difference between the predicted value and the true value.

[0162] (I3.3) Backward Propagation: Calculate the gradient of the loss function with respect to the model parameters.

[0163] (I3.4) Parameter update: Use an optimization algorithm to update the model parameters to reduce the loss function.

[0164] (I3.5) Iteration: Repeat the above steps until the model converges or reaches a predetermined number of training epochs.

[0165] (I4) AI model evaluation: After the training process, it is necessary to regularly evaluate the performance of the model to ensure that the model is not overfitting or underfitting. Commonly used evaluation metrics include:

[0166] (I4.1) Accuracy: The proportion of correctly predicted samples to the total number of samples in a classification task.

[0167] (I4.2) Precision: The proportion of samples correctly predicted as positive to the number of samples predicted as positive in a classification task.

[0168] (I4.3) Recall: The proportion of samples correctly predicted as positive to the number of samples actually positive in a classification task.

[0169] (I4.4) F1 Score: The harmonic mean of precision and recall.

[0170] (I4.5) Mean Squared Error (MSE): The average of the squared differences between the predicted values and the true values in a regression task.

[0171] (I4.6) R 2 -squared: The goodness of fit of the regression line in a regression task.

[0172] (I5) AI model tuning: According to the evaluation results, adjust the hyperparameters of the model, such as the learning rate, batch size, number of hidden layer nodes, etc., to optimize the model performance. Commonly used tuning methods include:

[0173] (I5.1) Grid Search: Search in a predefined hyperparameter grid.

[0174] (I5.2) Random Search: Randomly select in the hyperparameter space for search.

[0175] (I5.3) Bayesian optimization: Use Bayesian methods to optimize hyperparameters.

[0176] (I6) AI model deployment: After training is completed, the model is deployed to the production environment for actual applications, and the AI prediction results are compared with the traditional analysis results. During the deployment process, the performance, scalability, and maintainability of the model need to be considered.

[0177] (I7) External factor correction: The external factor data is also used as part of the AI training to correct the model.

[0178] Preferably, the embodiment of the present invention also establishes an integrated label: comprehensively labels the events, time, location, and services in the 5G-R network communication process, which is convenient for manual analysis or AI model training. The label includes two parts: identification type and attribute type.

[0179] Embodiment Two

[0180] The embodiment of the present invention also provides an analysis system for 5G-R multi-monitoring data fusion detection network situation awareness. This system is used to implement the method provided in the foregoing embodiment, and reference can also be made to Figure 2 , and this system mainly includes:

[0181] A multi-source data acquisition and division unit, which is used to acquire multi-source data of the 5G-R communication network and divide it into real-time monitoring data and non-real-time monitoring data;

[0182] An original data full life cycle management unit, which is used to store the real-time monitoring data and non-real-time monitoring data as original data and perform full life cycle management, and is also used for AI model training;

[0183] A real-time monitoring data processing unit, which is used for real-time monitoring data processing, including: extracting the real-time monitoring data in multiple dimensions, and providing it to step 5 after classification and identification, and then performing data analysis based on the normalized multi-dimensional data label fed back by step 5 to obtain the analysis result of the real-time monitoring data;

[0184] A non-real-time monitoring data processing unit, which is used for non-real-time monitoring data processing, including: preprocessing the non-real-time monitoring data and providing the corresponding multi-dimensional data to step 5, and combining the normalized multi-dimensional data label fed back by step 5 to perform data segmentation and storage, and establishing the association between different data sources based on the normalized multi-dimensional data label, and then dividing the data on different time scales, and respectively performing data analysis on the data under each time scale by combining the empirical model and the AI model to obtain the analysis result of the non-real-time monitoring data;

[0185] A label adaptation unit, which is used to establish a normalized label system around the multi-dimensional data centered on 5G-R services, and transfer the labels to step 3 and step 4 respectively;

[0186] A conclusion report output unit is configured to synthesize the analysis results of the real-time monitoring data and the analysis results of the non-real-time monitoring data, and output the analysis results of network situation awareness.

[0187] Considering that the main technical details of each part involved in the system have been introduced in detail in the previous embodiments, they will not be elaborated here.

[0188] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiments can be implemented by software or by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solutions of the above embodiments can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0189] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the above division of each functional module is used as an example. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the system is divided into different functional modules to complete all or part of the functions described above.

[0190] As mentioned above, the above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims. The information disclosed in the background art part of this article is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as an admission or any form of implication that this information constitutes prior art known to those skilled in the art.

Claims

1. A 5G-R multi-monitoring data fusion detection network situation awareness analysis method, characterized in that: include: Step 1: Obtain multivariate data of the 5G-R communication network and divide it into real-time monitoring data and non-real-time monitoring data; Step 2: The real-time monitoring data and the non-real-time monitoring data are stored as raw data and managed throughout their life cycle, and used for AI model training; Step 3, real-time monitoring data processing: extract the real-time monitoring data in multiple dimensions, classify and label it, and then provide it to step 5 for use, and then perform data analysis based on the normalized multi-dimensional data labels fed back from step 5 to obtain the analysis results of the real-time monitoring data; Step 4, non-real-time monitoring data processing: After pre-processing the non-real-time monitoring data, the corresponding multi-dimensional data is provided to step 5, and the data is segmented and stored in combination with the normalized multi-dimensional data labels fed back in step 5, and the association between different data sources is established based on the normalized multi-dimensional data labels, and then the data is divided according to different time scales, and the data at each time scale is analyzed respectively in combination with the empirical model and the AI ​​model to obtain the analysis results of the non-real-time monitoring data; Step 5: Establish a normalized label system around multi-dimensional data with 5G-R services as the center, and transfer labels with steps 3 and 4 respectively; Step 6: Combining the analysis results of the real-time monitoring data with the analysis results of the non-real-time monitoring data, outputting the analysis results of the network situation awareness.

2. According to claim 1, a 5G-R multi-monitoring data fusion detection network situation awareness analysis method is characterized in that: The 5G-R communication network multivariate data includes: 5G network northbound alarm data, 5G interface monitoring system data, 5G wireless probe data, 5G transmission network management system data, 5G network dynamic environment monitoring data, antenna and tower monitoring data, 5G-R network northbound performance data, 5G-R road test equipment data, 5G-R network configuration parameters, 5G-R equipment ledger, MDT minimum road test terminal measurement and collection system data, CIR system on-board equipment monitoring system data and MCX railway business dedicated processing service data.

3. According to claim 1, a 5G-R multi-monitoring data fusion detection network situation awareness analysis method is characterized in that: The real-time monitoring data processing includes: multi-dimensional data label extraction, anomaly and its weight identification, multi-source data adaptation and association, real-time event analysis and real-time event experience library maintenance; wherein: The multi-dimensional data extraction includes: extracting the three dimensions of time, location and event from various types of real-time monitoring data respectively; The anomaly and its weight identification include: determining the weight according to the importance of 5G-R communication and the carried business, and classifying and identifying it accordingly, performing label adaptation based on the classification and identification results, and judging whether there is an abnormal situation; The multi-source data adaptation and association includes: establishing associations between different data sources based on the normalized multi-dimensional data tags fed back in step 5; The real-time event analysis includes: realizing event analysis based on multi-source data adaptation and association, that is, when an abnormal event occurs in the monitoring data in a certain data source, analyzing the data of the same time period and location interval in the monitoring data in other data sources, combining the analysis results of different monitoring data, and obtaining the real-time event analysis results, that is, the analysis results of the real-time monitoring data; The real-time event experience database maintenance includes: recording and classifying the analysis results of each real-time event.

4. According to claim 1, a 5G-R multi-monitoring data fusion detection network situation awareness analysis method is characterized in that: The non-real-time monitoring data processing includes: preprocessing and data segmentation, multi-source data adaptation and association, short-term data organization, short-term data analysis, long-term data organization, long-term data analysis, ultra-long-term data organization and ultra-long-term data analysis; wherein: The preprocessing and data segmentation includes: preprocessing the non-real-time monitoring data and providing the corresponding multi-dimensional data to step 5, and performing data segmentation and storage in combination with the normalized multi-dimensional data labels fed back from step 5; The multi-source data adaptation and association includes: establishing associations between different data sources based on the normalized multi-dimensional data tags fed back in step 5; The time scale is divided into three categories, from small to large, called short-term, long-term and ultra-long-term; Short-term data organization, long-term data organization and ultra-long-term data organization include: organizing data according to the time scale to which it belongs; Short-term data analysis, long-term data analysis and ultra-long-term data analysis include: analyzing the data of the corresponding time scales through empirical models and AI models respectively, and combining the outputs of the two types of models to determine the final analysis results, and integrating the final analysis results of all time scale data to obtain the analysis results of non-real-time monitoring data.

5. According to claim 1, a 5G-R multi-monitoring data fusion detection network situation awareness analysis method is characterized in that: The 5G-R service is centered around multi-dimensional data to establish a normalized label system including: Multidimensional data includes three dimensions: time, location and event. The three dimensions are normalized separately to establish a normalized label system. The time normalization processing method includes: selecting the time of the monitoring data in the specified data source as the reference time, calculating the time deviation between the time of other data sources and the reference time, and performing time deviation compensation; The position normalization processing method includes: the position data tag contains multiple types of position information, selects the specified type of position information as the reference position, calculates the distance deviation between the position in each data source and the reference position, and performs distance deviation compensation; wherein, if the current data source does not contain the position, a synchronization algorithm is used to share the position in other data sources with the current data source; The event normalization processing method includes: defining various types of events, assigning a unique code to each defined event, identifying the event type through the unique code, and establishing an association between each event and time and location.

6. According to claim 1, a 5G-R multi-monitoring data fusion detection network situation awareness analysis method is characterized in that: The steps of event normalization processing include: Event coding: Assign a unique code to each event; Event attribute description: describe the specific attributes of the event; Event data standardization: standardize the time of different data sources, including: unify data formats and standardize data fields; Synchronize events with time and associate them with locations.

7. According to claim 1, a 5G-R multi-monitoring data fusion detection network situation awareness analysis method is characterized in that: The analysis results of the network situation awareness are output by integrating the analysis results of the real-time monitoring data with the analysis results of the non-real-time monitoring data, including: comprehensive scoring and comparison, report generation and push, prediction and alarm, and network optimization operation closed loop; wherein: Comprehensive scoring and comparison, including: multi-dimensional evaluation and comparison of the analysis results of the real-time monitoring data and the analysis results of the non-real-time monitoring data, that is, through the analysis results of the non-real-time monitoring data, establishing a health range benchmark, determining the boundaries of the health status of each operating dimension of the 5G communication network in different scenarios, and comparing the analysis results of the real-time monitoring data with each boundary to obtain a comprehensive scoring result of the 5G communication network; and multi-dimensional evaluation and comparison of the analysis results of the real-time monitoring data in the 5G-R communication network before and after adjustment and optimization and the analysis results of the non-real-time monitoring data to determine the effect after adjustment; Report generation and push, including: combining the analysis results of real-time monitoring data, the analysis results of non-real-time monitoring data and the comprehensive scoring results, generating and pushing analysis reports; Prediction and alarm, including: combining the analysis results of non-real-time monitoring data to predict problems and analyze trends, and combining the comprehensive scoring results of the 5G communication network to determine events that exceed the healthy range benchmark in the real-time monitoring data and generate corresponding alarm prompts; the outputs of the above analysis reports, predictions and alarms are collectively referred to as the analysis results of network situation awareness; The network optimization operation closed loop includes: adjusting and optimizing the 5G-R communication network based on the analysis results of network situation awareness.

8. According to claim 1, a 5G-R multi-monitoring data fusion detection network situation awareness analysis method is characterized in that: The method also includes: a step of inputting external environmental data, wherein the external environmental data includes: geographical data, meteorological data and environmental interference data; the external environmental data is provided to steps 3 and 4 for correcting corresponding analysis results, and the external environmental data is also used for AI model training.

9. The analysis method for 5G-R multi-monitoring data fusion detection network situation awareness according to claim 1 or 8, characterized in that: The method also includes: the steps of AI model training and deployment; during training, a training set is constructed using the stored original data to train the selected AI model, and during the training process, the AI ​​model is corrected in combination with external environment data; after the training is completed, the performance of the AI ​​model is evaluated, and the AI ​​model is tuned based on the evaluation results, and the AI ​​model is deployed to a production environment for processing non-real-time monitoring data.

10. A 5G-R multi-monitoring data fusion detection network situation awareness analysis system, characterized in that: The method for implementing any one of claims 1 to 9 comprises: A multivariate data acquisition and division unit, used to acquire multivariate data of the 5G-R communication network and divide it into real-time monitoring data and non-real-time monitoring data; A raw data full life cycle management unit, used to store the real-time monitoring data and non-real-time monitoring data as raw data and manage them throughout their life cycle, and to use them for AI model training; The real-time monitoring data processing unit is used for real-time monitoring data processing, including: extracting multiple dimensions of the real-time monitoring data, and providing the data for use in step 5 after classification and identification, and then performing data analysis based on the normalized multi-dimensional data labels fed back from step 5 to obtain the analysis results of the real-time monitoring data; The non-real-time monitoring data processing unit is used for non-real-time monitoring data processing, including: pre-processing the non-real-time monitoring data and providing the corresponding multi-dimensional data for use in step 5, and segmenting and storing the data in combination with the normalized multi-dimensional data labels fed back in step 5, and establishing associations between different data sources based on the normalized multi-dimensional data labels, and then dividing the data at different time scales, and analyzing the data at each time scale in combination with the empirical model and the AI ​​model to obtain the analysis results of the non-real-time monitoring data; The label adaptation unit is used to establish a normalized label system around multi-dimensional data with 5G-R services as the center, and transfer labels with step 3 and step 4 respectively; The conclusion report output unit is used to integrate the analysis results of the real-time monitoring data and the analysis results of the non-real-time monitoring data to output the analysis results of the network situation awareness.

Images