A network congestion control method and system based on mobile communication signaling analysis

Abstract

The application discloses a network congestion control method and system based on mobile communication signaling analysis, and can capture early micro signs of signaling congestion by collecting and analyzing control plane signaling data in real time, correlating and generating signaling event sequences of user granularity. On this basis, network load characteristics and user behavior characteristics are extracted, and the congestion prediction model is processed by pre-training, which can generate congestion risk indication information before the formation of signaling impact, and realize active prediction. When there is a congestion risk, the minimum regional overall signaling blocking rate and the highest load balancing degree are taken as the joint optimization target, the multi-objective optimization is carried out by using the swarm intelligence optimization algorithm, the global optimal differentiated control strategy is automatically solved, and the adjacent area chain congestion caused by single cell regulation is avoided. Therefore, the application solves the technical problem that the existing method lacks means for early perception of congestion signs from real-time signaling flow, cannot actively dredge before the formation of signaling impact, and delays the best control opportunity.

Description

Technical Field

[0001] This invention relates to the field of mobile communication technology, and more specifically to a network congestion control method and system based on mobile communication signaling analysis. Background Technology

[0002] In mobile communication networks, the explosive growth of services such as high-definition video and the Internet of Things (IoT) generates a massive amount of control signaling due to the frequent registration, service requests, and mobility management of a large number of terminals. This can easily overload the signaling processing nodes of the core network and base stations, leading to signaling congestion and severely affecting service establishment latency and user experience.

[0003] Traditional congestion control involves the network management system periodically collecting performance statistics for each cell, such as the number of connected users under radio resource control (RRR) and processor utilization, and comparing them with pre-set fixed thresholds. Once a metric exceeds its limit, it triggers adjustments to unified access prohibition parameters or initiates user migration. This process is entirely based on outdated statistical averages, representing a passive approach of addressing congestion only after it has occurred. Because changes in performance metrics lag behind actual signaling surges, signaling backlogs often occur before metrics worsen, making early prevention impossible. Furthermore, coarse-grained counters struggle to characterize different user behaviors and signaling burst patterns, leading to a one-size-fits-all control strategy. Therefore, existing methods lack the means to detect congestion signs in real-time signaling streams, failing to proactively alleviate congestion before signaling surges occur, thus delaying optimal control opportunities. Summary of the Invention

[0004] The purpose of this invention is to provide a network congestion control method and system based on mobile communication signaling analysis, in order to solve the technical problem that existing methods lack the means to detect congestion signs in advance from real-time signaling streams, and cannot actively alleviate congestion before signaling surges occur, thus delaying the best control opportunity.

[0005] The technical solution of this invention is implemented as follows:

[0006] In a first aspect, the present invention provides a network congestion control method based on mobile communication signaling analysis, comprising:

[0007] Control plane signaling data is collected in real time from the interface between the core network and the radio access network;

[0008] The signaling data is parsed, and a user signaling event sequence is generated based on the user identifier.

[0009] Based on the user signaling event sequence, network load characteristics and user behavior characteristics of the target cell are extracted according to a preset time window;

[0010] The network load characteristics and user behavior characteristics are input into a pre-trained congestion prediction model to obtain congestion risk indication information for a future preset time period.

[0011] When the congestion risk indication information determines that there is a congestion risk, the congestion control strategy is optimized by using a swarm intelligence optimization algorithm to find the multi-objective optimization of the congestion control strategy with the lowest overall regional signaling blocking rate and the highest load balance as the joint optimization objectives, and then generating and executing the globally optimal active congestion control strategy.

[0012] A further technical solution is that the acquisition of control plane signaling data and the generation of user signaling event sequences specifically include:

[0013] Collects radio resource control connection establishment requests, initial context establishment requests, service requests, handover requests, periodic registration update requests, and protocol data unit session establishment requests;

[0014] For each signaling message, extract the user's permanent identifier, temporary identifier, cell identifier, request type, timestamp, and request result;

[0015] A user state machine is established using the user's permanent identifier or temporary identifier as the primary key. The registration, service request, session establishment, switching, and release processes of the same user are concatenated into a complete state transition sequence in chronological order. Signaling timeout events and signaling rejection events are marked as exceptions to form the user signaling event sequence carrying exception tags.

[0016] A further technical solution is that the extraction of network load characteristics and user behavior characteristics of the target cell specifically includes:

[0017] Using the preset time window as the statistical granularity, the mean and variance of the number of various signaling requests, signaling processing success rate, average signaling processing latency, and signaling arrival interval per unit time are calculated for the target cell to constitute the network load characteristics.

[0018] The number of unique user identifiers within the preset time window for the target cell is counted as the number of active users. The group movement speed is estimated based on the number of cells traversed by users per unit time and divided into low-speed, medium-speed, and high-speed levels. The ratio of the signaling arrival rate of the current window to the average signaling arrival rate of multiple historical windows is calculated as the burst index. The proportion of IoT terminals and mobile terminals is counted based on user terminal type or behavior pattern to constitute the user behavior characteristics.

[0019] A further technical solution is that the congestion prediction model is a temporal neural network based on a multi-head self-attention mechanism, and the acquisition of congestion risk indication information specifically includes:

[0020] The network load features and user behavior features from multiple consecutive time windows are combined into a three-dimensional feature tensor, which is then input into the temporal neural network. The dependency relationship between different time windows is extracted through a self-attention layer, and the congestion probability of multiple future preset time windows is output.

[0021] The training process of the temporal neural network includes: acquiring historical signaling data, marking periods when the signaling processing delay exceeds the core network protection timer threshold and periods when signaling failures or timeouts occur consecutively as congestion positive samples, and marking the remaining periods as normal negative samples, and performing supervised training using binary cross-entropy as the loss function.

[0022] A further technical solution is that the swarm intelligence optimization algorithm is used to perform multi-objective optimization of the congestion control strategy, specifically including:

[0023] The adjustable control parameters of each target cell currently at risk of congestion are encoded as particle position vectors. The adjustable control parameters include the prohibition factor and prohibition duration of unified access control, cell reselection offset, and signaling flow control delay response ratio.

[0024] Construct a fitness function, which is a weighted sum of the predicted congestion probabilities of all risky cells, the normalized value of the total throughput of the regional system, and the weighted sum of the user experience rate compliance rate.

[0025] With the goal of maximizing the fitness function, the position and velocity of each particle are iteratively updated using the discrete particle swarm optimization algorithm. After a preset number of iterations, the position vector of the globally optimal particle is output, and the optimal combination of control parameters for each target cell is decoded.

[0026] A further technical solution is that the generation and execution of the globally optimal active congestion control strategy specifically includes:

[0027] Based on the optimal control parameter combinations obtained from the optimization process for each target cell, at least one of the following types of operation instructions is generated:

[0028] Send a radio resource control layer configuration update message containing the access prohibition factor and prohibition duration to the target base station;

[0029] Send a system information block update message containing cell reselection offset to the target base station; send a signaling flow control policy to the access and mobility management function network element, instructing to delay or extend the response interval for specific priority service requests from the target cell;

[0030] Send a load balancing instruction to the session management function network element or the policy control function network element to trigger the migration of some users of non-guaranteed rate services in the target cell to a neighboring cell;

[0031] The operation instructions are sent to the corresponding wireless access network element or core network element for execution through the operation and maintenance management interface or service interface.

[0032] Secondly, the present invention provides a network congestion control system based on mobile communication signaling analysis, comprising:

[0033] The signaling acquisition module is used to acquire control plane signaling data in real time from the interface between the core network and the radio access network;

[0034] The signaling parsing and association module is used to parse the signaling data and generate a user signaling event sequence based on the user identifier;

[0035] The feature extraction module is used to extract network load features and user behavior features of the target cell within a set time window based on the user signaling event sequence.

[0036] The congestion prediction module has a pre-trained congestion prediction model built in, which is used to generate congestion risk indication information for a future preset time period based on the network load characteristics and the user behavior characteristics.

[0037] The strategy decision module is used to optimize multiple candidate control strategy combinations and generate an active congestion control strategy when the congestion risk indication information determines that there is a congestion risk.

[0038] The strategy execution module is used to distribute the optimized congestion control strategy to the target network element for execution.

[0039] A further technical solution is that the swarm intelligence optimization algorithm includes a discrete particle swarm optimization algorithm. The strategy decision module encodes the access prohibition factor, cell reselection offset, and signaling flow control delay ratio of each cell to be controlled into the position vector of the particle, and searches for the globally optimal parameter combination by iteratively updating the velocity and position of the particle.

[0040] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a network congestion control method based on mobile communication signaling analysis.

[0041] Fourthly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a network congestion control method based on mobile communication signaling analysis.

[0042] The beneficial effects of this invention are as follows:

[0043] By collecting and analyzing control plane signaling data in real time and generating user-granular signaling event sequences, early signs of congestion can be captured from the microscopic details of signaling interactions. Based on this, network load characteristics and user behavior characteristics are extracted and processed by a pre-trained congestion prediction model to generate congestion risk indication information before signaling surges occur, enabling proactive prediction. When congestion risk exists, a swarm intelligence optimization algorithm is used to find the optimal global differentiated control strategy by employing the joint optimization objectives of minimizing the overall regional signaling congestion rate and maximizing load balancing. This avoids cascading congestion in neighboring cells caused by single-cell control. Therefore, this invention solves the technical problem of existing methods lacking the means to detect congestion signs in real-time signaling streams, failing to proactively alleviate congestion before signaling surges occur, and thus delaying the optimal control timing. Attached Figure Description

[0044] Figure 1 A flowchart illustrating the steps of a network congestion control method based on mobile communication signaling analysis, provided in this embodiment of the invention;

[0045] Figure 2 A block diagram of a network congestion control system based on mobile communication signaling analysis is provided for an embodiment of the present invention;

[0046] Figure 3 This is a structural block diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0047] To better understand the technical content of this invention, specific embodiments are provided below, and the invention will be further described in conjunction with the accompanying drawings.

[0048] This invention uses a standalone (SA) network scenario of a fifth-generation mobile communication system as an example for illustration, but the solution of this invention is also applicable to fourth-generation long-term evolution (LTE) networks or non-standalone (NSA) network architectures. In this example, the network contains multiple cells, each of which is provided by a corresponding base station. The base station connects to the access and mobility management function (AM) network element through the N2 interface and to the user plane function (MPF) network element through the N3 interface. The AMAM network element interacts with core network elements such as the session management function (SEM) network element and the policy control function (PCC) network element through service-oriented interfaces.

[0049] Example 1

[0050] See Figure 1 This invention provides a network congestion control method based on mobile communication signaling analysis.

[0051] Step 101: Collect control plane signaling data in real time from the interface between the core network and the radio access network;

[0052] The signaling acquisition module is deployed at the N2 interface between the core network and the radio access network, and acquires control plane signaling messages of the NGAP protocol in real time through optical splitting or port mirroring. The acquired signaling types include: initial user equipment messages corresponding to radio resource control connection establishment requests, initial context establishment request messages, service request messages, handover request messages based on the N2 interface, periodic registration update request messages, and protocol data unit session establishment request messages.

[0053] Meanwhile, the signaling acquisition module can also obtain signaling event notifications from the service interface of the access and mobility management function network element through a subscription method, as a supplementary data source.

[0054] Step 102: Parse the signaling data and generate a user signaling event sequence based on the user identifier;

[0055] The signaling parsing and association module performs in-depth parsing on each collected signaling message and extracts the following key fields: permanent user identifier such as SUPI, temporary user identifier such as 5G-GUTI, serving cell identifier such as NCI, request type such as initial registration, service request or handover, message timestamp accurate to milliseconds, and request result code such as success, timeout or rejection.

[0056] Using the aforementioned permanent or temporary user identifier as the primary key, an independent state machine is established for each user. The state machine tracks the user's signaling flow state. The initial state is idle; upon receiving an initial registration request, it enters the registration process state; after registration is complete, it enters the registered state; upon receiving a service request, it enters the service process state; and upon receiving a release message, it returns to the idle state. Through the state machine, the scattered signaling messages for the same user are concatenated into a complete state transition sequence in chronological order.

[0057] For example, for a user with the identifier 12345, the signaling event sequence is as follows: At time T1, a registration request is initiated in cell A, and the result is successful; at time T2, a protocol data unit session establishment request is initiated in cell A, and the result is successful; at time T3, a service request is initiated in cell A, and the result is successful; at time T5, a handover request is initiated to cell B, and the result is successful; at time T8, a release is initiated in cell B, and the result is normal. The entire process forms a complete user activity trajectory.

[0058] Simultaneously, abnormal events are specifically tagged. If the response waiting time for a signaling message exceeds 90% of the core network protection timer threshold, it is marked as a timeout event; if the signaling message returns a rejection result code, it is marked as a rejection event. These abnormal tags are stored along with the event sequence to provide a labeling basis for subsequent model training.

[0059] Step 103: Based on the user signaling event sequence, extract the network load characteristics and user behavior characteristics of the target cell according to a preset time window;

[0060] The feature extraction module uses a preset time window as the statistical granularity. In this example, the time window is set to one second, and features are calculated for each cell separately.

[0061] Regarding network load characteristics, taking cell A within a one-second window as an example, the following indicators are statistically analyzed: the total number of various signaling requests, for example, 30 initial user equipment messages, 15 service request messages, and 8 handover request messages received within this window; the signaling processing success rate, which is the number of successful signaling messages divided by the total number of signaling messages within this window, for example, 25 successful messages out of a total of 30, resulting in a success rate of 83.3%; the average signaling processing latency, which is the average time difference between the request and response for each signaling message within this window, for example, an average latency of 42 milliseconds; and the mean and variance of the signaling arrival interval, which are statistically analyzed based on the timestamp differences between two adjacent signaling messages, reflecting the stability of signaling arrival.

[0062] Regarding user behavior characteristics, the following indicators are statistically analyzed: Number of unique users within a cell, i.e., the number of different user identifiers appearing within a statistical window; for example, if there are 87 different users in cell A within a given second. Group movement speed estimate, calculated by counting the number of cells a user traverses per unit of time. If a user traverses more than three cells within one minute, they are marked as a high-speed user; if they traverse one or two cells, they are a medium-speed user; and if they do not traverse any cells, they are a low-speed user. The percentage of users at each speed level is calculated. Burst index, comparing the signaling arrival rate of the current window with the average signaling arrival rate of the previous ten historical windows. If the current arrival rate is 50 messages per second and the historical average is 20 messages per second, the burst index is 2.5, indicating a sudden increase in signaling. Terminal type distribution, determining the terminal category based on the user's device type identifier or behavioral pattern; for example, distinguishing between IoT terminals and smartphone terminals based on device capability information in the registration request, and calculating the percentage of each type of terminal.

[0063] Step 104: Input the network load characteristics and user behavior characteristics into the pre-trained congestion prediction model to obtain congestion risk indication information for the future preset time period;

[0064] The congestion prediction module deploys a pre-trained temporal neural network model based on a multi-head self-attention mechanism. During the inference phase, the module assembles the network load features and user behavior features of the target cell over the most recent sixty time windows into a three-dimensional feature tensor. The three dimensions of this tensor represent the number of time steps, the number of feature categories, and the one-dimensional data depth, respectively. This tensor is input into the temporal neural network, first passing through a location encoding layer to add temporal and location information to each time window. Then, a multi-layer multi-head self-attention mechanism extracts the dependencies between different time windows. For example, the model can learn a strong correlation between the signaling burstiness of the current window and the number of handover requests in the previous three windows. Finally, a fully connected layer and activation function output the congestion probability for multiple future time windows; for example, predicting a congestion probability of 0.85 for the first second, 0.72 for the second second, and 0.31 for the fifth second.

[0065] The training process of the temporal neural network is as follows: Signaling data from multiple consecutive days in history is acquired. Periods where the signaling processing latency exceeds 90% of the core network protection timer threshold are labeled as positive congestion samples. Periods with three or more consecutive signaling failures or timeouts within one minute are also labeled as positive congestion samples. All other normal periods are labeled as negative samples. The labeled samples are then divided into training and validation sets according to time windows. Using binary cross-entropy as the loss function, gradient descent optimization algorithm is employed for multiple rounds of iterative training until the model converges.

[0066] During real-time operation, the congestion prediction module compares the output congestion probability with a preset threshold of 0.75. If the predicted probability exceeds 0.75, a congestion risk indication is generated, indicating that there is a congestion risk in the corresponding cell in the future, and this information is sent to the policy decision module.

[0067] Step 105: When the congestion risk indication information determines that there is a congestion risk, the congestion control strategy is optimized by using a swarm intelligence optimization algorithm to find the multi-objective optimization of the congestion control strategy with the lowest overall regional signaling blocking rate and the highest load balance as the joint optimization objectives, and the globally optimal active congestion control strategy is generated and executed.

[0068] Upon receiving the congestion risk indication information, the strategy decision module assumes that three cells, A, B, and C, simultaneously face congestion risk. The module then initiates a discrete particle swarm optimization algorithm to perform multi-objective optimization.

[0069] First, the adjustable control parameters of each risk cell are encoded as particle position vectors. In this example, the adjustable parameters of each cell include three dimensions: the access prohibition factor of unified access control, ranging from 0.1 to 1.0; the cell reselection offset, ranging from 0 to 6 dB; and the delay response ratio of signaling flow control, ranging from 1 to 5 times. The parameter combination of the three cells constitutes a nine-dimensional position vector. During initialization, thirty particles are randomly generated, each particle representing a set of candidate control parameter combinations.

[0070] Secondly, a fitness function is constructed to evaluate the performance of each set of parameters. The fitness function is calculated as follows: fitness equals the sum of the first, second, and third terms. The first term is the negative of the weighted sum of the predicted congestion probabilities of all risky cells, which is calculated by multiplying the predicted congestion probabilities of cells A, B, and C by their respective service weights, summing the results, and then taking the negative value. The lower the congestion probability, the higher the score for this term. The second term is the normalized value of the total throughput of the regional system, which is obtained by dividing the current total throughput of the region by the historical maximum throughput to obtain a normalized value between zero and one. The third term is the user experience rate compliance rate, which is the proportion of users in the statistical region whose user experience rate exceeds a preset threshold. The three terms are assigned different weights and then summed to obtain a comprehensive fitness value.

[0071] Then, the algorithm enters the iterative optimization process. In each iteration, each particle updates its velocity and position based on its own historical best position and the global best position of the swarm. The velocity update formula introduces inertia weights and a learning factor, allowing particles to maintain a certain inertia in the search space while moving towards both the individual optimal and the global optimal direction. After the position is updated, the fitness value is calculated for each particle's new position. If it is better than the particle's historical best fitness, the individual best position is updated; if it is better than the swarm's global best fitness, the global best position is updated. This example sets the number of iterations to one hundred, or terminates the iteration early when the global best fitness value does not improve significantly for twenty consecutive rounds.

[0072] After iteration, the position vector of the globally optimal particle is output, and this nine-dimensional vector is decoded to restore the optimal control parameter combination for each target cell. For example, the final optimization result might be: cell A has an injunction factor of 0.5, an injunction duration of four seconds, a cell reselection offset increased by 3 dB, and a delay response multiplier of 2; cell B has an injunction factor of 0.3, an injunction duration of two seconds, a cell reselection offset increased by 1 dB, and a delay response multiplier of 1; and cell C has an injunction factor of 0.7, an injunction duration of six seconds, a cell reselection offset increased by 5 dB, and a delay response multiplier of 3. This set of parameters achieves a balance between the lowest regional signaling congestion rate and the highest load balancing.

[0073] The strategy execution module transforms the optimal combination of control parameters obtained through optimization into specific operation instructions, which are then issued and executed through the corresponding interface.

[0074] For example, for cell A, the access prohibition factor is 0.5 and the prohibition duration is four seconds. The policy execution module constructs a radio resource control layer configuration update message. The message body includes a unified access control parameter information unit, sets the prohibition factor field to 0.5 and the prohibition duration field to four seconds, and sends it to the base station of cell A through the operation and maintenance management interface. After receiving it, the base station immediately updates the corresponding parameters in the system information block and broadcasts it to the user equipment within the coverage area.

[0075] For cell A, the cell reselection offset is increased by 3 dB. The policy execution module constructs a system information block update message, modifies the offset field in the cell reselection configuration information unit to 3 dB, and sends it to the corresponding base station for execution.

[0076] For signaling flow control policies, the policy execution module constructs a policy request message and sends it to the service interface of the access and mobility management function network element. The message indicates a delayed response to low-priority periodic registration update requests from cell A, with a delay factor of two. Upon receiving the request, the access and mobility management function network element does not process it immediately but postpones the response time to twice the original timer duration.

[0077] For load balancing commands, the policy execution module sends a request to the session management function network element or the policy control function network element, instructing users of non-guaranteed rate services within cell A to migrate to a neighboring cell. In practice, the core network smoothly migrates eligible users from cell A to cell B with lower load by modifying the user plane path or triggering a handover procedure, ensuring uninterrupted service during the migration process.

[0078] After each operation command is issued, the system continuously monitors the changes in the congestion probability of each cell. If the congestion probability falls below the safety threshold, the original parameter configuration is gradually restored.

[0079] Example 2

[0080] See Figure 2This invention provides a network congestion control system based on mobile communication signaling analysis, comprising: a signaling acquisition module for real-time acquisition of control plane signaling data from the interface between the core network and the radio access network; a signaling parsing and association module for parsing signaling data and generating user signaling event sequences based on user identifiers; a feature extraction module for extracting network load features and user behavior features of the target cell within a set time window based on the user signaling event sequences; a congestion prediction module, internally equipped with a pre-trained congestion prediction model for generating congestion risk indication information for a future preset time period based on network load features and user behavior features; a strategy decision module for optimizing multiple candidate control strategy combinations using a swarm intelligence optimization algorithm, with the goal of maximizing the overall network utility, to generate an active congestion control strategy when the congestion risk indication information indicates the existence of congestion risk; and a strategy execution module for distributing the optimized congestion control strategy to the target network element for execution. Among them, the swarm intelligence optimization algorithm includes the discrete particle swarm algorithm. The strategy decision module encodes the access prohibition factor, cell reselection offset and signaling flow control delay ratio of each cell to be controlled into the position vector of the particle, and searches for the global optimal parameter combination by iteratively updating the velocity and position of the particle.

[0081] The system modules of this invention can be deployed on network edge computing platforms or centralized network intelligent platforms. The signaling acquisition module employs high-speed data packet capture technology; the signaling parsing and correlation module performs real-time processing based on a stream computing framework; the feature extraction module writes the calculation results into a time-series database for the prediction module to use; the congestion prediction module provides an inference interface in the form of microservices; the policy decision-making module embeds a particle swarm optimization engine; and the policy execution module has the ability to adapt to base station operation and maintenance management interfaces and core network service interfaces. The modules communicate asynchronously and decoupledly through message queues to ensure the real-time performance and reliability of the entire control loop.

[0082] Example 3

[0083] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the network congestion control method based on mobile communication signaling analysis as described in any of the above embodiments.

[0084] Please see Figure 3 , Figure 3 This is a structural block diagram of an electronic device provided in an embodiment of the present invention.

[0085] An electronic device according to an embodiment of the present invention includes: a memory 201 and a processor 202. The memory 201 stores a computer program. When the computer program is executed by the processor 202, the processor 202 executes a network congestion control method based on mobile communication signaling analysis as described in any of the above embodiments.

[0086] Memory 201 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Memory 201 has storage space 203 for program code 213 for performing any of the method steps described above. For example, storage space 203 for program code may include individual program codes 213 for implementing the various steps in the methods described above. This program code can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When run by a computing processing device, this code causes the computing processing device to perform the various steps in the methods described above. This program code can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When this code is run by a computing device, it causes the device to perform the various steps in the style transfer-based method for generating realistic interior design renderings described above.

[0087] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A network congestion control method based on mobile communication signaling analysis, characterized in that, include: Control plane signaling data is collected in real time from the interface between the core network and the radio access network; The signaling data is parsed, and a user signaling event sequence is generated based on the user identifier. Based on the user signaling event sequence, network load characteristics and user behavior characteristics of the target cell are extracted according to a preset time window; The network load characteristics and user behavior characteristics are input into a pre-trained congestion prediction model to obtain congestion risk indication information for a future preset time period. When the congestion risk indication information determines that there is a congestion risk, the congestion control strategy is optimized by using a swarm intelligence optimization algorithm to find the multi-objective optimization of the congestion control strategy with the lowest overall regional signaling blocking rate and the highest load balance as the joint optimization objectives, and then generating and executing the globally optimal active congestion control strategy.

2. The network congestion control method based on mobile communication signaling analysis according to claim 1, characterized in that, The acquisition of control plane signaling data and the generation of user signaling event sequences specifically include: Collects radio resource control connection establishment requests, initial context establishment requests, service requests, handover requests, periodic registration update requests, and protocol data unit session establishment requests; For each signaling message, extract the user's permanent identifier, temporary identifier, cell identifier, request type, timestamp, and request result; A user state machine is established using the user's permanent identifier or temporary identifier as the primary key. The registration, service request, session establishment, switching, and release processes of the same user are concatenated into a complete state transition sequence in chronological order. Signaling timeout events and signaling rejection events are marked as exceptions to form the user signaling event sequence carrying exception tags.

3. The network congestion control method based on mobile communication signaling analysis according to claim 1, characterized in that, The extraction of network load characteristics and user behavior characteristics of the target cell specifically includes: Using the preset time window as the statistical granularity, the mean and variance of the number of various signaling requests, signaling processing success rate, average signaling processing latency, and signaling arrival interval per unit time are calculated for the target cell to constitute the network load characteristics. The number of unique user identifiers within the preset time window for the target cell is counted as the number of active users. The group movement speed is estimated based on the number of cells traversed by users per unit time and divided into low-speed, medium-speed, and high-speed levels. The ratio of the signaling arrival rate of the current window to the average signaling arrival rate of multiple historical windows is calculated as the burst index. The proportion of IoT terminals and mobile terminals is counted based on user terminal type or behavior pattern to constitute the user behavior characteristics.

4. The network congestion control method based on mobile communication signaling analysis according to claim 1, characterized in that, The congestion prediction model is a temporal neural network based on a multi-head self-attention mechanism, and the congestion risk indication information obtained specifically includes: The network load features and user behavior features from multiple consecutive time windows are combined into a three-dimensional feature tensor, which is then input into the temporal neural network. The dependency relationship between different time windows is extracted through a self-attention layer, and the congestion probability of multiple future preset time windows is output. The training process of the temporal neural network includes: acquiring historical signaling data, marking periods when the signaling processing delay exceeds the core network protection timer threshold and periods when signaling failures or timeouts occur consecutively as congestion positive samples, and marking the remaining periods as normal negative samples, and performing supervised training using binary cross-entropy as the loss function.

5. The network congestion control method based on mobile communication signaling analysis according to claim 1, characterized in that, The method of employing a swarm intelligence optimization algorithm to perform multi-objective optimization of the congestion control strategy specifically includes: The adjustable control parameters of each target cell currently at risk of congestion are encoded as particle position vectors. The adjustable control parameters include the prohibition factor and prohibition duration of unified access control, cell reselection offset, and signaling flow control delay response ratio. Construct a fitness function, which is a weighted sum of the predicted congestion probabilities of all risky cells, the normalized value of the total throughput of the regional system, and the weighted sum of the user experience rate compliance rate. With the goal of maximizing the fitness function, the position and velocity of each particle are iteratively updated using the discrete particle swarm optimization algorithm. After a preset number of iterations, the position vector of the globally optimal particle is output, and the optimal combination of control parameters for each target cell is decoded.

6. The network congestion control method based on mobile communication signaling analysis according to claim 1, characterized in that, The generation and execution of the globally optimal proactive congestion control strategy specifically includes: Based on the optimal control parameter combinations obtained from the optimization process for each target cell, at least one of the following types of operation instructions is generated: Send a radio resource control layer configuration update message containing the access prohibition factor and prohibition duration to the target base station; Send a system information block update message containing cell reselection offset to the target base station; send a signaling flow control policy to the access and mobility management function network element, instructing to delay or extend the response interval for specific priority service requests from the target cell; Send a load balancing instruction to the session management function network element or the policy control function network element to trigger the migration of some users of non-guaranteed rate services in the target cell to a neighboring cell; The operation instructions are sent to the corresponding wireless access network element or core network element for execution through the operation and maintenance management interface or service interface.

7. A network congestion control system based on mobile communication signaling analysis, characterized in that, include: The signaling acquisition module is used to acquire control plane signaling data in real time from the interface between the core network and the radio access network; The signaling parsing and association module is used to parse the signaling data and generate a user signaling event sequence based on the user identifier; The feature extraction module is used to extract network load features and user behavior features of the target cell within a set time window based on the user signaling event sequence. The congestion prediction module has a pre-trained congestion prediction model built in, which is used to generate congestion risk indication information for a future preset time period based on the network load characteristics and the user behavior characteristics. The strategy decision module is used to optimize multiple candidate control strategy combinations and generate an active congestion control strategy when the congestion risk indication information determines that there is a congestion risk. The strategy execution module is used to distribute the optimized congestion control strategy to the target network element for execution.

8. The network congestion control system based on mobile communication signaling analysis according to claim 7, characterized in that, The swarm intelligence optimization algorithm includes the discrete particle swarm algorithm. The strategy decision module encodes the access prohibition factor, cell reselection offset and signaling flow control delay ratio of each cell to be controlled into the position vector of the particle, and searches for the globally optimal parameter combination by iteratively updating the velocity and position of the particle.

9. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the network congestion control method based on mobile communication signaling analysis as described in any one of claims 1 to 6.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the network congestion control method based on mobile communication signaling analysis as described in any one of claims 1 to 6.

Images