An end-to-end network slice sla real-time monitoring method and system

Abstract

The present application relates to network slice service level agreement (SLA) real-time monitoring method and system, and solves the technical problem that the prior art cannot identify the macro queuing deterioration trend caused by the burst traffic in advance without excessive consumption of network resources. The method comprises: determining a discrete coordinate point set according to the base station downlink buffer occupancy rate and the core network traffic instantaneous burst coefficient of each sampling period; determining the convex hull envelope area of the discrete coordinate point set, and determining the center metric value of the data point distribution center of the discrete coordinate point set; determining the congestion early warning evaluation factor according to the coordinate radix of the latest state point, the convex hull envelope area, the congestion aggravation direction projection component, and the center metric value of the data point distribution center; and when the congestion early warning evaluation factor exceeds the preset alarm threshold, adjusting the resource configuration parameters of the low priority slice of the same base station according to the measurement result. The present application is used in the network slice monitoring scene.

Description

Technical Field

[0001] This invention relates to the field of network slicing service level agreement (SLA) assurance technology, specifically to an end-to-end network slicing SLA real-time monitoring method and system. Background Technology

[0002] In the context of 5G network slicing operation and maintenance, ensuring end-to-end latency of Service Level Agreements (SLAs) for services such as industrial control has become one of the core requirements of network operation. With the large-scale deployment of network slicing technology, different services have increasingly stringent requirements for quality of service, especially in high-dynamic traffic environments where agreed-upon latency, bandwidth, and other key performance indicators must be continuously met. This places higher demands on network monitoring and resource scheduling capabilities.

[0003] In specific application scenarios, control command traffic often exhibits high-intensity bursts, with a large number of packets flooding the core network gateway and the base station downlink within a short period. This can easily exceed the capacity limit of the base station's MAC layer downlink buffer, leading to queuing degradation and service interruptions. To address this queuing accumulation caused by burst traffic, existing network management platforms typically rely on periodically retrieving low-frequency one-dimensional performance indicators such as the average interface throughput or statistical variance to monitor network element status. Some solutions also employ high-frequency deployment of bidirectional active measurement protocol probes to capture congestion. However, existing technologies struggle to balance monitoring accuracy and resource overhead in practical applications, resulting in poor overall SLA performance. On one hand, low-frequency statistical indicators cannot accurately depict the state span experienced by the base station scheduler during rapid switching between idle and high-load periods, easily masking the queuing deterioration trend behind the average. On the other hand, routine high-frequency probe deployment severely strains the limited service backhaul air interface bandwidth and increases the gateway's computing power burden. Therefore, existing technologies struggle to identify macro-level queuing degradation trends caused by burst traffic in advance without excessively consuming network resources. Summary of the Invention

[0004] To address the technical problem that existing methods struggle to identify macroscopic queuing degradation trends caused by intermittent traffic in advance without excessively consuming network resources, the present invention aims to provide an end-to-end network slicing SLA real-time monitoring method and system. The specific technical solution adopted is as follows: In a first aspect, the present invention provides an end-to-end network slice SLA real-time monitoring method, the method comprising: determining the base station downlink buffer occupancy rate and the core network traffic instantaneous burst coefficient for multiple sampling periods within a sliding time window; generating two-dimensional state feature coordinate points and determining a discrete coordinate point set based on the base station downlink buffer occupancy rate and the core network traffic instantaneous burst coefficient for each sampling period; determining the convex hull envelope area of ​​the discrete coordinate point set and determining the center metric of the data point distribution center of the discrete coordinate point set; determining the congestion aggravation direction projection component of the latest state point in the discrete coordinate point set relative to the data point distribution center; the latest state point being the last coordinate point in the time sequence of the discrete coordinate point set; determining a congestion warning assessment factor based on the coordinate base, convex hull envelope area, congestion aggravation direction projection component, and center metric of the data point distribution center of the latest state point; and triggering end-to-end active latency measurement of the target network slice when the congestion warning assessment factor exceeds a preset alarm threshold, and adjusting the resource configuration parameters of the low-priority slice under the same base station according to the measurement results.

[0005] In conjunction with the first aspect mentioned above, in one possible implementation, the method specifically includes: at the end of the time segment of each sampling period, reading the current backlog of bytes and the maximum downlink buffer capacity of the base station; determining the downlink buffer occupancy rate of the base station in each sampling period based on the ratio of the current backlog of bytes to the maximum downlink buffer capacity; within the time segment of each sampling period, reading the total number of incoming bytes of the core network gateway multiple times at preset time intervals to generate a throughput micro-sequence; and determining the instantaneous burst coefficient of core network traffic in each sampling period based on the ratio of the standard deviation to the arithmetic mean of the throughput micro-sequence.

[0006] In conjunction with the first aspect mentioned above, in one possible implementation, the method specifically includes: determining a first-dimensional coordinate value based on the downlink buffer occupancy rate of the base station and the degree of difference between it and a preset benchmark value; the first-dimensional coordinate value is used to quantify the queuing time limit; determining a second-dimensional coordinate value based on the amplified fluctuation intensity of the instantaneous burst coefficient of core network traffic; the second-dimensional coordinate value is used to quantify the burst interference intensity of traffic; combining the first-dimensional coordinate value and the second-dimensional coordinate value to generate two-dimensional state feature coordinate points; arranging the two-dimensional state feature coordinate points generated in each sampling period in chronological order to determine a discrete coordinate point set.

[0007] In conjunction with the first aspect mentioned above, in one possible implementation, the method specifically includes: performing deduplication on the coordinate points in the discrete coordinate point set, determining whether the number of remaining coordinate points after deduplication is less than three, or whether all deduplicated coordinate points are located on the same straight line; when the number of remaining coordinate points after deduplication is less than three, or all deduplicated coordinate points are located on the same straight line, setting the convex hull area to zero; when the number of remaining coordinate points after deduplication is greater than or equal to three, and all deduplicated coordinate points are not located on the same straight line, determining the outer vertex sequence of the discrete coordinate point set using a preset convex hull boundary extraction method; and determining the convex hull area based on the spatial distribution characteristics of each vertex in the outer vertex sequence.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, the method specifically includes: determining the center value of the horizontal coordinate based on the distribution of the horizontal coordinates of all coordinate points in the discrete coordinate point set; determining the center value of the vertical coordinate based on the distribution of the vertical coordinates of all coordinate points in the discrete coordinate point set; combining the center value of the horizontal coordinate and the center value of the vertical coordinate as the data point distribution center; and determining the center measurement value based on the center value of the horizontal coordinate and the center value of the vertical coordinate.

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, the method specifically includes: generating an instantaneous directional offset vector based on the positional offset between the coordinates of the latest state point and the coordinates of the data point distribution center; obtaining a preset congestion reference direction vector; the preset congestion reference direction vector points in a direction where both the horizontal and vertical coordinates change towards the direction of congestion deterioration; and determining the projection component of the congestion aggravation direction based on the degree of directional correlation between the instantaneous directional offset vector and the preset congestion reference direction vector.

[0010] In conjunction with the first aspect mentioned above, in one possible implementation, the method specifically includes: determining a directional penalty activation coefficient based on the numerical sign characteristics of the projection component in the direction of congestion aggravation; the directional penalty activation coefficient is used to activate the incremental penalty mechanism when the projection component in the direction of congestion aggravation indicates a trend of congestion deterioration; determining a smoothing adjustment factor based on the center metric of the data point distribution center and a preset bias constant; the smoothing adjustment factor is used to adjust the intensity of the incremental penalty according to the current network load baseline level; determining a fluctuation increment coefficient based on the smoothing adjustment factor, the directional penalty activation coefficient, and the convex hull envelope area; the fluctuation increment coefficient is used to characterize the additional penalty weight jointly determined by the historical state fluctuation range, the current trend direction, and the load baseline level; when the convex hull envelope area is zero or close to zero, the coordinate base of the latest state point is used as the congestion warning assessment factor; when the convex hull envelope area is greater than zero, a congestion warning assessment factor is generated based on the coordinate base of the latest state point and the fluctuation increment coefficient.

[0011] In conjunction with the first aspect mentioned above, in one possible implementation, the method specifically includes: responding to a congestion warning assessment factor exceeding a preset alarm threshold, parsing the network slice selection auxiliary information of the target network slice, querying and obtaining the associated user equipment Internet Protocol address based on the network slice selection auxiliary information; issuing an active measurement protocol activation command to the core network user plane function gateway to trigger the sending of probe packets to the user equipment Internet Protocol address; and obtaining the end-to-end latency measurement data returned by the core network user plane function gateway.

[0012] In conjunction with the first aspect mentioned above, in one possible implementation, the method specifically includes: comparing the measured end-to-end latency data with a preset red-line latency threshold; when the measured end-to-end latency data reaches or exceeds the red-line latency threshold, determining the resource limiting weight based on the congestion warning assessment factor; sending an adjustment instruction to the core network policy control function gateway; the adjustment instruction carrying the resource limiting weight, used to instruct the reduction of resource guarantee configuration for other slices that share the same base station with the target network slice and have a lower priority than the target network slice.

[0013] In a second aspect, the present invention provides an end-to-end network slicing SLA real-time monitoring system for implementing the end-to-end network slicing SLA real-time monitoring method of the first aspect above; the system includes: a determination module for determining the base station downlink buffer occupancy rate and the core network traffic instantaneous burst coefficient for multiple sampling periods within a sliding time window; a coordinate generation module for generating two-dimensional state feature coordinate points and determining a discrete coordinate point set based on the base station downlink buffer occupancy rate and the core network traffic instantaneous burst coefficient for each sampling period; and an area and center determination module for determining the convex hull area of ​​the discrete coordinate point set and the center of the data point distribution of the discrete coordinate point set. The system includes: a centrality metric module; a projection component determination module, used to determine the congestion aggravation direction projection component of the latest state point in the discrete coordinate point set relative to the center of the data point distribution; the latest state point is the last coordinate point in the time sequence of the discrete coordinate point set; a congestion warning assessment factor determination module, used to determine the congestion warning assessment factor based on the coordinate base, convex hull area, congestion aggravation direction projection component, and centrality metric of the data point distribution center of the latest state point; and an intervention execution module, used to trigger end-to-end active latency measurement of the target network slice when the congestion warning assessment factor exceeds a preset alarm threshold, and to adjust the resource configuration parameters of low-priority slices under the same base station based on the measurement results.

[0014] The present invention has the following beneficial effects: This invention, by simultaneously collecting the downlink buffer occupancy rate of base stations and the burst coefficient of core network traffic, constructs a two-dimensional state coordinate point set and extracts the convex hull envelope area. This transforms queuing degradation trends, which are difficult to capture by traditional low-frequency one-dimensional statistical indicators, into quantifiable geometric features. Based on this, it integrates the coordinate base, envelope area, projection components, and center metric of the data point distribution center of the latest state points to generate a congestion warning assessment factor. This effectively characterizes the state span experienced by the base station scheduler during rapid switching between idle and high load. When the assessment factor exceeds the alarm threshold, it triggers end-to-end proactive latency measurement and adjusts the resource configuration of low-priority slices on demand. This allows for early identification of macro-level queuing degradation trends caused by burst traffic without increasing the network-wide normalized measurement bandwidth overhead. It achieves a closed-loop on-demand reconfiguration of cross-slice air interface resources, significantly improving the real-time performance and resource utilization efficiency of network slice SLA guarantees. This solves the technical problem of existing methods failing to identify macro-level queuing degradation trends caused by burst traffic in advance without excessively consuming network resources. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating an end-to-end network slicing SLA real-time monitoring method according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the system architecture of an end-to-end network slicing SLA real-time monitoring system provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an end-to-end network slicing SLA real-time monitoring method and system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of the end-to-end network slicing SLA real-time monitoring method and system provided by the present invention.

[0020] Please see Figure 1 The illustration shows a schematic flowchart of an end-to-end network slice SLA real-time monitoring method according to an embodiment of the present invention. The method includes the following S101-S106, which will be described in detail below.

[0021] S101. Determine the base station downlink buffer occupancy rate for multiple sampling periods within the sliding time window, as well as the instantaneous burst coefficient of core network traffic.

[0022] In one possible implementation, for each sub-sampling period within the sliding monitoring window, the system synchronously collects two types of raw feature data on both the base station side and the core network side: the base station side reads the ratio of the current backlog of bytes to the maximum downlink buffer capacity to determine the downlink buffer occupancy rate; the core network side reads the total number of incoming bytes from the gateway multiple times at fixed small time intervals within the same period, generates a throughput micro-sequence, and calculates the ratio of the standard deviation of the micro-sequence to its arithmetic mean to determine the instantaneous burst coefficient of traffic.

[0023] For example, the above data acquisition can be achieved by combining hardware counter reading with statistical coefficient of variation calculation. During the determination of the downlink buffer occupancy rate, the system truncates values ​​approaching full capacity to prevent the denominator from being zero during subsequent mapping. In determining the instantaneous burst coefficient of traffic, if the arithmetic mean is zero, the burst coefficient is directly set to zero; otherwise, the ratio of the standard deviation to the mean is used as the burst coefficient. The granularity of the sub-sampling period and the micro-sampling frequency are directly related to the accuracy and time alignment requirements of data acquisition: the finer the time window division and the shorter the micro-sampling interval, the more sensitive the acquired burst coefficient is to characterizing the burst characteristics of traffic bursts. For typical burst traffic scenarios, the total duration of the sliding window can be set to 10 seconds, divided into 20 equal sub-sampling periods, and high-frequency micro-sampling can be performed at 50-millisecond intervals within each 500-millisecond period to capture sub-second details of traffic bursts.

[0024] S102. Based on the base station downlink buffer occupancy rate and the instantaneous burst coefficient of core network traffic in each sampling period, generate two-dimensional state feature coordinate points and determine the discrete coordinate point set.

[0025] In one possible implementation, nonlinear coordinate mapping is performed on each set of time-aligned base station downlink buffer occupancy rates and core network traffic burst coefficients acquired within the sliding monitoring window. Specifically, the base station downlink buffer occupancy rate and its difference from a preset benchmark value are used as the basis to calculate a first-dimensional coordinate value. This coordinate value is used to quantify the queuing time limit, and its physical meaning is approximately equivalent to a normalized queuing delay estimate. This coordinate value increases sharply when the buffer occupancy rate approaches full capacity. Simultaneously, the fluctuation intensity of the core network traffic burst coefficient is nonlinearly amplified to calculate a second-dimensional coordinate value. This coordinate value is used to quantify the traffic burst interference intensity, reflecting the impact of the uneven temporal distribution of packet injection into the gateway on the base station scheduling order. The first-dimensional coordinate value and the second-dimensional coordinate value calculated in each sampling period are combined to generate a two-dimensional state feature coordinate point. The system arranges the coordinate points generated in all sampling periods in chronological order to construct a discrete coordinate point set.

[0026] For example, the above mapping can be achieved using a nonlinear stretching transformation and a quadratic enhancement transformation based on queuing theory. For the buffer occupancy rate, the ratio obtained by dividing it by 1 and subtracting itself is used as the first-dimensional coordinate value. This ensures that when the occupancy rate increases slightly from 0.9 to 0.99, the first-dimensional coordinate value increases dramatically from 9 to 99, thus achieving a quantitative stretching and amplification of the high-load queuing time limit. For the traffic burst coefficient, its square value is calculated first, and then a fixed reference value (such as 1) is added as the second-dimensional coordinate value. The squaring operation amplifies the quadratic interference of burst unevenness on the scheduling order, while adding the reference value ensures that the coordinate value is always greater than or equal to 1, avoiding zero values ​​in subsequent calculations. After completing the coordinate mapping for all sampling periods, the coordinate points are arranged in time index order to form a discrete coordinate point set. The distribution range and convex hull shape of this point set directly reflect the congestion state fluctuation trajectory experienced by the base station within the sliding window.

[0027] S103. Determine the area of ​​the convex hull of the discrete coordinate point set, and determine the center metric of the data point distribution center of the discrete coordinate point set.

[0028] In one possible implementation, the validity of the constructed discrete coordinate point set is determined. If the number of valid points in the discrete coordinate point set is insufficient or all points are collinear, the convex hull area is directly set to zero. If the number of valid points is sufficient and they are not collinear, a pre-defined convex hull boundary extraction method is used to determine the sequence of outer vertices of the discrete coordinate point set, and the convex hull area is calculated based on the spatial distribution characteristics of each vertex in the outer vertex sequence. Simultaneously, the abscissa center value is determined based on the abscissa distribution of all coordinate points in the discrete coordinate point set, and the ordinate center value is determined based on the ordinate distribution. The abscissa center value and the ordinate center value are combined as the data point distribution center, and the center metric is determined based on their product.

[0029] S104. Determine the projection component of the latest state point in the discrete coordinate point set relative to the congestion aggravation direction of the data point distribution center.

[0030] The latest state point is the last coordinate point in the time sequence of the discrete coordinate point set.

[0031] In one possible implementation, the latest coordinate point in time sequence is extracted from the discrete coordinate point set as the latest state point, representing the network congestion state at the current sampling moment. Simultaneously, the calculated data point distribution center is used as the average congestion baseline within this time window. The positional offset between the latest state point coordinates and the data point distribution center coordinates is calculated to generate an instantaneous directional offset vector, reflecting the instantaneous direction of the system load state deviating from the average baseline at the current moment. A preset congestion reference direction vector is obtained, pointing in a direction where both the horizontal and vertical coordinates change towards congestion worsening (i.e., the direction where queuing time limits increase and traffic burst interference intensity increases). Subsequently, the directional correlation between the instantaneous directional offset vector and the preset congestion reference direction vector is calculated, and this correlation is determined as the projection component of the congestion aggravation direction. When this projection component is greater than zero, it indicates that the network state at the latest moment is evolving towards congestion worsening; when the projection component is less than or equal to zero, it indicates that network congestion is easing or oscillating laterally.

[0032] S105. Determine the congestion warning assessment factor based on the coordinate base of the latest state point, the area of ​​the convex hull envelope, the projection component of the congestion aggravation direction, and the center metric of the data point distribution center.

[0033] In one possible implementation, firstly, a directional penalty activation coefficient is determined based on the numerical sign characteristics of the projection component in the direction of congestion aggravation. This coefficient is used to activate the incremental penalty mechanism when the projection component indicates a worsening congestion trend, while having no incremental impact on the direction of mitigation or oscillation. Simultaneously, a smoothing adjustment factor is determined based on the center metric of the data point distribution center and a bias constant (e.g., the bias constant can be 0.001, its function being to prevent division by zero errors in the denominator when the center metric approaches zero). This factor is used to adaptively adjust the strength of the incremental penalty according to the current network load baseline level: when the load baseline is low, the smoothing adjustment factor amplifies the contribution of small fluctuations, increasing sensitivity; when the load baseline is high, the smoothing adjustment factor suppresses the contribution of regular fluctuations, reducing false alarms. Then, a fluctuation increment coefficient is determined based on the smoothing adjustment factor, the directional penalty activation coefficient, and the convex hull envelope area. This coefficient comprehensively characterizes the additional penalty weight jointly determined by the historical state fluctuation range, the current trend direction, and the load baseline level. Finally, a congestion early warning assessment factor is generated based on the coordinate base of the latest state point and the fluctuation increment coefficient. In particular, when the area of ​​the convex hull is zero or close to zero, the coordinate base of the latest state point is directly used as the congestion warning assessment factor to ensure that the assessment factor does not become invalid under extreme congestion conditions.

[0034] S106. When the congestion warning assessment factor exceeds the preset alarm threshold, trigger end-to-end active latency measurement of the target network slice, and adjust the resource configuration parameters of the low-priority slice under the same base station according to the measurement results.

[0035] In one possible implementation, the congestion warning assessment factor calculated within the current sliding window is compared with a preset alarm threshold. For example, the preset alarm threshold can be 10, calibrated based on simulation or measured data. Under normal steady-state conditions, the congestion warning assessment factors for multiple sliding windows are statistically analyzed, and their mean plus 3 to 5 times the standard deviation is used as a benchmark. Alternatively, in intermittent traffic testing, the threshold is selected when the assessment factor grows non-linearly and the corresponding end-to-end latency approaches the red line threshold (e.g., 50 milliseconds). In this example, a value of 10 is used to achieve a balance between sensitivity and stability. If the congestion warning assessment factor is less than or equal to the preset alarm threshold, the current network queuing fluctuation is determined to be within the normal range, the current processing round ends, and the system waits for data inflow in the next sampling period. If the congestion warning assessment factor is greater than the preset alarm threshold, a significant queuing deterioration trend has been identified within the current sliding window, and end-to-end active latency measurement is immediately triggered. Specifically, the network slice selection auxiliary information of the target network slice is parsed, the associated user equipment Internet Protocol address is obtained by querying, an active measurement protocol activation command is sent to the core network user plane function gateway, a probe packet is sent to the user equipment Internet Protocol address, and the end-to-end latency measurement data is obtained.

[0036] After obtaining the end-to-end latency measurement data, it is compared with a preset red-line latency threshold. For example, the red-line latency threshold can be set according to the service level agreement requirements of the target network slice, such as 50 milliseconds. If the end-to-end latency measurement data is less than the red-line latency threshold, it is determined that the current queuing backlog has not yet caused a substantial service violation, and the active measurement operation is cancelled, returning to the regular feature data collection process. If the end-to-end latency measurement data reaches or exceeds the red-line latency threshold, it is determined that a congestion violation has indeed occurred. Based on the congestion warning assessment factor, the resource limiting weight is determined, and an adjustment instruction is sent to the core network policy control function gateway. This instruction carries the resource limiting weight and is used to instruct the reduction of resource guarantee configurations for other slices sharing the same base station with the target network slice and having a lower priority than the target network slice.

[0037] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment, by simultaneously collecting the downlink buffer occupancy rate of the base station and the core network traffic burst coefficient, constructs a two-dimensional state coordinate point set and extracts the convex hull envelope area, which can transform the queuing deterioration trend, which is difficult to capture by traditional low-frequency one-dimensional statistical indicators, into a quantifiable geometric feature. On this basis, by integrating the coordinate base, envelope area, projection components, and center metric value of the data point distribution center of the latest state point, a congestion warning evaluation factor is generated, which realizes an effective characterization of the state span experienced by the base station scheduler during rapid switching between idle and high load. When the evaluation factor exceeds the alarm threshold, end-to-end active latency measurement is triggered on demand and the resource configuration of low-priority slices is adjusted, thereby identifying the macro-queuing deterioration trend caused by burst traffic in advance without increasing the bandwidth overhead of normal measurement across the entire network, realizing the closed-loop on-demand reconfiguration of cross-slice air interface resources, and significantly improving the real-time performance and resource utilization efficiency of network slice SLA guarantee. This solves the technical problem that existing methods are unable to identify the macro-queuing deterioration trend caused by burst traffic in advance without excessively consuming network resources.

[0038] In one possible implementation, the process of determining the base station downlink buffer occupancy rate and the instantaneous burst coefficient of core network traffic for multiple sampling periods within the sliding time window can be specifically implemented through the following S201-S204, which will be explained in detail below.

[0039] S201. At the end of the time segment of each sampling period, read the current number of accumulated bytes and the maximum downlink buffer capacity of the base station.

[0040] In one possible implementation, for each sub-sampling period divided within the sliding monitoring window, at the end of the time segment of that period, the current backlog of bytes for the base station corresponding to the target network slice and the maximum downlink buffer capacity of that base station are read through the base station's northbound interface.

[0041] For example, if the maximum downlink buffer capacity of the base station is 10 megabytes and the current backlog of bytes is 8 megabytes, then two original values ​​are obtained directly; if the current backlog of bytes is close to the maximum capacity (such as reaching more than 99%), then upper limit truncation can be performed later to prevent the denominator from being zero.

[0042] S202. Determine the base station downlink buffer occupancy rate for each sampling period based on the ratio of the current number of accumulated bytes to the maximum downlink buffer capacity.

[0043] In one possible implementation, for each sampling period, the current number of accumulated bytes and the maximum downlink buffer capacity are read, and the current number of accumulated bytes is divided by the maximum downlink buffer capacity. The resulting ratio is the base station downlink buffer occupancy rate for that sampling period.

[0044] For example, if the maximum downlink buffer capacity is 10 megabytes and the current backlog is 8 megabytes, the buffer occupancy rate is 0.8; if the current backlog is 9.9 megabytes, the occupancy rate is 0.99. To prevent the denominator from being zero during subsequent nonlinear mapping, the upper limit of the occupancy rate can be truncated to 0.99, that is, if the calculated value is greater than 0.99, it will be forcibly set to 0.99.

[0045] S203. Within the time segment of each sampling period, read the total number of incoming bytes from the core network gateway multiple times according to a preset time interval to generate a throughput micro-sequence.

[0046] In one possible implementation, for the same sub-sampling period that is exactly the same as the aforementioned base station sampling period, within the time segment of this period, the total number of incoming bytes to the core network gateway device is read multiple times at a fixed preset time interval. Each read obtains the number of bytes arriving at the gateway within that tiny interval, and these data points are recorded in chronological order to generate a throughput micro-sequence specific to the current sampling period.

[0047] For example, if the subsampling period is 500 milliseconds, the preset time interval can be set to 50 milliseconds. Then, 10 read operations are performed continuously within one subsampling period to obtain 10 inflow byte count values, forming a throughput micro-sequence. Each value in this sequence reflects the intensity of traffic injection within an extremely short time window, which can capture sub-second burst characteristics.

[0048] S204. Determine the instantaneous burst coefficient of core network traffic for each sampling period based on the ratio of the standard deviation to the arithmetic mean of the throughput microsequence.

[0049] In one possible implementation, for each throughput micro-sequence generated in each sampling period, the arithmetic mean and standard deviation of all values ​​in the sequence are calculated. If the arithmetic mean is equal to zero, it indicates that there is no downlink traffic injection into the target network slice in that sampling period, and the instantaneous burst coefficient of the core network traffic is directly determined to be zero; if the arithmetic mean is not equal to zero, the standard deviation is divided by the arithmetic mean, and the resulting ratio is used as the instantaneous burst coefficient of the core network traffic in that sampling period.

[0050] For example, if the throughput micro-sequence is 100, 200, 50, 150, 100 (unit: bytes), with an arithmetic mean of 120, a standard deviation of approximately 53.5, and a ratio of 0.446, then the burst coefficient is 0.446; if the micro-sequence is 0,0,0,0,0,0,0,0,0, with an arithmetic mean of 0, then the burst coefficient is directly set to 0. A larger burst coefficient indicates a more uneven concentration of traffic within the sampling period, resulting in a stronger pulse-like impact on the base station.

[0051] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment defines the specific collection method for the base station downlink buffer occupancy rate and the core network traffic burst coefficient. By reading the number of backlogged bytes and the maximum capacity of the base station at the end of the time segment of each sampling period, and determining the buffer occupancy rate based on the ratio, the static queuing waiting degree of the base station can be accurately reflected. At the same time, within the same sampling period, the total number of inflow bytes of the core network gateway is read multiple times at preset time intervals to generate a throughput micro-sequence, and the ratio of its standard deviation to its arithmetic mean is calculated as the burst coefficient, which can accurately characterize the non-uniformity of the time distribution of packets injected into the gateway. This synchronous collection method under a unified time reference system provides a reliable data foundation for the subsequent construction of a two-dimensional coordinate point set with strict time alignment, avoiding the problem of congestion assessment time misalignment caused by differences in network element sampling frequency.

[0052] In one possible implementation, the process of generating two-dimensional state feature coordinate points and determining the discrete coordinate point set based on the base station downlink buffer occupancy rate and the instantaneous burst coefficient of core network traffic in each sampling period can be specifically implemented through the following S301-S304, which will be explained in detail below.

[0053] S301. Determine the first-dimensional coordinate value based on the downlink buffer occupancy rate of the base station and the degree of difference between it and the preset reference value.

[0054] The first dimension coordinate value is used to quantify the queuing time limit.

[0055] In one possible implementation, a preset baseline value is first set to measure the saturation level of the buffer occupancy rate. For example, the preset baseline value is set to 1, representing a theoretically fully loaded state. The difference between the base station downlink buffer occupancy rate and the preset baseline value is calculated, i.e., the base station downlink buffer occupancy rate is subtracted from the preset baseline value to obtain the remaining idle ratio. Then, the base station downlink buffer occupancy rate is divided by the aforementioned difference, and the resulting ratio is used as the first-dimensional coordinate value. This coordinate value is used to quantify the queuing time limit, and its physical meaning approximates the average waiting time factor in queuing theory.

[0056] Understandably, the nonlinear mapping function (hyperbolic transformation) used in this step accurately characterizes the mathematical relationship between latency and utilization in the queuing system: when the utilization approaches 100%, the queuing latency approaches infinity. By embedding this physical law into the coordinate transformation, the subsequent calculation of the geometric envelope area and distribution center can sensitively reflect the trend of queuing deterioration. Compared to directly using linear ratios, this mapping significantly amplifies the differences in high-load areas, solving the problem that traditional mean indicators change slowly and are difficult to trigger warnings when the buffer is close to full capacity. At the same time, combined with the upper limit truncation process in S201, it ensures that the mapping results are always within a controllable range, guaranteeing the numerical stability of the algorithm.

[0057] S302. Determine the second-dimensional coordinate value based on the amplified result of the fluctuation intensity of the instantaneous burst coefficient of the core network traffic.

[0058] The second dimension coordinate value is used to quantify the intensity of traffic burst interference.

[0059] One possible implementation involves amplifying the fluctuation intensity of the instantaneous burst coefficient of core network traffic. This involves performing a nonlinear enhancement operation on the coefficient to highlight the impact of traffic burst unevenness on base station scheduling. Based on the amplified result, a fixed reference value is further added to ensure that the second-dimensional coordinate value always remains in the positive range, avoiding zero values ​​in subsequent calculations.

[0060] For example, let the instantaneous burst coefficient of core network traffic be... The amplification of wave intensity is achieved by squaring, i.e. Then add the value 1 to get the second dimension coordinate value. The squaring operation further amplifies the large burst coefficient, while adding 1 ensures that the ordinate is always not less than 1, eliminating the risk of zero denominators in subsequent calculations (for example, the product of the x and y coordinates of the data point distribution center may be zero).

[0061] Understandably, this step quantifies the quadratic interference intensity caused by traffic bursts and unevenness on base station scheduling through squaring. Physically, this corresponds to the ratio of traffic variance to the square of the mean, which is proportional to the square of the coefficient of variation. Adding a baseline value of 1 not only prevents zero values ​​but also ensures that the coordinate points have a basic ordinate even in the absence of bursts, maintaining a reasonable distance from the x-axis in geometric space, facilitating convex hull construction and area calculation. This mapping transforms the dynamic traffic characteristics of the core network into ordinate values ​​in geometric space, which, together with the first-dimensional coordinate values, constitute two-dimensional state characteristic coordinate points.

[0062] S303. Combine the first-dimensional coordinate value with the second-dimensional coordinate value to generate two-dimensional state feature coordinate points.

[0063] In one possible implementation, the first-dimensional coordinate value and the second-dimensional coordinate value calculated in the same sampling period are treated as an ordered pair, with the first-dimensional coordinate value as the horizontal axis and the second-dimensional coordinate value as the vertical axis, forming a two-dimensional state feature coordinate point for that sampling period.

[0064] In one possible implementation, before generating the two-dimensional state feature coordinate points, the first-dimensional coordinate values ​​and the second-dimensional coordinate values ​​for all sampling periods within the current sliding window are normalized. For example, maximum / minimum value normalization can be used: for the first-dimensional coordinate values, the minimum value within the window is recorded. and maximum value , each Mapped to This ensures the value range is [0, 1]; the same applies to the second-dimensional coordinate values. If all values ​​within the window... If the denominator is zero due to equality, all normalized values ​​are set to 0.5. This process ensures that the two axes are completely consistent and the convex hull area calculation is not affected by dimensions.

[0065] S304. Arrange the two-dimensional state feature coordinate points generated in each sampling period in chronological order to determine the discrete coordinate point set.

[0066] In one possible implementation, the two-dimensional state feature coordinate points corresponding to all sampling periods within the sliding monitoring window are arranged sequentially according to the time index order (from 1 to N) of their respective sampling periods to form a set of discrete coordinate points with temporal relationships.

[0067] For example, if the sliding window contains 20 sampling periods, and each period generates a coordinate point, then these points are arranged in order of time index 1 to 20 to obtain a discrete coordinate point set containing 20 coordinate points. Each point in this set carries both the congestion state information at that moment and preserves the evolution trajectory of the state over time through the arrangement order.

[0068] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment defines a specific method for mapping the original collected data into two-dimensional state feature coordinate points. By determining the first-dimensional coordinate value based on the difference between the base station downlink buffer occupancy rate and the preset benchmark value, the queuing time limit can be quantified, and nonlinear stretching and amplification can be achieved when the buffer is under high load, making potential congestion features easier to capture. At the same time, by determining the second-dimensional coordinate value based on the amplification result of the fluctuation intensity of the core network traffic burst coefficient, the intensity of traffic burst interference can be quantified, highlighting the quadratic impact of external traffic unevenness on the base station scheduling order. Arranging the coordinate points generated in each sampling period in chronological order to form a discrete coordinate point set provides a structured input for subsequent convex hull area extraction and center calculation, so that the congestion state distribution in the two-dimensional space can intuitively reflect the load span limit experienced by the system within the current time window.

[0069] In one possible implementation, the process of determining the convex hull area of ​​the discrete coordinate point set can be specifically implemented through the following S401-S404, which will be explained in detail below.

[0070] S401. Perform deduplication on the coordinate points in the discrete coordinate point set, and determine whether the number of remaining coordinate points after deduplication is less than three, or whether all deduplicated coordinate points are located on the same straight line.

[0071] In one possible implementation, for the constructed set of discrete coordinate points, a deduplication operation is first performed to remove duplicate points with identical coordinates (i.e., equal x and y coordinates), resulting in a unique set of coordinate points after deduplication. Then, the number of remaining coordinate points after deduplication is counted, and it is determined whether the number is less than three. Simultaneously, by calculating the slope of the line connecting any two points, it is determined whether all deduplicated coordinate points lie on the same straight line.

[0072] S402. When the number of remaining coordinate points after deduplication is less than three, or when all deduplicated coordinate points are located on the same straight line, the area of ​​the convex hull is set to zero.

[0073] In one possible implementation, based on the judgment result of step S401, if the number of remaining coordinate points after deduplication is less than three, or if all coordinate points are located on the same straight line, then it is determined that the current set of discrete coordinate points cannot form a valid polygon. In this case, the convex hull area is directly set to zero, and subsequent convex hull boundary extraction and area calculation operations are not performed.

[0074] Understandably, this step avoids the algorithm throwing exceptions or producing meaningless results when it cannot form a convex hull by setting the envelope area to zero in invalid or degenerate cases. At the same time, the zero area serves as a conditional signal in subsequent exponential fusion, which can activate the extreme value degradation mechanism (i.e., directly using the coordinate base of the latest state point as the evaluation factor), ensuring that effective output can still be produced even in extremely congested conditions (high overlap of data points or dead boundaries) or when samples are insufficient.

[0075] S403. When the number of remaining coordinate points after deduplication is greater than or equal to three, and all deduplicated coordinate points are not located on the same straight line, the outer vertex sequence of the discrete coordinate point set is determined by a preset convex hull boundary extraction method.

[0076] In one possible implementation, after the discrete coordinate point set passes the validity check, a convex hull scan algorithm based on vector cross product is used to extract the outer vertices. This algorithm calculates the turning directions between adjacent points and filters out the set of vertices that constitute the smallest convex polygon. It should be noted that convex hull scan algorithms (such as the GrahamScan algorithm) are common and well-known techniques in this field, and their detailed derivation will not be elaborated here.

[0077] S404. Determine the area of ​​the convex hull based on the spatial distribution characteristics of each vertex in the outer vertex sequence.

[0078] In one possible implementation, after obtaining the sequence of outer vertices, the x and y coordinates of each vertex are extracted sequentially according to their order on the convex hull boundary. The area enclosed by the convex polygon is calculated by summing the cross products of vectors. Specifically, the vertex sequence is arranged in a counter-clockwise or clockwise direction, and the directed area of ​​the parallelogram formed by adjacent vertices and the origin is calculated sequentially. The calculation results of all adjacent vertex pairs (including the first and last vertex pairs) are accumulated, and half of the absolute value of the accumulated result is taken as the convex hull area. This area reflects the degree of expansion of the discrete coordinate point set in two-dimensional space. The larger the area, the wider the range of congestion fluctuations experienced by the base station within the sliding time window.

[0079] For example, if the outer vertex sequence contains M points, they are denoted as follows: , ,…, Then calculate in sequence. and , and ... and The cross product contribution between the polygons is calculated by summing all cross product values, taking the absolute value, and then multiplying by 0.5 to obtain the area of ​​the convex hull. It should be noted that the Gaussian area formula for calculating the area of ​​a polygon is a well-known technique in this field, and its mathematical derivation will not be elaborated here.

[0080] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment defines the method for determining the convex hull envelope area and introduces an abnormal degradation mechanism. By deduplicating the coordinate point set, it is determined whether the number of remaining points is less than three or whether they are collinear. When the conditions for forming a polygon are not met, the envelope area is directly assigned to zero, avoiding algorithm crashes when data points are highly overlapping or collinear. When the conditions are met, a preset convex hull boundary extraction method is used to determine the sequence of outer vertices and calculate the area. This area directly quantifies the switching amplitude of the base station scheduler between idle and high-load states within the sliding time window. This degradation mechanism ensures that the system can still produce effective output without errors in extreme heavy congestion scenarios. At the same time, the special handling when the envelope area is zero also provides a precondition for extreme value degradation in subsequent exponential fusion, enhancing the robustness of the algorithm.

[0081] In one possible implementation, the process of determining the center metric value of the data point distribution center of the discrete coordinate point set can be specifically implemented through the following S501-S504, which will be explained in detail below.

[0082] S501. Determine the center value of the horizontal coordinate based on the distribution of the horizontal coordinates of all coordinate points in the discrete coordinate point set.

[0083] In one possible implementation, each coordinate point in the discrete coordinate set is traversed, the x-coordinate value of each point is extracted, all these x-coordinates are added together, and then divided by the total number of coordinate points. The resulting arithmetic mean is the center value of the x-coordinate. This center value represents the average congestion level across all sampling periods within the sliding time window in terms of queuing time limit.

[0084] For example, if the discrete coordinate point set contains 5 points with x-coordinates of 2, 5, 3, 8, and 2 respectively, then the sum of the x-coordinates is 20, the number of points is 5, and the center value of the x-coordinate is 4.

[0085] Understandably, the horizontal axis center value, as a component of the data point distribution center, reflects the average congestion baseline of the system within the time window, providing basic parameters for subsequent calculation of the center metric and the adaptive smoothing adjustment factor.

[0086] S502. Determine the center value of the ordinate based on the distribution of the ordinates of all coordinate points in the discrete coordinate point set.

[0087] In one possible implementation, the system iterates through each coordinate point in the discrete coordinate set, extracts the ordinate value of each point, sums all these ordinates, and then divides by the total number of coordinate points. The resulting arithmetic mean is the center value of the ordinate. This center value represents the average congestion level in terms of the burst traffic interference intensity dimension across all sampling periods within the sliding time window.

[0088] For example, if the discrete coordinate point set contains 5 points with ordinates of 1.2, 1.5, 1.3, 2.0, and 1.6 respectively, then the sum of the ordinates is 7.6, the number of points is 5, and the center value of the ordinate is 1.52.

[0089] Understandably, the center value of the vertical axis, as another component of the data point distribution center, together with the center value of the horizontal axis, determines the average congestion baseline position within the sliding window. This center value will subsequently be used to calculate the center metric (the product of the center values ​​of the horizontal and vertical axes), which will then serve as the basis for the smoothing adjustment factor, enabling adaptive gain adjustment under different load levels.

[0090] S503. Combine the center values ​​of the horizontal and vertical axes to form the distribution center of the data points.

[0091] In one possible implementation, the center values ​​of the horizontal and vertical coordinates are combined into an ordered pair, with the center value of the horizontal coordinate as the horizontal coordinate and the center value of the vertical coordinate as the vertical coordinate. This coordinate point represents the data point distribution center of the discrete coordinate set within the sliding time window. This center represents the average congestion baseline position of the system within the current time window.

[0092] S504. Determine the center measurement value based on the center values ​​of the horizontal and vertical axes.

[0093] In one possible implementation, the center value of the horizontal axis is multiplied by the center value of the vertical axis, and the product is used as a center metric for the distribution center of the data points. This center metric is a dimensionless numerical value used to characterize the overall level of the average congestion baseline within the current sliding time window.

[0094] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment defines the method for determining the data point distribution center and its center metric. By calculating the abscissa center value and ordinate center value of all coordinate points in the discrete coordinate point set, the data point distribution center is obtained. This center represents the average congestion baseline level of the system within this time window. Further, the center metric is determined based on the abscissa center value and ordinate center value. This metric serves as the basis for smoothing adjustment in the subsequent exponential fusion formula. Its core function is: when the network as a whole is in a low-load stable period, the center metric is smaller, which can adaptively amplify the fluctuation gain caused by small bursts of traffic, improving the sensitivity of capturing abnormal bursts during low-load periods; conversely, when the network is in a high-load saturation period, the center metric is larger, automatically smoothing out unnecessary oscillations caused by regular traffic fluctuations, thereby solving the technical problem of frequent false alarms in traditional fixed threshold monitoring at high levels.

[0095] In one possible implementation, the process of determining the projection component of the latest state point in the discrete coordinate point set relative to the congestion aggravation direction of the data point distribution center can be specifically implemented through the following S601-S603, which will be explained in detail below.

[0096] S601. Generate an instantaneous direction offset vector based on the positional offset between the coordinates of the latest state point and the coordinates of the data point distribution center.

[0097] In one possible implementation, the latest coordinate point in time sequence is extracted from the set of discrete coordinate points as the latest state point. Its x and y coordinates are subtracted from the x and y coordinates of the data point distribution center, respectively. The resulting differences are used as the two components of the instantaneous directional offset vector. This vector points from the data point distribution center to the latest state point, representing the instantaneous direction of the system load state deviating from the average baseline at the current moment.

[0098] For example, if the latest state point coordinates are (15, 2.5) and the data point distribution center coordinates are (4, 1.52), then the instantaneous direction offset vector is (11, 0.98).

[0099] S602. Obtain the preset congestion reference direction vector.

[0100] The preset congestion reference direction vector points in a direction in which both the horizontal and vertical coordinates change in the direction of congestion worsening.

[0101] In one possible implementation, a fixed direction vector is pre-defined as the congestion reference direction. This vector points in a two-dimensional coordinate system where both the horizontal and vertical coordinates change in the direction of congestion worsening, i.e., the direction in which queuing time limits increase and the intensity of sudden traffic disruptions increases. This pre-defined vector remains unchanged throughout the monitoring process and is used for direction comparison with the instantaneous direction offset vector.

[0102] For example, the preset congestion reference direction vector is set to (1, 1), pointing to the diagonal direction of the first quadrant. This direction requires both an increase in the horizontal coordinate (longer queuing time) and an increase in the vertical coordinate (stronger sudden interference), which is consistent with the physical meaning of increased congestion.

[0103] S603. Determine the congestion aggravation direction projection component based on the degree of directional correlation between the instantaneous directional offset vector and the preset congestion reference directional vector.

[0104] In one possible implementation, the dot product of the instantaneous directional offset vector and the preset congestion reference direction vector is calculated, and the dot product result is used as the projection component of the congestion aggravation direction. This dot product reflects the degree of agreement between the two vectors in direction: a positive dot product indicates that the instantaneous offset direction and the congestion aggravation direction have components in the same direction, and the larger the positive value, the more obvious the deterioration trend; a negative or zero dot product indicates that the instantaneous offset direction does not point to the congestion aggravation direction, that is, the congestion is easing or oscillating laterally.

[0105] For example, if the instantaneous directional offset vector is (11, 0.98) and the preset congestion reference direction vector is (1, 1), then the dot product is 11×1 + 0.98×1 = 11.98, and the projected component is 11.98, which is greater than 0, indicating that the latest state point is shifting in the direction of worsening congestion. If the instantaneous directional offset vector is (-3, 1), the dot product is -3×1 + 1×1 = -2, and the projected component is -2, which is less than 0, indicating that the congestion is showing a trend of easing.

[0106] Understandably, the sign of the projection component in the direction of congestion aggravation directly determines the value of the directional penalty activation coefficient: when the projection component is greater than 0, the incremental penalty mechanism is activated; when the projection component is less than or equal to 0, the penalty term is zeroed. This design ensures that additional penalties are introduced only during the congestion worsening phase, avoiding false triggering of interventions during congestion relief or stable states, and improving the targeting and accuracy of early warnings.

[0107] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment defines the method for determining the projection component of the congestion aggravation direction. An instantaneous direction offset vector is generated by calculating the positional offset between the latest state point coordinates and the data point distribution center coordinates. This vector represents the instantaneous movement direction of the system load state deviating from the average baseline at the latest moment. Combined with a preset congestion reference direction vector pointing in both horizontal and vertical coordinates towards the direction of congestion deterioration, the directional correlation between the two vectors is calculated as the projection component. When the projection component is greater than zero, it indicates that the network state at the latest moment is evolving in the direction of congestion deterioration; when it is less than or equal to zero, it indicates that congestion is easing or oscillating laterally. This feature enables the early warning mechanism to distinguish the directionality of congestion trends, avoids false triggering of interventions during congestion easing phases, and improves the accuracy of early warnings.

[0108] In one possible implementation, the process of determining the congestion warning assessment factor based on the coordinate base of the latest state point, the area of ​​the convex hull envelope, the projection component of the congestion aggravation direction, and the center metric of the data point distribution center can be specifically implemented through the following S701-S705, which will be explained in detail below.

[0109] S701. Determine the directional penalty activation coefficient based on the numerical sign characteristics of the projection component in the direction of congestion aggravation.

[0110] Among them, the directional penalty activation coefficient is used to activate the incremental penalty mechanism when the directional projection component of congestion aggravation indicates a trend of congestion deterioration.

[0111] In one possible implementation, the value of the projection component in the direction of congestion aggravation is extracted, and it is determined whether it is greater than zero. If the projection component is greater than zero, it indicates that the offset direction of the latest state point relative to the data point distribution center is consistent with the preset congestion reference direction, that is, the network state is evolving in the direction of congestion deterioration. At this time, the direction penalty activation coefficient is determined to be a positive number (for example, a value of 1) to activate the subsequent incremental penalty mechanism. If the projection component is less than or equal to zero, it indicates that the network congestion is easing or oscillating laterally. At this time, the direction penalty activation coefficient is determined to be zero, so that the incremental penalty term is zeroed.

[0112] For example, if the projection component of the congestion aggravation direction is 11.98, which is greater than zero, then the direction penalty activation coefficient is 1; if the projection component is -2, then the direction penalty activation coefficient is 0.

[0113] S702. Determine the smoothing adjustment factor based on the center measurement value of the data point distribution center and the preset bias constant.

[0114] The easing factor is used to adjust the intensity of the incremental penalty based on the current network load baseline level.

[0115] In one possible implementation, the centrality metric is added to the bias constant, and the reciprocal is taken to obtain the smoothing adjustment factor. The smoothing adjustment factor is used to adaptively adjust the strength of the incremental penalty based on the current network load baseline: when the centrality metric is small (corresponding to a low-load stable period), the smoothing adjustment factor is larger, which can amplify the fluctuation gain caused by small bursts of traffic; when the centrality metric is large (corresponding to a high-load saturation period), the smoothing adjustment factor is smaller, which can suppress unnecessary oscillations caused by regular traffic fluctuations. The bias constant is used to prevent the reciprocal from becoming infinitely large when the centrality metric is zero, ensuring the numerical stability of the algorithm.

[0116] For example, if the centrality metric is 6.08 and the bias constant is 0.001, the sum is 6.081, and the suppression factor is approximately 0.1644. If the centrality metric is close to 0, the suppression factor is 1000, significantly amplifying the increment. If the centrality metric is very large (e.g., 100), the suppression factor is 0.01, effectively suppressing the increment.

[0117] S703. Determine the fluctuation increment coefficient based on the pacing adjustment factor, the directional penalty activation coefficient, and the convex hull area.

[0118] Among them, the fluctuation increment coefficient is used to characterize the additional penalty weight determined by the historical fluctuation range, the current trend direction, and the load baseline level.

[0119] In one possible implementation, the smoothing adjustment factor, the directional penalty activation coefficient, and the convex hull area are multiplied together, and the resulting product is used as the fluctuation increment coefficient. This coefficient comprehensively represents the additional penalty weight determined by the historical fluctuation range (convex hull area), the current trend direction (directional penalty activation coefficient), and the load baseline level (smoothing adjustment factor).

[0120] S704. When the area of ​​the convex hull is zero or close to zero, the coordinate base of the latest state point is used as the congestion warning assessment factor.

[0121] In one possible implementation, the numerical value of the convex hull area is determined. If the convex hull area is equal to zero or less than a preset minimum threshold (e.g., 0.001), it indicates that the discrete coordinate point set cannot form an effective convex polygon (e.g., insufficient number of points, collinearity, or all points having the same height). In this case, the fluctuation increment coefficient is no longer calculated, and the coordinate base of the latest state point (i.e., the product of the horizontal and vertical coordinates of the latest state point) is directly used as the congestion warning assessment factor.

[0122] For example, if the latest state point is (15, 2.5), the coordinate base is 37.5, and the convex hull area is 0, then the congestion warning assessment factor is 37.5.

[0123] Understandably, this step provides an extreme value degradation mechanism. In extremely congested scenarios (such as a continuously full buffer and all data points anchored at the upper limit boundary), the convex hull area may be zero or close to zero. In this case, calculating using the conventional fusion formula would cause the incremental term to become invalid. By degrading to a coordinate base, it ensures that the evaluation factor can still reflect the current severity of congestion, preventing the early warning system from failing and missing reports due to algorithm limitations.

[0124] S705. When the area of ​​the convex hull is greater than zero, a congestion warning assessment factor is generated based on the coordinate base of the latest state point and the fluctuation increment coefficient.

[0125] In one possible implementation, determining that the area of ​​the convex hull is greater than zero indicates that the set of discrete coordinate points can form an effective convex polygon, and that historical fluctuation information has reference value. In this case, the system uses the coordinate base of the latest state point as the baseline value, and then algebraically combines this baseline value with the fluctuation increment coefficient to generate the final congestion warning assessment factor. Specifically, the assessment factor can be equal to the baseline value multiplied by an adjustment coefficient, which is obtained by adding a fixed baseline value (e.g., 1) to the fluctuation increment coefficient, thereby achieving a multiplicative amplification of the baseline value.

[0126] For example, the congestion warning assessment factor satisfies the following formula: in, This represents the total number of sampling periods within the sliding window; and The x and y coordinates represent the two-dimensional state features of the latest sampling period (the Nth period). is the x-coordinate of the center of the data point distribution, and is the arithmetic mean of all x-coordinates in the discrete coordinate point set; The ordinate is the center of the data point distribution, and the arithmetic mean of all the ordinates in the discrete coordinate point set. This is a bias constant to prevent the denominator from being zero; The convex hull area is the area of ​​the polygon enclosed by the outer vertices of the discrete coordinate point set, reflecting the range of historical load fluctuations. For congestion aggravation direction projection component, and for the instantaneous direction vector and the preset congestion reference direction vector (1,1); Set negative values ​​to 0 to ensure that increments are only contributed in the direction of worsening congestion. The coordinate base of the latest state point represents the basic congestion level at the current moment; The denominator of the gain adjustment is 1, and its reciprocal decreases as the central product increases, thus achieving adaptive smoothing. The sum of the fixed base value 1 and the fluctuation increment coefficient constitutes the adjustment coefficient. It is an additional penalty, only in the direction of worsening congestion ( And historical fluctuations are significant. It will take effect when ( ).

[0127] Congestion warning assessment factors It is a dimensionless comprehensive indicator that considers the current congestion level (baseline), historical fluctuation range (convex hull area), direction of deterioration trend (projected component), and load baseline level (distribution center). The larger the value, the closer the network slice is to SLA violation and the more obvious the queuing degradation trend.

[0128] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment defines the fusion calculation method of the congestion warning assessment factor. By determining the directional penalty activation coefficient according to the numerical sign of the projection component, the incremental penalty mechanism is activated only when the congestion worsens, avoiding meaningless directional penalties. The smoothing adjustment factor is determined according to the center metric value of the data point distribution center and the preset bias constant, realizing load-adaptive gain adjustment. The fluctuation increment coefficient is determined by combining the smoothing adjustment factor, the directional penalty activation coefficient and the convex hull envelope area. This coefficient comprehensively reflects the additional penalty weight jointly determined by the historical state fluctuation range, the current trend direction and the load baseline level. When the envelope area is zero or close to zero, the assessment factor degenerates into the coordinate base of the latest state point, ensuring that the warning does not fail under extreme congestion conditions; when the envelope area is greater than zero, the comprehensive generation method of the coordinate base and the fluctuation increment coefficient enables the assessment factor to simultaneously reflect the current congestion base and the historical fluctuation trend, effectively overcoming the technical bias that fixed threshold alarms are prone to false alarms during high load periods.

[0129] In one possible implementation, the process of triggering end-to-end active latency measurement of the target network slice when the congestion warning assessment factor exceeds the preset alarm threshold can be specifically implemented through the following S801-S803, which will be described in detail below.

[0130] S801: In response to the congestion warning assessment factor exceeding the preset alarm threshold, the network slice selection auxiliary information of the target network slice is parsed, and the associated user equipment Internet Protocol address is obtained based on the network slice selection auxiliary information.

[0131] In one possible implementation, when the congestion warning assessment factor exceeds a preset alarm threshold (e.g., a value of 10), an active measurement process is immediately triggered. First, the Network Slice Selection Auxiliary Information (NSSAI) of the target network slice is parsed. This information is the unique identifier of the network slice in the control plane, containing fields such as slice type, service type, and distinguisher. Based on the parsed NSSAI, the core network is queried for all active user equipment (UEs) currently associated with that slice. A preset number (e.g., 3) of representative IP addresses are selected from the returned UE list as the target endpoints for the active measurement.

[0132] S802: Issue an active measurement protocol activation command to the core network user plane function gateway to trigger the sending of probe packets to the Internet Protocol address of the user equipment.

[0133] In one possible implementation, after obtaining the associated Internet Protocol (IP) address of the user equipment (UE), an active measurement protocol activation command is sent to the core network user plane function gateway (UPF) carrying the target network slice service. This command includes a list of UE IP addresses to be measured, probe packet sending parameters (such as sending frequency, packet size, and test duration), and the result reporting address. Upon receiving the command, the UPF sends probe packets to the specified UE IP address in the user plane tunnel corresponding to the target network slice and records measurement data such as round-trip time.

[0134] For example, if the preset test time window is 2 seconds and the probe message sending interval is 100 milliseconds, the user plane function gateway will continuously send about 20 probe messages and count the latency of each message. Finally, the average latency or worst latency will be calculated as the end-to-end latency measurement value.

[0135] S803. Obtain the end-to-end latency measurement data returned by the core network user plane function gateway.

[0136] In one possible implementation, after sending a probe message to the user equipment's Internet Protocol address, the core network user plane function gateway collects round-trip time (RTT) information according to a preset measurement period and sends the processed end-to-end delay measurement data back to the monitoring system. This measurement data can be the maximum delay, average delay, or a certain quantile delay for a single probe, depending on the measurement configuration.

[0137] For example, if the test time window is 2 seconds and a total of 20 probe messages are sent, and the user plane function gateway calculates a maximum latency of 52 milliseconds and an average latency of 48 milliseconds, then the maximum or average value will be reported as the actual latency measurement data. After receiving this data, the monitoring system will store it in its memory for subsequent comparison with the red-line latency threshold.

[0138] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment defines the specific process of triggering end-to-end active latency measurement on demand. Only when the congestion warning assessment factor exceeds the preset alarm threshold is the network slice selection auxiliary information of the target network slice parsed and the associated user equipment IP address queried. Then, an active measurement protocol activation command is sent to the core network user plane function gateway to send probe packets, and finally, end-to-end latency measurement data is obtained. This on-demand activation method completely avoids the backhaul air interface bandwidth congestion and gateway computing power burden caused by the normalized high-frequency deployment of probes in all network slices. Precise measurement is only initiated when the assessment factor indicates a queuing degradation trend, which not only ensures the accuracy of congestion violation confirmation, but also greatly reduces measurement overhead, achieving a balance between monitoring accuracy and resource consumption.

[0139] In one possible implementation, the process of adjusting the resource configuration parameters of low-priority slices under the same base station based on the measurement results can be specifically described in S901-S903 below.

[0140] S901. Compare the end-to-end latency measurement data with the preset red-line latency threshold.

[0141] In one possible implementation, after obtaining the measured end-to-end latency data returned by the core network user plane function gateway, the data is compared with a pre-stored red-line latency threshold. This red-line latency threshold is set according to the maximum tolerable latency parameter explicitly specified in the Service Level Agreement (SLA) of the target network slice, and is used to determine whether the current end-to-end service latency constitutes a substantial violation.

[0142] For example, if the SLA of the target network slice requires a maximum end-to-end latency of no more than 50 milliseconds, then the preset red-line latency threshold is set to 50 milliseconds. When the measured latency is 48 milliseconds, it is determined to be less than the threshold and there is no violation; when the measured latency is 52 milliseconds, it is determined to have reached or exceeded the threshold, and a congestion violation is confirmed to have occurred.

[0143] S902. When the measured end-to-end latency data reaches or exceeds the red line latency threshold, the resource constraint weight is determined based on the congestion warning assessment factor.

[0144] In one possible implementation, if the measured end-to-end latency is greater than or equal to a preset red-line latency threshold (e.g., 50 milliseconds), it is determined that the current queuing backlog has resulted in a substantial violation of service regulations. The congestion warning assessment factor (e.g., 191.6) calculated within the current sliding window is extracted, and the resource limiting weight for restricting low-priority slice bandwidth is determined based on this factor. To prevent dimensionless assessment factors from being directly distributed to physical network elements, causing parameter out-of-bounds errors, they are subject to scaling: for example, the assessment factor is multiplied by a preset scaling factor (e.g., 0.01) to obtain an initial reduction ratio; this ratio is then compared with a preset maximum reduction ratio threshold (e.g., 80%), and the smaller of the two values ​​is taken as the final resource limiting weight.

[0145] For example, if the congestion warning assessment factor is 191.6 and the scaling factor is 0.01, the initial deduction ratio is 1.916 (i.e., 191.6%), which exceeds the maximum deduction ratio of 80%. Therefore, the resource constraint weight is 80%. If the assessment factor is 50 and the initial deduction ratio is 0.5 (50%), which is less than 80%, the weight is 50%.

[0146] S903: Send adjustment instructions to the core network policy control function gateway.

[0147] In one possible implementation, after determining the resource limiting weight, the system encapsulates the resource limiting weight into a policy control rule modification instruction and sends this instruction to the policy control function gateway of the core network. This instruction carries the resource limiting weight and the identification information of the target network slice, instructing the policy control function gateway to query other slice instances that share the same base station hardware entity as the target network slice and have a lower priority than the target network slice, and to reduce the guaranteed bit rate configuration of these low-priority slices according to the resource limiting weight.

[0148] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment defines the specific method for adjusting the resource configuration of low-priority slices based on active measurement results, forming a complete closed-loop intervention. By comparing the end-to-end latency measurement data with the red-line latency threshold, resource limiting weights are determined based on evaluation factors only when a substantial violation is confirmed, and adjustment instructions are sent to the policy control function gateway to instruct the reduction of resource guarantee configurations for other slices that share the same base station with the target network slice and have a lower priority than the target network slice. This mechanism ensures that the intervention action has solid evidence of violation, avoids blindly adjusting normal services, and at the same time uses evaluation factors to guide the limiting weights, realizing a secure mapping from the warning index to physical network element configuration parameters. It can accelerate the consumption of backlogged packets in the target network slice by depriving low-priority slices of air interface resources without changing the base station's default scheduling mechanism, thereby alleviating SLA violations caused by burst traffic.

[0149] Please see Figure 2This document illustrates a system architecture diagram of an end-to-end network slicing SLA real-time monitoring system 200 according to an embodiment of the present invention. This system is used to implement the aforementioned end-to-end network slicing SLA real-time monitoring method. The system includes: a determination module 201, used to determine the base station downlink buffer occupancy rate and the core network traffic instantaneous burst coefficient for multiple sampling periods within a sliding time window; a coordinate generation module 202, used to generate two-dimensional state feature coordinate points and determine a discrete coordinate point set based on the base station downlink buffer occupancy rate and the core network traffic instantaneous burst coefficient for each sampling period; and an area and center determination module 203, used to determine the convex hull area of ​​the discrete coordinate point set and the data points of the discrete coordinate point set. The distribution center's center metric value; the projection component determination module 204, used to determine the congestion aggravation direction projection component of the latest state point in the discrete coordinate point set relative to the data point distribution center; the latest state point is the last coordinate point in the time sequence of the discrete coordinate point set; the congestion warning assessment factor determination module 205, used to determine the congestion warning assessment factor based on the coordinate base, convex hull envelope area, congestion aggravation direction projection component, and the center metric value of the data point distribution center of the latest state point; the intervention execution module 206, used to trigger end-to-end active latency measurement of the target network slice when the congestion warning assessment factor exceeds the preset alarm threshold, and adjust the resource configuration parameters of the low-priority slices under the same base station according to the measurement results.

[0150] The technical solution provided by the above embodiments can bring at least the following beneficial effects: This embodiment protects the technical solution of the aforementioned method in the form of a system. The system includes a determination module, a coordinate generation module, an area and center determination module, a projection component determination module, a congestion warning assessment factor determination module, and an intervention execution module. Each module executes the corresponding steps in the method to form a complete end-to-end network slice SLA real-time monitoring system. This system can generate a two-dimensional geometric envelope area using low-frequency statistical indicators, adaptively calculate the congestion warning assessment factor, and trigger active measurement and cross-slice resource reconfiguration as needed. Thus, without increasing the bandwidth overhead of normal measurement across the entire network, it can accurately identify the queuing congestion trend hidden behind the macro average, realizing a closed loop of network slice SLA degradation warning and cross-slice air interface resource reconfiguration for intermittent traffic.

[0151] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0152] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for real-time monitoring of end-to-end network slicing SLA, characterized in that, The method includes: Determine the downlink buffer occupancy rate of the base station for multiple sampling periods within the sliding time window, as well as the instantaneous burst coefficient of core network traffic; Based on the base station downlink buffer occupancy rate and the instantaneous burst coefficient of the core network traffic in each sampling period, two-dimensional state feature coordinate points are generated to determine the discrete coordinate point set; Determine the convex hull area of ​​the discrete coordinate point set, and determine the center metric value of the data point distribution center of the discrete coordinate point set; Determine the congestion aggravation direction projection component of the latest state point in the discrete coordinate point set relative to the center of the data point distribution; the latest state point is the last coordinate point in the time sequence of the discrete coordinate point set. Based on the coordinate base of the latest state point, the area of ​​the convex hull envelope, the projection component of the congestion aggravation direction, and the center metric of the data point distribution center, the congestion warning assessment factor is determined. When the congestion warning assessment factor exceeds the preset alarm threshold, an end-to-end active latency measurement of the target network slice is triggered, and the resource configuration parameters of the low-priority slices under the same base station are adjusted according to the measurement results.

2. The end-to-end network slicing SLA real-time monitoring method according to claim 1, characterized in that, The determination of the base station downlink buffer occupancy rate for multiple sampling periods within the sliding time window, and the instantaneous burst coefficient of core network traffic, includes: At the end of each sampling period's time segment, the current backlog of bytes and the maximum downlink buffer capacity of the base station are read; The base station downlink buffer occupancy rate for each sampling period is determined based on the ratio of the current number of backlogged bytes to the maximum downlink buffer capacity. Within each sampling period's time segment, the total number of incoming bytes from the core network gateway is read multiple times at preset time intervals to generate a throughput micro-sequence; The instantaneous burst coefficient of core network traffic for each sampling period is determined based on the ratio of the standard deviation to the arithmetic mean of the throughput microsequence.

3. The end-to-end network slicing SLA real-time monitoring method according to claim 1, characterized in that, The step of generating two-dimensional state feature coordinate points and determining a discrete coordinate point set based on the base station downlink buffer occupancy rate and the instantaneous burst coefficient of the core network traffic in each sampling period includes: The first-dimensional coordinate value is determined based on the downlink buffer occupancy rate of the base station and the degree of difference between it and the preset benchmark value; the first-dimensional coordinate value is used to quantify the queuing time limit. The second-dimensional coordinate value is determined based on the amplified result of the fluctuation intensity of the instantaneous burst coefficient of the core network traffic; the second-dimensional coordinate value is used to quantify the intensity of traffic burst interference. The first-dimensional coordinate value is combined with the second-dimensional coordinate value to generate the two-dimensional state feature coordinate point; The two-dimensional state feature coordinate points generated in each sampling period are arranged in chronological order to determine the discrete coordinate point set.

4. The end-to-end network slicing SLA real-time monitoring method according to claim 1, characterized in that, Determining the convex hull area of ​​the discrete coordinate point set includes: The coordinate points in the discrete coordinate point set are deduplicated, and it is determined whether the number of remaining coordinate points after deduplication is less than three, or whether all the deduplicated coordinate points are located on the same straight line. When the number of remaining coordinate points after deduplication is less than three, or when all deduplicated coordinate points are located on the same straight line, the area of ​​the convex hull is determined to be zero. When the number of remaining coordinate points after deduplication is greater than or equal to three, and all the deduplicated coordinate points are not located on the same straight line, the outer vertex sequence of the discrete coordinate point set is determined by a preset convex hull boundary extraction method. The area of ​​the convex hull is determined based on the spatial distribution characteristics of each vertex in the outer vertex sequence.

5. The end-to-end network slicing SLA real-time monitoring method according to claim 1, characterized in that, The centrality metric value for determining the data point distribution center of the discrete coordinate point set includes: The center value of the horizontal coordinate is determined based on the distribution of the horizontal coordinates of all coordinate points in the discrete coordinate point set. Based on the distribution of the ordinates of all coordinate points in the discrete coordinate point set, determine the center value of the ordinate; The center value of the horizontal axis and the center value of the vertical axis are combined to form the distribution center of the data points; The center measurement value is determined based on the center value of the horizontal axis and the center value of the vertical axis.

6. The end-to-end network slicing SLA real-time monitoring method according to claim 1, characterized in that, Determining the projection component of the latest state point in the discrete coordinate point set relative to the congestion aggravation direction of the data point distribution center includes: Based on the positional offset between the coordinates of the latest state point and the coordinates of the data point distribution center, an instantaneous direction offset vector is generated; Obtain a preset congestion reference direction vector; the preset congestion reference direction vector points in a direction in which both the horizontal and vertical coordinates change in the direction of congestion worsening; The congestion aggravation direction projection component is determined based on the degree of directional correlation between the instantaneous directional offset vector and the preset congestion reference directional vector.

7. The end-to-end network slicing SLA real-time monitoring method according to claim 1, characterized in that, The step of determining the congestion warning assessment factor based on the coordinate base of the latest state point, the area of ​​the convex hull, the projection component of the congestion aggravation direction, and the center metric of the data point distribution center includes: Based on the numerical sign characteristics of the projection component in the congestion aggravation direction, a directional penalty activation coefficient is determined; the directional penalty activation coefficient is used to activate the incremental penalty mechanism when the projection component in the congestion aggravation direction indicates a congestion deterioration trend. The shaving adjustment factor is determined based on the center metric value of the data point distribution center and the preset bias constant; the shaving adjustment factor is used to adjust the intensity of the incremental penalty according to the current network load baseline level. The fluctuation increment coefficient is determined based on the smoothing adjustment factor, the direction penalty activation coefficient, and the convex hull envelope area; the fluctuation increment coefficient is used to characterize the additional penalty weight jointly determined by the historical state fluctuation range, the current trend direction, and the load baseline level. When the area of ​​the convex hull is zero or close to zero, the coordinate base of the latest state point is used as the congestion warning assessment factor. When the area of ​​the convex hull is greater than zero, the congestion warning assessment factor is generated based on the coordinate base of the latest state point and the fluctuation increment coefficient.

8. The end-to-end network slicing SLA real-time monitoring method according to claim 1, characterized in that, When the congestion warning assessment factor exceeds a preset alarm threshold, the end-to-end active latency measurement of the target network slice is triggered, including: In response to the congestion warning assessment factor exceeding the preset alarm threshold, the network slice selection auxiliary information of the target network slice is parsed, and the associated user equipment Internet Protocol address is obtained by querying the network slice selection auxiliary information. An active measurement protocol activation command is sent to the core network user plane function gateway to trigger the sending of probe packets to the Internet Protocol address of the user equipment; Obtain the end-to-end latency measurement data returned by the core network user plane function gateway.

9. The end-to-end network slicing SLA real-time monitoring method according to claim 8, characterized in that, The adjustment of resource configuration parameters for low-priority slices under the same base station based on measurement results includes: The measured end-to-end delay data is compared with the preset red-line delay threshold; When the measured end-to-end latency data reaches or exceeds the red line latency threshold, the resource constraint weight is determined according to the congestion warning assessment factor. Send an adjustment instruction to the core network policy control function gateway; the adjustment instruction carries the resource limiting weight, which is used to instruct to reduce the resource guarantee configuration of other slices that share the same base station with the target network slice and have a lower priority than the target network slice.

10. An end-to-end network slicing SLA real-time monitoring system, characterized in that, The system is used to implement an end-to-end network slicing SLA real-time monitoring method as described in any one of claims 1 to 9, the system comprising: The determination module is used to determine the base station downlink buffer occupancy rate and the instantaneous burst coefficient of core network traffic for multiple sampling periods within a sliding time window; The coordinate generation module is used to generate two-dimensional state feature coordinate points and determine the discrete coordinate point set based on the base station downlink buffer occupancy rate and the instantaneous burst coefficient of the core network traffic in each sampling period. The area and center determination module is used to determine the area of ​​the convex hull of the discrete coordinate point set, and to determine the center metric value of the data point distribution center of the discrete coordinate point set. The projection component determination module is used to determine the congestion aggravation direction projection component of the latest state point in the discrete coordinate point set relative to the center of the data point distribution; the latest state point is the last coordinate point in the time sequence of the discrete coordinate point set. The congestion warning assessment factor determination module is used to determine the congestion warning assessment factor based on the coordinate base of the latest state point, the area of ​​the convex hull envelope, the projection component of the congestion aggravation direction, and the center metric value of the data point distribution center. The intervention execution module is used to trigger end-to-end active latency measurement of the target network slice when the congestion warning assessment factor exceeds the preset alarm threshold, and adjust the resource configuration parameters of the low-priority slice under the same base station according to the measurement results.

Images