A method for predicting a cellular internet of things traffic burst

Abstract

The application discloses a kind of cellular internet of things traffic burst prediction method, comprising the following steps: step one, the number of redundant sim cards Rr that current remaining bandwidth of gateway can allow to join is calculated;Step two, Rr-map-1 model is constructed;Step three, Rr-map-1 model is trained;Step four, the record creation moment in Rr-map-1 model, the discontinuous data in Rr-map-1 model is eliminated, for each record associated risk level data, obtain Rr-map-2 model;Step five, update time continuity step, update Rr-map-2 model;Step six, with the data of sim card business activities as input, Rr-map-2 model outputs risk level Lr;Step seven, according to Lr provides multi-dimensional congestion reduction measures, ISP executes congestion reduction measures.The application is used to predict the risk level Lr and time of traffic burst, provides quantifiable reference for ISP to take evasive action in advance, can reduce the frequency of congestion event caused by traffic burst, finally reduce the loss caused by congestion caused by traffic burst.

Description

Technical Field

[0001] This invention belongs to the field of data communication technology, and specifically relates to a method for predicting bursts of traffic in cellular Internet of Things (IoT) systems. Background Technology

[0002] Cellular Internet of Things (IoT) is a terrestrial radio communication network consisting of base stations and mobile stations. It provides mobile phone connectivity, data transmission, and internet access. The development of IoT application technologies has spurred numerous new human-computer interaction methods. Any object that needs to be monitored, connected, or interacted with will access the network through IoT access nodes. The traffic characteristics generated by these terminal access devices are highly variable, posing significant challenges to the access nodes.

[0003] Some traffic behaviors of IoT nodes are triggered by external commands or driven by internal programs, which can easily lead to a batch of nodes accessing the same service in a short period of time, causing severe congestion in the ISP's data forwarding link and affecting the stability of the ISP's network.

[0004] The growth trend in the number of IoT nodes connected is closely related to a company's business strategy. In the initial stage of a company's deployment, because the number of nodes with similar traffic behavior is relatively small, the characteristics of sudden traffic spikes are not obvious, and their consequences are relatively minor, making them difficult to detect. In this context, ISPs cannot accurately predict congestion based on a company's business activities, nor can they make targeted adjustments in advance. After a company deploys a large number of similar IoT nodes in a short period, the harm of sudden traffic spikes to the company's IoT business will be rapidly amplified.

[0005] Some IoT platform companies offer products that may be used by multiple enterprise customers, such as cloud computing, cloud storage, cloud security, and database modules for cloud data storage. Therefore, when congestion occurs, affected customers may also include companies that haven't activated many nodes but are simply using the same IoT platform, which can damage the platform company's reputation.

[0006] ISPs typically expand their capacity based on their own macro-planning and current resource utilization. Existing technologies mainly adjust and control congestion caused by sudden traffic surges. When congestion occurs due to sudden traffic surges, the urgency of the event often leads to the adoption of a crude and indiscriminate restriction strategy, resulting in an expanded range of affected customers. Summary of the Invention

[0007] The technical problem to be solved by this invention is to provide a method for predicting traffic bursts in cellular IoT, which addresses the shortcomings of the prior art. The method uses data from SIM card business activities as input, and the Rr-map-2 model outputs the risk level Lr of traffic bursts caused by incremental access to the target server caused by the business activities. This provides a quantifiable reference for ISPs to take preventative measures in advance, reduces the frequency of congestion events caused by traffic bursts, and ultimately reduces the losses caused to ISPs, enterprises, and platform providers.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for predicting bursts of cellular Internet of Things (IoT) traffic, characterized by comprising the following steps: Step 1: Calculate Rr, where Rr represents the number of redundant SIM cards that the gateway's current remaining bandwidth can allow to be added. Rr=round(max(Bu' / Eu,Bd' / Ed),0), where (Bu',Bd') represents the remaining bandwidth obtained through the BTE component, and (Eu,Ed) represents the evaluated rate pair of the destination server obtained through the BTE component. Step 2: Retrieve the customer code and MSISDN relationship data table C j The basic structure of the Rr-map model, Rr-map-1, is constructed as follows: Rr-map-1 = [Num, ip, Rr, Mr], where Num represents the number of IoT SIM cards used by a customer to access a target server IP, ip represents the destination server IP, and Mr = |C j ∩M2ASet|,M r Represents relational data table C j The number of elements in the intersection of the msisdn in the target server and the msisdn in the global or local data table of the destination server is given by relational data table C. j This is a data table showing the relationship between a customer's unique code and a SIM card, where j represents the customer's unique code. Step 3: Build a communication network for mutual access between f BTE components in the SAP system. After the gateway is put into network service, the q-th BTE component periodically sends the Rr-map-1 model data to the p-th BTE component, 1≤q≤f-1, where q is a positive integer. The p-th BTE component updates the Rr-map-1 model in its storage according to the received Rr-map-1 model data, thus completing the training of the Rr-map-1 model. Step 4: For the record creation time in the Rr-map-1 model, determine the continuity of time using a step size Δt, and remove discontinuous data from the Rr-map-1 model. Then, associate risk level data with each record in the basic structure of the Rr-map model to obtain the final structure of the Rr-map model, Rr-map-2, where Rr-map-2 = [Num, ip, Rr, Mr, Tc]. i ,Lr], where Tc i =(t i Rr i , t i+1 Rr i+1 ), Rr i Let t represent the i-th Rr, where Rr represents the number of redundant SIM cards that the gateway's current remaining bandwidth can allow to be added. i This indicates that the prediction module received Rr i The moment, t i+1 Indicates the relationship with t i For consecutive moments, Lr represents the risk level of the incremental access to the destination server to be predicted; Step 5: Collect SIM card operation activities and obtain order information. Input the order information into the Rr-map-2 model. The Rr-map-2 model outputs the risk level Lr of the traffic burst caused by the predicted incremental access to the destination server due to the operation activities. This means that sudden bursts of cellular IoT traffic are considered to cause congestion. <e<6; Step 6: Update Δt, return to Step 4, and update the Rr-map-2 model; Step 7: When traffic surges, based on the multi-dimensional congestion reduction measures provided by Lr to the ISP, the ISP implements the congestion reduction measures.

[0009] The aforementioned method for predicting bursts of cellular IoT traffic is characterized by the following specific training method for the Rr-map-1 model: Step 301: Construct a communication network that supports mutual access between f BTE components in the SAP system, and elect the p-th BTE component as the centralized data storage module. Step 302: After the gateway puts the network service into operation, the q-th BTE component periodically sends Rr-map-1 model data to the p-th BTE component, where 1≤q≤f-1 and q is a positive integer. Step 303: The p-th BTE component receives the Rr-map-1 model data sent by the q-th BTE component and merges the received Rr-map-1 model data. If the Rr-map-1 model data already exists in the Rr-map-1 model data of the p-th BTE component, then return to step one. The p-th BTE component recalculates Rr and Mr and sends the recalculated Rr and Mr back to the q-th BTE component; otherwise, the merged row records are updated in the Rr-map-1 model data of the p-th BTE component. Step 304: The q-th BTE component receives the recalculated Rr and Mr, which are used to update the Rr-map-1 model data in its database; Step 305: Iterate through f-1 BTE components and repeat steps 302-304 to complete the training of the Rr-map-1 model.

[0010] The aforementioned method for predicting bursts of cellular IoT traffic is characterized by the following: the specific method for removing discontinuous data from the Rr-map-1 model in step four is as follows: Step 401: The prediction module receives each record from the trained Rr-map-1 model and creates a time step t for each record. i , t i Indicates receipt of Rr i At any given time, a storage table Tb is formed, where Tb = [(t1, Rr1), ..., (t...]. i ,Rr i ), ..., (t n ,Rr n )]; Step 402: Process the data in storage table Tb according to Rr i Arrange the data in ascending order, and take the first h data to obtain the storage table Tb'; Step 403: Process the data in storage table Tb' according to t i Arrange the data in ascending order and update the storage table Tb'. Step 404: If in the storage table Tb', t i +Δt <t i+1 Then (t) i ,Rr i Remove from storage table Tb'; Step 405: Traverse the times in storage table Tb', repeat step 404, to obtain storage table Tb'' and set Tc. i Tc is used to store elements that exist at least two consecutive time points. i =(t i Rr i , t i+1 Rri+1 )t i+1 Indicates the relationship with t i A series of moments.

[0011] The above-mentioned method for predicting bursts of cellular IoT traffic is characterized in that: the specific method for calculating Lr in step four is as follows: ,in P represents the increment to be predicted.

[0012] The above-mentioned method for predicting bursts of cellular IoT traffic is characterized in that: the specific method for obtaining the increment P to be predicted is as follows: The BTE component periodically calls the order information of the EC platform, extracts the msisdn list, and identifies three types of work orders: purchase and account cancellation. The mirror data package is obtained through the BTE component of the SAP system to obtain a four-tuple summary. The abstract_info_table is constructed based on the four-tuple summary. The row records in the abstract_info_table are classified according to preset rules to obtain a global data table and a local data table. The intersection of the row records in the global data table or the local data table of the msisdn list is calculated, and the number of intersection elements W is calculated. If the work order type is account closure, P = -W; if the work order type is purchasing, P = W.

[0013] The aforementioned method for predicting bursts of cellular IoT traffic is characterized by the following: the specific method for obtaining (Bu', Bd') in step one is as follows: Obtain the port physical forwarding rate (B / T) of the first port of the BTE component in the SAP system. u B d ), where B u This indicates the maximum uplink bandwidth of the first port of the BTE component in the SAP system. d This indicates the maximum downlink bandwidth of the first port of the BTE component in the SAP system; The remaining bandwidth (Bu', Bd') is obtained by subtracting the currently used bandwidth of the first port from the port's physical forwarding rate (Bu, Bd), where Bu' represents the remaining uplink bandwidth and Bd' represents the remaining downlink bandwidth.

[0014] The aforementioned method for predicting bursts of cellular IoT traffic is characterized by the following: the specific method for obtaining (Eu, Ed) in step one is as follows: The mirror data package is obtained through the BTE component of the SAP system to obtain a four-tuple summary. The abstract_info_table is constructed based on the four-tuple summary. The row records in the abstract_info_table are classified according to preset rules to obtain a global data table and a local data table. Randomly sample the actual network rate corresponding to msisdn in the global or local data table, and calculate the estimated rate pair (Eu, Ed) of the global or local data, where Eu represents the average uplink rate and Ed represents the average downlink rate.

[0015] The aforementioned method for predicting bursts of cellular IoT traffic is characterized by the following method for obtaining the abstract_info_table: Step a: Start SAP, connect to the first port of the SAP system's BTE component, mirror and export the data from the first port of SAP to obtain data packets, remove the ISP intranet transport protocol label from the header of the data packets to obtain the data packets between the node and the server; capture the four-tuple digest in the data packets, which includes: source IP, destination server IP, protocol, and destination server port, and write the four-tuple digest into memory. Step b: Connect to the second port of the BTE component of the SAP system to export the data mirror of the SAP internal Radius link, obtain the billing stop request message of Radius. The billing stop request message includes AVP. Identify the AVP, obtain the msisdn assigned to the terminal, establish the correspondence between the terminal msisdn and the source IP, and store it in the database. Step c: Determine whether the storage size of the quadruple digest written in memory has reached the preset memory limit. If yes, package the quadruple digest and send it to the database; otherwise, return to step a. Step d: Query the mapping between the terminal msisdn and the source IP in the database. Replace the source IP with the terminal msisdn in the quad tuple digest. After the replacement, compare each digest in the quad tuple digest with the row record in the abstract_info_table table in the database. If they are the same, discard the digest. If they are different, write the digest to the abstract_info_table table.

[0016] The aforementioned method for predicting bursts of cellular IoT traffic is characterized by: classifying the row records in the abstract_info_table using preset rules to obtain a data partition table T. F F represents the access type, F≥d, where d is a positive integer, data table T1 represents the global data table, and data table T2-T d This represents a local data table.

[0017] The aforementioned method for predicting bursts of cellular IoT traffic is characterized by the following specific method for providing ISPs with multi-dimensional congestion reduction measures based on Lr: According to the formula calculate , To indicate the number of customers causing congestion, the ISP can choose one of the following measures: Measure 1: Based on the customer code, Migrate customer traffic forwarding to other gateways, or... Implement personalized speed limiting policies for each customer's SIM card; Measure 2: Based on the IP address of the visited server, Forward customer traffic to other gateways, or perform bandwidth channel isolation on customer IPs; Measure 3: Based on customer purchasing and account cancellation activities, Resource optimization for incompletely activated SIM cards and SIM cards selected for cancellation by individual customers; Measure 4: Locate Tc i， ,Will Deploy flexible congestion control strategies in the time period following the corresponding moment. Measure 5, according to The ISP adopts different schemes for the value of .

[0018] Compared with the prior art, the present invention has the following advantages: 1. The present invention has a simple structure, reasonable design, and is convenient to implement and use.

[0019] 2. This invention uses data from SIM card business activities as input, and the Rr-map-2 model outputs the risk level Lr of traffic bursts caused by the predicted incremental access to the destination server due to business activities. This provides quantifiable references for ISPs to take preventative actions in advance, which can reduce the frequency of congestion events caused by traffic bursts and ultimately reduce the losses caused to ISPs, enterprises and platform providers.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a flowchart of the Rr-map-1 model training method of the present invention. Detailed Implementation

[0023] The method of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] For ease of description, spatial relative terms such as "above," "on top of," "on the upper surface of," "above," etc., are used herein to describe the spatial positional relationship of a device or feature as shown in the figures to other devices or features. It should be understood that spatial relative terms are intended to encompass different orientations in use or operation beyond the orientation of the device as described in the figures. For example, if the device in the figures were inverted, a device described as "above" or "on top of" other devices or structures would subsequently be positioned as "below" or "under" other devices or structures. Thus, the exemplary term "above" can include both "above" and "below." The device may also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatial relative descriptions used herein will be interpreted accordingly.

[0028] like Figure 1 As shown, this invention is implemented by a computer and includes the following steps: Step 1: Calculate Rr, where Rr represents the number of redundant SIM cards that the gateway's current remaining bandwidth can allow to be added. Rr=round(max(Bu' / Eu,Bd' / Ed),0), where (Bu',Bd') represents the remaining bandwidth obtained through the BTE component, and (Eu,Ed) represents the evaluated rate pair of the destination server obtained through the BTE component.

[0029] This application combines sampling and processing of gateway logs to statistically and calculate the number of redundant SIM cards that the current remaining bandwidth of the gateway can allow to be added, denoted as Rr.

[0030] The specific method for obtaining (Bu', Bd') in step one is as follows: Obtain the port physical forwarding rate (Bd') of the first port of the BTE component in the SAP system. u B d ), where B u This indicates the maximum uplink bandwidth of the first port of the BTE component in the SAP system. d This represents the maximum downlink bandwidth of the first port of the BTE component in the SAP system; the remaining bandwidth (Bu', Bd') is obtained by subtracting the currently used bandwidth value of the first port from the port physical forwarding rate (Bu, Bd), where Bu' represents the remaining uplink bandwidth and Bd' represents the remaining downlink bandwidth.

[0031] The BTE component has two ports. The first port is the Publish and Subscribe Interface, which provides an SAP data source that can be used by external programs or for data inspection purposes. Mapping internal SAP data to external databases is a common solution for this data mirroring export in one possible implementation: the SAP SLT server. It supports the use of third-party replication tools, such as SymmetricDS, to extract data from SAP systems to SAP or non-SAP targets. External databases include DB2, SAP MaxDB, or Microsoft SQL Server.

[0032] The second port of the BTE component is the Process Interface, which enables data modification and is used to enhance standard business processes.

[0033] This application uses the BTE component to obtain the interface traffic to be predicted of the cellular IoT gateway and the AAA traffic between the gateway and the radius, thereby obtaining a dataset of multiple features of the IoT visited server to be processed.

[0034] The method for obtaining the port physical forwarding rate is as follows: First, the first port of the SAP system's BTE component is the port to be predicted. The LLDP of the first port of the SAP system's BTE component is enabled, i.e., the link layer discovery protocol is enabled. At this time, the port physical forwarding rate will be sent out through LLDP. Since the data of the SAP internal first port is mirrored and exported in step 101, the exported data packets also contain identical LLDP messages. According to the LLDP convention, '0x88cc' is the Ethernet protocol type of the LLDP frame. Based on this characteristic, the LLDP message is filtered out from the raw traffic to reduce the performance overhead of subsequent processing. Finally, the filtered LLDP message is further processed. Since the key parameters carried in this frame are in plaintext, the negotiated rate between the port to be predicted and the peer can be extracted to obtain the port physical forwarding rate, i.e., (B... u B d ).

[0035] It should be noted that, normally, the maximum uplink bandwidth refers to the port's physical forwarding rate; however, if a rate-limiting policy is set for the port, then the maximum uplink bandwidth equals the rate-limited physical forwarding rate; if the administrator configures a forwarding rate for the first port of the BTE component, and the configured forwarding rate value is less than the forwarding rate (B... u B d If the value of ) is obtained, then the forwarding rate (B) u B d The forwarding rate is based on the configured value; if the configured value of the forwarding rate is not less than the forwarding rate (B) u B d If the value of ) is obtained, then the forwarding rate (B) u B d The physical forwarding rate of the port will still be used as the standard.

[0036] The specific method for obtaining (Eu, Ed) in step one is as follows: The mirrored data packet is obtained through the BTE component of the SAP system to obtain a four-tuple digest. An abstract_info_table is constructed based on the four-tuple digest. Row records in the abstract_info_table are classified according to preset rules to obtain a global data table and a local data table. The actual network rate corresponding to the msisdn in the global or local data table is randomly sampled, and the estimated rate pair (Eu, Ed) of the global or local data is calculated, where Eu represents the average uplink rate and Ed represents the average downlink rate.

[0037] The MSISDN is a unique number that can identify a mobile user in the public telephone network switching network numbering scheme. In one possible embodiment, the MSISDN can be the mobile user's number.

[0038] As an example, an MSISDN can include the following components: CC + NDC + SN, where CC (country code) is the country code, NDC (National Destination Code) is the domestic destination address code, which can also be called the network access number, and SN is the user number.

[0039] As a specific example, the MSISDN is: 86+134+11111111, where 86 represents the country code for China, 134 is the domestic destination address code, and 11111111 is the user number. By removing the country code CC from this MSISDN, the mobile station's domestic identification number, which is the mobile phone number, can be obtained.

[0040] In one possible embodiment, the actual network rate corresponding to msisdn in data table T2 is randomly sampled: the gateway's session log function is enabled, the session log is copied to the BTE's log processing module, the number of elements N in data table T2 is calculated, and the number of elements N is stored in data table T2; the BTE's log processing module queries data table T2, filters out matching sessions based on the destination server IP stored in the table, sorts the sessions according to their time, and numbers them starting from 1 to obtain a sequence Ss, Ss=[(1,S1),(2,S2),...,(n,Sn)]. Finally, sessions are sampled with a step size N, and the duration Pj, upload bytes Ubj, and download bytes Dbj of the session are extracted from the sequence Ss according to their numbers. The uplink and downlink rates of the sampled sessions are calculated respectively, and the average uplink and downlink rates of all sampled values ​​are calculated, converted to Mbps, and accurate to two decimal places to obtain (Eu,Ed) of the local data in data table T2.

[0041] In another possible embodiment, the actual network rate corresponding to msisdn in data table T1 is randomly sampled: the gateway's session log function is enabled, the session log is copied to the BTE's log processing module, the number of elements N in data table T1 is calculated, and the number of elements N is stored in data table T1; the BTE's log processing module queries data table T1, filters out matching sessions based on the destination server IP stored in the table, sorts the sessions according to their time, and numbers them starting from 1 to obtain a sequence Ss, Ss=[(1,S1),(2,S2),...,(n,Sn)]. Finally, sessions are sampled with a step size N, and the duration Pj, upload bytes Ubj, and download bytes Dbj of the session are extracted from the sequence Ss according to their numbers. The uplink and downlink rates of the sampled sessions are calculated respectively, and the average uplink and downlink rates of all sampled values ​​are calculated, converted to Mbps, and accurate to two decimal places to obtain the global data (Eu,Ed).

[0042] The row records in the abstract_info_table are classified according to preset rules, dividing them into global data tables and local data tables. This application randomly samples the actual network rates of the global data tables and / or local data tables, and calculates the estimated rate pairs (Eu, Ed) of the global data tables and / or local data tables based on the randomly sampled log session data. The estimated rate pairs (Eu, Ed) of the local data tables can be obtained without statistical analysis of all log sessions, reducing the performance overhead of the module in the calculation process. Secondly, the step size of N takes into account both the number of mssisdn sets and the possible number of sessions, and the calculation results of sets with different N values ​​will not show significant deviations.

[0043] In one possible embodiment, the method for obtaining the abstract_info_table is as follows: Step a: Start SAP, connect to the first port of the SAP system's BTE component, mirror and export the data from the first port of SAP to obtain data packets, remove the ISP intranet transport protocol label from the header of the data packets to obtain the data packets between the node and the server; capture the four-tuple digest in the data packets, which includes: source IP, destination server IP, protocol, and destination server port, and write the four-tuple digest into memory. The intranet transport protocol uses TCP / IP. TCP / IP supports multi-layer encapsulation, meaning a single data packet may contain multiple protocols due to these encapsulation layers. In this application, when stripping labels, each layer's protocol is concatenated using the ">" symbol, and the resulting string is used as the protocol for the entire data packet. For example, if a raw ICMP data packet is encapsulated by GRE on a forwarding link, the captured protocol string should be recorded as 'ip>gre>ip>icmp'.

[0044] Step b: Connect to the second port of the BTE component of the SAP system to export the data mirror of the SAP internal Radius link, obtain the billing stop request message of Radius. The billing stop request message includes AVP. Identify the AVP, obtain the msisdn assigned to the terminal, establish the correspondence between the terminal msisdn and the source IP, and store it in the database. It should be noted that the mapping between the terminal's MSISDN and the source IP is established, forming the msisdn_ip_table. The msisdn_ip_table is stored in the database. When the terminal's SIM card goes offline, the MSISDN allocated to the terminal is released, and its corresponding IP will be confirmed in the subsequent account-stop message. Then the database will delete this record.

[0045] This application directly obtains the mapping between the terminal's MSISDN and the source IP from the mirrored traffic, instead of obtaining it from the RADIUS processing events. The advantages of this approach are: First, RADIUS uses the plaintext UDP protocol, which can be directly read and processed without considering potential event processing delays, ensuring the most timely updates to the mapping between the terminal's MSISDN and the source IP. Second, when a terminal attaches to the network, in multi-attachment scenarios, the MSISDN assigned to the same SIM card may result in two different IPs due to two dial-ups. If the interval between the two dial-ups is too short for RADIUS to process the data, the mapping obtained from RADIUS events using traditional methods may be inaccurate. This application effectively avoids this situation.

[0046] Step c: Determine whether the storage size of the quadruple digest written in memory has reached the preset memory limit. If yes, package the quadruple digest and send it to the database; otherwise, return to step a. Step d: Query the mapping between the terminal msisdn and the source IP in the database. Replace the source IP with the terminal msisdn in the quad tuple digest. After the replacement, compare each digest in the quad tuple digest with the row record in the abstract_info_table table in the database. If they are the same, discard the digest. If they are different, write the digest to the abstract_info_table table.

[0047] In one possible embodiment, the row records in the abstract_info_table are classified according to preset rules to obtain a data partition table T. F F represents the access type, F≥d, where d is a positive integer, data table T1 represents the global data table, and data table T2-T d This represents a local data table.

[0048] The method for classifying records in the abstract_info_table table using preset rules is as follows: Dataset 1, accessing the destination server IP of the MSISDN, the data is partitioned into a table named msisdn_dstip_table; There are two types of datasets, one for accessing the destination server IP and protocol of the MSISDN, and the data is partitioned into a table named msisdn_dstprotocol_table; The dataset is divided into three categories, accessing the MSISDN server with the IP address and port number of the destination server. The data is partitioned into a table named msisdn_dstport_table. According to the aforementioned preset rules, the row records in the abstract_info_table are divided into three categories. As can be seen, the scope of dataset 1 includes dataset 2 and dataset 3. Dataset 1 represents global data, while dataset 2 and dataset 3 represent two types of local data. Thus, the records in the abstract_info_table are divided into global data and two types of local data. This can reduce the error of a single data point in the prediction results. On the other hand, the prediction results based on the protocol and the port of the destination server relative to the destination server IP are more refined, making it easier for the ISP network to take more specific actions based on the prediction results.

[0049] Step 2: Retrieve the customer code and MSISDN relationship data table C j The basic structure of the Rr-map model, Rr-map-1, is constructed as follows: Rr-map-1 = [Num, ip, Rr, Mr], where Num represents the number of IoT SIM cards used by a customer to access a target server IP, ip represents the destination server IP, and Mr = |C j ∩M2ASet|,M r Represents relational data table C j The number of elements in the intersection of the msisdn in the target server and the msisdn in the global or local data table of the destination server is given by relational data table C. j This is a data table showing the relationship between a customer's unique code and a SIM card, where j represents the customer's unique code.

[0050] In the EC (Internet Protocol version control) IoT SIM card management platform system, a web API interface is developed. This interface contains a program that retrieves the customer's unique code and the set of MSIDs corresponding to the SIM cards owned and used by the customer from the EC database, forming a relational data table C. j The subscript j represents the customer's unique code. Once the computer establishes communication with the IoT SIM card management platform EC, it can obtain the relational data table C. j , relational data table C j A database stored in a computer.

[0051] Information on customer SIM cards used is obtained from the ISP's IoT SIM card management platform. The analysis establishes a mapping relationship between the statistical characteristics of customers and their SIM cards used and the Rr value, forming the basic structure of the Rr-map model.

[0052] The basic structure of the Rr-map-1 model only includes the basic parts of Rr-map-2, as shown in the example below: 027836 1.1.1.1 Rr=11250 Mr=3; 027836 2.1.1.1 Rr=9378 Mr=1; 027836 3.1.1.1 Rr=24772 Mr=1; 027836 4.1.1.1 Rr=61053 Mr=1; 027836 6.1.1.1 Rr=67821 Mr=0; 027837 2.1.1.1 Rr=67821 Mr=0.

[0053] Step 3: Build a communication network for mutual access between f BTE components in the SAP system. After the gateway is put into network service, the q-th BTE component periodically sends the Rr-map-1 model data to the p-th BTE component, where 1≤q≤f-1 and q is a positive integer. The p-th BTE component updates its stored Rr-map-1 model according to the received Rr-map-1 model data, thus completing the training of the Rr-map-1 model.

[0054] Generally, in the initial stage of gateway deployment for network services, since the number of IoT SIM cards connected is small and the gateway bandwidth resources are sufficient, gateway congestion caused by traffic surges is almost non-existent. Therefore, BTE can be started in the early stage of gateway deployment for network services, and the accuracy of Rr can be improved by increasing the amount of data and time.

[0055] In one possible embodiment, such as Figure 2 As shown, the specific method for training the Rr-map-1 model is as follows: Step 301: Construct a communication network that supports mutual access between f BTE components in the SAP system, and elect the p-th BTE component as the centralized data storage module.

[0056] It should be noted that the centralized data storage module is confirmed through an election. The election method is as follows: each BTE component periodically sends the number of bytes of storage space occupied by its current data storage module's database to other BTEs. If the p-th BTE component has the largest number of bytes, then the p-th BTE component is elected as the centralized data storage module. This minimizes the overhead when most BTE databases synchronize with the centralized data storage module's database.

[0057] Step 302: After the gateway puts the network service into operation, the q-th BTE component periodically sends Rr-map-1 model data to the p-th BTE component, where 1≤q≤f-1 and q is a positive integer. Step 303: The p-th BTE component receives the Rr-map-1 model data sent by the q-th BTE component and merges the received Rr-map-1 model data. If the Rr-map-1 model data already exists in the Rr-map-1 model data of the p-th BTE component, then return to step one. The p-th BTE component recalculates Rr and Mr and sends the recalculated Rr and Mr back to the q-th BTE component; otherwise, the merged row records are updated in the Rr-map-1 model data of the p-th BTE component. Step 304: The q-th BTE component receives the recalculated Rr and Mr, which are used to update the Rr-map-1 model data in its database; Step 305: Iterate through f-1 BTE components and repeat steps 302-304 to complete the training of the Rr-map-1 model.

[0058] By enabling communication between f BTE components in the SAP system and using cross-training between BTEs, the amount of data and training time involved are increased, thereby improving the accuracy of the Rr value and the accuracy of prediction.

[0059] Step 4: For the record creation time in the Rr-map-1 model, determine the continuity of time using a step size Δt, and remove discontinuous data from the Rr-map-1 model. Then, associate risk level data with each record in the basic structure of the Rr-map model to obtain the final structure of the Rr-map model, Rr-map-2, where Rr-map-2 = [Num, ip, Rr, Mr, Tc]. i ,Lr], where Tc i =(t i Rr i , t i+1 Rr i+1 ), Rr i Let t represent the i-th Rr, where Rr represents the number of redundant SIM cards that the gateway's current remaining bandwidth can allow to be added. i This indicates that the prediction module received Rr i The moment, t i+1 Indicates the relationship with t i In consecutive moments, Lr represents the risk level of the incremental access to the destination server to be predicted.

[0060] An example of the Rr-map-2 model is as follows: [027836 1.1.1.1 Rr=11250 Mr=3Tc i 1).

[0061] Tc i Example as follows: Tc i=[((1683504000,7672,1683504480,8890),(1683763400,9001,(1683763980,6987)).

[0062] By adding time points to the final structure of the Rr-map model and removing discontinuous data from the Rr-map-1 model, the continuity of Rr in the time distribution is ensured, and Tc is... i As a basis for further determining the time points with high congestion risk, the final structure of the Rr-map model is realized.

[0063] In one possible embodiment, the specific method for removing discontinuous data in the Rr-map-1 model in step four is as follows: Step 401: The prediction module receives each record from the trained Rr-map-1 model and creates a time step t for each record. i , t i Indicates receipt of Rr i At any given time, a storage table Tb is formed, where Tb = [(t1, Rr1), ..., (t...]. i ,Rr i ), ..., (t n ,Rr n )]; Step 402: Process the data in storage table Tb according to Rr i Arrange the data in ascending order, and take the first h data to obtain the storage table Tb'; Step 403: Process the data in storage table Tb' according to t i Arrange the data in ascending order and update the storage table Tb'. Step 404: If in the storage table Tb', t i +Δt <t i+1 Then (t) i ,Rr i Remove from storage table Tb'; Step 405: Traverse the times in storage table Tb', repeat step 404, to obtain storage table Tb'' and set Tc. i Tc is used to store elements that exist at least two consecutive time points. i =(t i Rr i , t i+1 Rr i+1 )t i+1 Indicates the relationship with t i A series of moments.

[0064] In one possible embodiment, the specific method for calculating Lr in step four is as follows: ,in P represents the increment to be predicted.

[0065] It should be noted that when Lr=1, it means there is no congestion and no risk; when Lr=2, it means there is low risk of congestion; when Lr=3, it means there is low to medium risk of congestion; when Lr=4, it means there is high risk of congestion; when Lr=5, it means there is high risk of congestion; and when Lr=6, it means that congestion has occurred.

[0066] The specific method for obtaining the increment P to be predicted is as follows: The BTE component periodically calls the order information of the EC platform, extracts the msisdn list, and identifies three types of work orders: purchase and account cancellation. The mirror data package is obtained through the BTE component of the SAP system to obtain a four-tuple summary. The abstract_info_table is constructed based on the four-tuple summary. The row records in the abstract_info_table are classified according to preset rules to obtain a global data table and a local data table. The intersection of the row records in the global data table or the local data table of the msisdn list is calculated, and the number of intersection elements W is calculated. If the work order type is account closure, P = -W; if the work order type is purchasing, P = W.

[0067] Step 5: Collect SIM card operation activities and obtain order information. Input the order information into the Rr-map-2 model. The Rr-map-2 model outputs the risk level Lr of the traffic burst caused by the predicted incremental access to the destination server due to the operation activities. This means that sudden bursts of cellular IoT traffic are considered to cause congestion. <e<6。

[0068] The system synchronously acquires customer procurement and account cancellation operation activities from the IoT SIM card management platform EC. After processing the activity information, it inputs it into the Rr-map-2 model. The Rr-map-2 model outputs the prediction of gateway burst bandwidth based on the operation activities.

[0069] Step 6: Update Δt, return to Step 4, and update the Rr-map-2 model. It should be noted that the value of Δt is taken from the beginning to the end according to the array (300s, 250s, 200s, 150s, 100s, 50s).

[0070] Updating Δt allows for the prediction of the duration of events at different risk levels.

[0071] Step 7: When traffic surges, based on the multi-dimensional congestion reduction measures provided by Lr to the ISP, the ISP implements the congestion reduction measures.

[0072] The specific methods for Lr to provide ISPs with multi-dimensional congestion reduction measures are as follows: According to the formula calculate , To indicate the number of customers causing congestion, the ISP can choose one of the following measures: Measure 1: Based on the customer code, Migrate customer traffic forwarding to other gateways, or... Personalized rate-limiting policies are implemented for each customer's in-use SIM card to reduce the impact of traffic congestion caused by a small number of customers on the overall customer base. Measure 2: Based on the IP address of the visited server, Forwarding customer traffic to other gateways or implementing bandwidth channel isolation for customer IPs can reduce the reputational impact of traffic congestion on specific IPs on third-party platforms. Measure 3: Based on customer purchasing and account cancellation activities, Resource optimization will be implemented for incomplete or unactivated SIM cards of individual customers and SIM cards that have been cancelled, in order to reduce the blind allocation of ISP resources. Measure 4: Locate Tc i ,Will Deploy flexible congestion control strategies in the time period following the corresponding moment; Tc i It further describes the possible timing and duration of the risk, thereby improving the ISP's return on investment.

[0073] Measure 5, according to The ISP adopts different schemes for the value of .

[0074] The above description is merely an embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for predicting bursts of traffic in cellular IoT, characterized in that, Includes the following steps: Step 1: Calculate Rr, where Rr represents the number of redundant SIM cards that the gateway's current remaining bandwidth can allow to be added. Rr=round(max(Bu' / Eu,Bd' / Ed),0), where (Bu',Bd') represents the remaining bandwidth obtained through the BTE component, and (Eu,Ed) represents the evaluated rate pair of the destination server obtained through the BTE component. Step 2: Retrieve the customer code and MSISDN relationship data table C j The basic structure of the Rr-map model, Rr-map-1, is constructed as follows: Rr-map-1 = [Num, ip, Rr, Mr], where Num represents the number of IoT SIM cards used by a customer to access a target server IP, ip represents the destination server IP, and Mr = |C j ∩M2ASet|,M r Represents relational data table C j The number of elements in the intersection of the `msisdn` in the target server and the `msisdn` in the global or local data table of the destination server, where `M2Aset` is the `msisdn` in the global or local data table of the destination server, and `relational data table C` is the relational data table. j This is a data table showing the relationship between a customer's unique code and a SIM card, where j represents the customer's unique code. Step 3: Build a communication network for mutual access between f BTE components in the SAP system. After the gateway is put into network service, the q-th BTE component periodically sends the Rr-map-1 model data to the p-th BTE component, 1≤q≤f-1, where q is a positive integer. The p-th BTE component updates the Rr-map-1 model in its storage according to the received Rr-map-1 model data, thus completing the training of the Rr-map-1 model. Step 4: For the record creation time in the Rr-map-1 model, determine the continuity of time using a step size Δt, and remove discontinuous data from the Rr-map-1 model. Then, associate risk level data with each record in the basic structure of the Rr-map model to obtain the final structure of the Rr-map model, Rr-map-2, where Rr-map-2 = [Num, ip, Rr, Mr, Tc]. i ,Lr], where Tc i =(t i Rr i , t i+1 Rr i+1 ), Rr i Let t represent the i-th Rr, where Rr represents the number of redundant SIM cards that the gateway's current remaining bandwidth can allow to be added. i This indicates that the prediction module received Rr i The moment, t i+1 Indicates the relationship with t i In consecutive time intervals, Lr represents the risk level of the incremental access to the destination server to be predicted; Step 5: Collect SIM card operation activities and obtain order information. Input the order information into the Rr-map-2 model. The Rr-map-2 model outputs the risk level Lr of the traffic burst caused by the predicted incremental access to the destination server due to the operation activities. This means that sudden bursts of cellular IoT traffic are considered to cause congestion. <e<6; Step 6: Update Δt, return to Step 4, and update the Rr-map-2 model; Step 7: When traffic surges, based on the multi-dimensional congestion reduction measures provided by Lr to the ISP, the ISP implements the congestion reduction measures.

2. The method for predicting bursts of cellular IoT traffic according to claim 1, characterized in that: The specific method for training the Rr-map-1 model is as follows: Step 301: Construct a communication network that supports mutual access between f BTE components in the SAP system, and elect the p-th BTE component as the centralized data storage module. Step 302: After the gateway puts the network service into operation, the q-th BTE component periodically sends Rr-map-1 model data to the p-th BTE component, where 1≤q≤f-1 and q is a positive integer. Step 303: The p-th BTE component receives the Rr-map-1 model data sent by the q-th BTE component and merges the received Rr-map-1 model data. If the Rr-map-1 model data already exists in the Rr-map-1 model data of the p-th BTE component, then return to step one. The p-th BTE component recalculates Rr and Mr and sends the recalculated Rr and Mr back to the q-th BTE component; otherwise, the merged row records are updated in the Rr-map-1 model data of the p-th BTE component. Step 304: The q-th BTE component receives the recalculated Rr and Mr, which are used to update the Rr-map-1 model data in its database; Step 305: Iterate through f-1 BTE components and repeat steps 302-304 to complete the training of the Rr-map-1 model.

3. The method for predicting bursts of cellular IoT traffic according to claim 1, characterized in that: The specific method for removing discontinuous data in the Rr-map-1 model in step four is as follows: Step 401: The prediction module receives each record from the trained Rr-map-1 model and creates a time step t for each record. i , t i Indicates receipt of Rr i At any given time, a storage table Tb is formed, where Tb = [(t1, Rr1), ..., (t...]. i ,Rr i ), ..., (t n ,Rr n )]; Step 402: Process the data in storage table Tb according to Rr i Arrange the data in ascending order, and take the first h data to obtain the storage table Tb'; Step 403: Process the data in storage table Tb' according to t i Arrange the data in ascending order and update the storage table Tb'. Step 404: If in the storage table Tb', t i +Δt <t i+1 Then (t) i ,Rr i Remove from storage table Tb'; Step 405: Traverse the times in storage table Tb', repeat step 404, to obtain storage table Tb'' and set Tc. i Tc is used to store elements that exist at least two consecutive time points. i =(t i Rr i , t i+1 Rr i+1 )t i+1 Indicates the relationship with t i A series of moments.

4. The method for predicting bursts of cellular IoT traffic according to claim 1, characterized in that: The specific method for calculating Lr in step four is as follows: ,in P represents the increment to be predicted.

5. A method for predicting bursts of cellular IoT traffic according to claim 2, characterized in that: The specific method for obtaining the increment P to be predicted is as follows: The BTE component periodically calls the order information of the EC platform, extracts the msisdn list, and identifies three types of work orders: purchase and account cancellation. The mirror data package is obtained through the BTE component of the SAP system to obtain a four-tuple summary. The abstract_info_table is constructed based on the four-tuple summary. The row records in the abstract_info_table are classified according to preset rules to obtain a global data table and a local data table. The intersection of the row records in the global data table or the local data table of the msisdn list is calculated, and the number of intersection elements W is calculated. If the work order type is account closure, P = -W; if the work order type is purchasing, P = W.

6. The method for predicting bursts of cellular IoT traffic according to claim 1, characterized in that: The specific method for obtaining (Bu', Bd') in step one is as follows: Obtain the port physical forwarding rate (B / T) of the first port of the BTE component in the SAP system. u B d ), where B u This indicates the maximum uplink bandwidth of the first port of the BTE component in the SAP system. d This indicates the maximum downlink bandwidth of the first port of the BTE component in the SAP system; The remaining bandwidth (Bu', Bd') is obtained by subtracting the currently used bandwidth of the first port from the port's physical forwarding rate (Bu, Bd), where Bu' represents the remaining uplink bandwidth and Bd' represents the remaining downlink bandwidth.

7. A method for predicting bursts of cellular IoT traffic according to claim 1, characterized in that: The specific method for obtaining (Eu, Ed) in step one is as follows: The mirror data package is obtained through the BTE component of the SAP system to obtain a four-tuple summary. The abstract_info_table is constructed based on the four-tuple summary. The row records in the abstract_info_table are classified according to preset rules to obtain a global data table and a local data table. Randomly sample the actual network rate corresponding to msisdn in the global or local data table, and calculate the estimated rate pair (Eu, Ed) of the global or local data, where Eu represents the average uplink rate and Ed represents the average downlink rate.

8. A method for predicting bursts of cellular IoT traffic according to claim 5 or 7, characterized in that: The method for obtaining the abstract_info_table table is as follows: Step a: Start SAP, connect to the first port of the SAP system's BTE component, mirror and export the data from the first port inside SAP, obtain data packets, remove the ISP intranet transport protocol label from the header of the data packets, and obtain the data packets between the node and the server. Capture the four-tuple digest from the data packet. The four-tuple digest includes: source IP, destination server IP, protocol, and destination server port. Write the four-tuple digest into memory. Step b: Connect to the second port of the BTE component of the SAP system to export the data mirror of the SAP internal Radius link, obtain the billing stop request message of Radius. The billing stop request message includes AVP. Identify the AVP, obtain the msisdn assigned to the terminal, establish the correspondence between the terminal msisdn and the source IP, and store it in the database. Step c: Determine whether the storage size of the quadruple digest written in memory has reached the preset memory limit. If yes, package the quadruple digest and send it to the database; otherwise, return to step a. Step d: Query the mapping between the terminal msisdn and the source IP in the database. Replace the source IP with the terminal msisdn in the quad tuple digest. After the replacement, compare each digest in the quad tuple digest with the row record in the abstract_info_table table in the database. If they are the same, discard the digest. If they are different, write the digest to the abstract_info_table table.

9. A method for predicting bursts of cellular IoT traffic according to claim 8, characterized in that: The rows in the abstract_info_table are categorized using preset rules to obtain a data partition table T. F F represents the access type, F≥d, where d is a positive integer, data table T1 represents the global data table, and data table T2-T d This represents a local data table.

10. A method for predicting bursts of cellular IoT traffic according to claim 1, characterized in that: The specific methods for Lr to provide ISPs with multi-dimensional congestion reduction measures are as follows: According to the formula calculate , To indicate the number of customers causing congestion, the ISP can choose one of the following measures: Measure 1: Based on the customer code, Migrate customer traffic forwarding to other gateways, or... Implement personalized speed limiting policies for each customer's SIM card; Measure 2: Based on the IP address of the visited server, Forward customer traffic to other gateways, or perform bandwidth channel isolation on customer IPs; Measure 3: Based on customer purchasing and account cancellation activities, Resource optimization for incompletely activated SIM cards and SIM cards selected for cancellation by individual customers; Measure 4: Locate Tc i ,Will Deploy flexible congestion control strategies in the time period following the corresponding moment. Measure 5, according to The ISP adopts different schemes for the value of .

Images