A method and system for dynamic allocation of bandwidth of CPE based on user behavior prediction

Abstract

The application relates to the technical field of bandwidth resource management, and discloses a CPE bandwidth dynamic allocation method and system based on user behavior prediction. The method comprises the following steps: forming a terminal short-period bandwidth sequence on a time slice scale, and identifying a starting window of an action-triggered burst on the sequence, so that the bandwidth allocation is no longer completely dependent on a passive response after congestion, and the method has the capability of early identification of the starting burst; meanwhile, the window normalization and the sequence of the inter-terminal direction communication volume are counted, a regional communication dependence matrix is constructed, and key nodes and key edges are derived, so that the bandwidth allocation can perform hard reservation and priority guarantee towards the key nodes of the dependence chain, thereby improving the availability and stability of the key service starting stage. Therefore, the problems that the existing CPE bandwidth allocation cannot identify the starting window in advance in the action-triggered burst scenario, cannot express the inter-terminal communication dependence relationship, and thus the bandwidth of the key service starting stage cannot be guaranteed are solved.

Description

Technical Field

[0001] This invention relates to the field of bandwidth resource management technology, and in particular to a method and system for dynamic allocation of CPE bandwidth based on user behavior prediction. Background Technology

[0002] In home broadband, SME access, and campus edge networks, CPE (Customer Premises Equipment) typically serves as the egress node for multi-terminal shared access links, carrying concurrent transmissions of various services such as video conferencing, live streaming and video-on-demand, cloud disk synchronization, and IoT reporting. CPE-side bandwidth allocation strategies often rely on "current traffic volume" or "fixed service priority," meaning they apply rate limiting, shaping, or queue scheduling to terminal traffic based on throughput, queue length, or preset weights at a given moment. While this approach can maintain basic fairness in relatively stable service scenarios, it often leads to adjustment lag, sudden congestion, and a significant increase in the latency of the first packet for critical services when user behavior undergoes short-term, drastic changes.

[0003] Existing bandwidth allocation strategies generally lack the ability to identify "action-triggered bursts." Action-triggered bursts refer to a sudden increase in bandwidth demand within a very short time window when a user performs a specific action or operation. Examples include clicking to join a video conference before it begins, the player launching at the start of a live stream, or a home security device uploading video clips after triggering an alarm. These bursts share the common characteristic of being preceded by a brief "startup warning," but this warning may not manifest as high throughput. Traditional methods passively adjust based solely on instantaneous throughput, often only starting rate limiting or adjustments after congestion has already occurred, leading to stuttering or handshake timeouts for critical services during the startup phase.

[0004] On the other hand, there are often obvious communication dependencies between terminals in actual home or campus networks. For example, from a camera to a gateway / storage, from a projection source device to a display terminal, and from an authentication terminal to an authentication service. Such dependencies are not simply a matter of traffic volume, but rather manifest as "certain terminals having a stable order and proportion of communication to certain terminals / services within an action window." Traditional methods often treat terminals as independent competitors, lacking the expression and utilization of the dependency structure between terminals. This leads to a situation where, when a shared link is contested, critical nodes (depending nodes) are not given priority protection, resulting in the phenomenon of "critical nodes being restricted, causing the entire link to fail." Even if other terminals are allocated bandwidth, they cannot effectively complete the business actions on the dependent chain.

[0005] In summary, existing technologies have the following shortcomings in "short-term sudden scenarios caused by users performing a certain action or operation": First, they lack the ability to identify the time window of sudden start-up symptoms, resulting in a lag in bandwidth allocation; second, they lack the ability to characterize the communication dependency structure between several operating devices connected to CPE terminals within a region, making it difficult for critical nodes to obtain available bandwidth during congestion; third, they lack a mechanism to unify the "action window identification results," "dependency matrix results," and "bandwidth prediction results" into the same allocation logic, making it difficult for the allocation strategy to simultaneously take into account both foresight and critical business availability in sudden scenarios. Summary of the Invention

[0006] This invention provides a CPE bandwidth dynamic allocation method and system based on user behavior prediction, which at least solves the problem that existing CPE bandwidth allocation cannot identify the start-up window in advance and cannot express the communication dependency between terminals in user-triggered burst scenarios, thus resulting in the bandwidth not being guaranteed during the critical service start-up phase.

[0007] To achieve the above objectives, this invention provides a CPE bandwidth dynamic allocation method based on user behavior prediction, the method comprising the following steps:

[0008] The time-slice communication volume of each terminal under each CPE is obtained according to the preset time-slice length, and the terminal short-cycle bandwidth sequence is generated. The allocable bandwidth limit of each CPE is obtained, and the inter-terminal directional communication volume is summarized based on the time-slice directional communication volume statistics.

[0009] Based on the terminal's short-period bandwidth sequence, the terminal initiates burst execution action window identification processing to obtain an action window marker;

[0010] A regional communication dependency matrix is ​​constructed based on the aggregated inter-terminal directional communication traffic, and a set of key nodes and a set of key edges are generated based on the regional communication dependency matrix.

[0011] Based on the action window marker and the terminal short-cycle bandwidth sequence, generate the predicted terminal demand value for the next time slice;

[0012] Based on the demand forecast, the allocable bandwidth limit, the set of key nodes, and the set of key edges, a terminal bandwidth allocation result is generated, and the terminal bandwidth allocation result is converted into executable rate limiting parameters or queue guarantee parameters to take effect in the corresponding time slice.

[0013] Optionally, the time-slice communication volume of each terminal under each CPE is obtained according to a preset time-slice length, and a short-cycle bandwidth sequence of the terminal is generated. The allocatable bandwidth limit of each CPE is obtained, and a summary of inter-terminal directional communication volume is generated based on the time-slice directional communication volume statistics. Specifically, this includes:

[0014] For each terminal of each CPE, the cumulative number of bytes in each time slice is obtained, and the time slice rate is converted on the cumulative number of bytes to generate a short-cycle bandwidth sequence for the terminal.

[0015] A sliding window buffer write process is performed on the terminal's short-cycle bandwidth sequence to form a bandwidth sequence segment of the most recent W time slices;

[0016] The allocable bandwidth limit of the CPE is obtained and time-slice aligned storage is performed. Statistical summarization processing is performed on the directional communication traffic between terminals within the time slice to generate a summary of directional communication traffic between terminals.

[0017] Optionally, based on the terminal's short-period bandwidth sequence, the terminal initiates burst execution action window identification processing to obtain action window markers, specifically including:

[0018] The bandwidth subsequence of the terminal's most recent K time slices is obtained based on the sliding window buffer, and the relative burst ratio is calculated on the bandwidth subsequence to obtain the burst ratio;

[0019] Perform a continuously rising count operation on the bandwidth subsequence to obtain a continuously rising count;

[0020] A startup feature score is generated based on the surge ratio and the continuous rise count, and a threshold persistence determination process is performed on the startup feature score to obtain an action window marker.

[0021] Specifically, when the activation feature score exceeds the threshold and lasts for at least L time slices, the action window is marked as 1; otherwise, the action window is marked as 0.

[0022] Optionally, a regional communication dependency matrix is ​​constructed based on the aggregated inter-terminal directional communication traffic, specifically including:

[0023] Based on the aggregation of directional communication between terminals, obtain the directional communication window of the most recent W time slices, and perform window accumulation and normalization processing on each pair of terminals to calculate the amount of dependency evidence.

[0024] The quantity of dependent evidence is output as input to generate the elements of the regional communication dependency matrix, and the regional communication dependency matrix is ​​constructed.

[0025] Optionally, a regional communication dependency matrix is ​​constructed, specifically including:

[0026] The time slice set during the action window is determined based on the action window marker, and the directional communication within the time slice set is used to perform time slice positioning for the first time exceeding the starting threshold.

[0027] Based on the time slice positioning result of the first time exceeding the starting threshold, the terminal pair consisting of terminal i and terminal j is subjected to statistical processing of the order relationship to generate the order factor;

[0028] The sequence factor is used to characterize the frequency proportion of communication from terminal i to terminal j before communication in other directions in multiple action window events.

[0029] Based on the amount of dependency evidence and the prior factors, a synthesis process is performed to generate dependency strength. The dependency strengths of each terminal pair are then arranged according to the regional terminal index to form a regional communication dependency matrix.

[0030] Optionally, generating a set of key nodes and a set of key edges based on the regional communication dependency matrix specifically includes:

[0031] Based on the regional communication dependency matrix, column summation is performed on each terminal to calculate the degree of dependency. Threshold filtering is then performed on the degree of dependency to add the terminal index that meets the key node determination criteria to the key node set.

[0032] A threshold filtering process is performed on the dependency strength in the regional communication dependency matrix to add directed edges that meet the key edge determination criteria to the key edge set.

[0033] Output the set of key nodes and the set of key edges for demand prediction and bandwidth allocation.

[0034] Optionally, based on the action window marker and the terminal short-cycle bandwidth sequence, a predicted value for the terminal demand in the next time slice is generated, specifically including:

[0035] A basic prediction value is set for each terminal based on the terminal short-cycle bandwidth sequence, and the basic prediction value is then subjected to exponential smoothing update processing.

[0036] Based on the action window markers, the base forecast value is either adjusted upwards or maintained to obtain the demand forecast value.

[0037] Optionally, a terminal bandwidth allocation result is generated based on the demand forecast, the allocable bandwidth limit, the key node set, and the key edge set, specifically including:

[0038] Based on the demand forecast, the allocable bandwidth limit, and the action window marker, hard reservation quota is calculated for key terminals to generate hard reservation quota.

[0039] The remaining capacity is calculated based on the allocable bandwidth limit and the hard reserved quota. For non-critical terminals, a proportional allocation process is performed based on the remaining capacity and the demand forecast value to generate terminal bandwidth allocation results.

[0040] Optionally, the terminal bandwidth allocation result is converted into executable rate limiting parameters or queue guarantee parameters to take effect in the corresponding time slice, specifically including:

[0041] Based on the action window, the trigger ratio of the statistical area is marked, and the ratio allocation of non-critical terminals is processed by the area throttling scaling.

[0042] The terminal bandwidth allocation result is converted into a token bucket rate or a queue minimum guarantee parameter and sent out so that the parameter takes effect in the next time slice.

[0043] Furthermore, to achieve the above objectives, the present invention also provides a CPE bandwidth dynamic allocation system based on user behavior prediction, comprising:

[0044] The acquisition module is used to acquire the time slice communication volume of each terminal under each CPE according to the preset time slice length and generate the terminal short-cycle bandwidth sequence, acquire the allocable bandwidth limit of each CPE, and generate the inter-terminal directional communication volume summary based on the time slice directional communication volume statistics.

[0045] The identification module is used to identify the burst execution action window of the terminal based on the short-period bandwidth sequence of the terminal, so as to obtain the action window mark;

[0046] The construction module is used to construct a regional communication dependency matrix based on the summary of inter-terminal directional communication traffic, and to generate a set of key nodes and a set of key edges based on the regional communication dependency matrix.

[0047] The generation module is used to generate a predicted value of terminal demand for the next time slice based on the action window marker and the terminal short-cycle bandwidth sequence.

[0048] The allocation module is used to generate terminal bandwidth allocation results based on the demand forecast value, the allocable bandwidth limit, the set of key nodes, and the set of key edges, and to convert the terminal bandwidth allocation results into executable rate limiting parameters or queue guarantee parameters to take effect in the corresponding time slice.

[0049] The beneficial effects of this invention are as follows: It proposes a CPE bandwidth dynamic allocation method and system based on user behavior prediction. By forming a short-cycle bandwidth sequence of terminals on a time-slice scale and identifying the start-up window of action-triggered bursts in this sequence, bandwidth allocation no longer relies entirely on passive response after congestion, but has the ability to identify start-up bursts in advance. At the same time, by normalizing the window and statistically analyzing the sequence of inter-terminal directional communication traffic, a regional communication dependency matrix is ​​constructed and key nodes and key edges are derived. This enables bandwidth allocation to perform hard reservation and priority guarantee for key nodes in the dependency chain, thereby improving the availability and stability of the critical service start-up phase in action-triggered burst scenarios. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the method of an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0053] This invention provides a method for dynamic allocation of CPE bandwidth based on user behavior prediction, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the CPE bandwidth dynamic allocation method based on user behavior prediction, as described in an embodiment of the present invention.

[0054] In this embodiment, a CPE bandwidth dynamic allocation method based on user behavior prediction is described, the method comprising the following steps:

[0055] S1: Obtain the time slice communication volume of each terminal under each CPE according to the preset time slice length and generate the terminal short-cycle bandwidth sequence. Obtain the upper limit of the allocable bandwidth of each CPE and generate the summary of inter-terminal directional communication volume based on the time slice directional communication volume statistics.

[0056] It should be noted that step S1 is used to generate key data objects required for subsequent action window identification, dependency matrix construction, short-cycle demand prediction and bandwidth allocation at a unified time slice scale.

[0057] Specifically, for each terminal of each CPE, the cumulative number of bytes within each time slice is obtained, and time slice rate conversion is performed on the cumulative number of bytes to generate a short-cycle bandwidth sequence for the terminal; a sliding window cache write process is performed on the short-cycle bandwidth sequence for the terminal to form a bandwidth sequence segment of the most recent W time slices; the allocatable bandwidth limit of the CPE is obtained and time slice aligned storage is performed, and statistical summary processing is performed on the directional communication traffic between terminals within the time slice to generate a summary of directional communication traffic between terminals.

[0058] For the construction of the terminal short-period bandwidth sequence, the time slice length is set to... For terminal d under the c-th CPE, obtain the cumulative number of bytes within time slice t. And convert it into time slice bandwidth rate. For example, calculate using the following formula:

[0059] ;

[0060] in, This represents the bandwidth rate of terminal d within time slice t; ΔT represents the cumulative number of bytes in the terminal within time slice t; ΔT represents the time slice length; the coefficient 8 is used for the conversion from bytes to bits;

[0061] In this embodiment of the invention, by converting discrete traffic counts into a rate sequence with uniform granularity, subsequent steps can directly compare and determine bandwidth change trends, burst start characteristics, and prediction errors on the same time axis.

[0062] In one embodiment, ΔT can be 5 seconds, 10 seconds, or other fixed values. Those skilled in the art can select an appropriate time slice length based on the CPE's processing capabilities and burst response requirements. An excessively large time slice may reduce the sensitivity of the startup window, while an excessively small time slice may increase the update frequency and computational overhead.

[0063] To obtain and align the allocatable bandwidth limit of a CPE, obtain the allocatable bandwidth limit of the c-th CPE in time slice t. and compare it with the obtained time slice bandwidth rate. Align storage along the time-slice dimension. This represents the upper limit of the total available capacity of the CPE for terminal bandwidth allocation within time slice t. This upper limit can be determined by the access link synchronization rate, uplink configuration, or policy rate limiting; in some implementations, if there are independent constraints in different directions (uplink / downlink), they can be established separately. and And in subsequent steps, prediction and allocation are performed respectively.

[0064] In this embodiment of the invention, by obtaining and aligning the upper limit of the allocable bandwidth of the CPE, a clear total constraint is provided for the subsequent allocation stage, so that the allocation result can ensure that the sum of the bandwidth allocation results of all terminals does not exceed the upper limit of the allocable bandwidth, thus avoiding over-allocation caused by relying solely on demand prediction.

[0065] To generate the summary of directional communication traffic between terminals, the directional communication traffic from terminal i to terminal j within time slice t is statistically summarized to obtain the directional communication traffic summary. .in, This represents the cumulative number of bytes or equivalent communication volume sent from terminal i to terminal j within time slice t. In an optional implementation, terminal j may also represent a preset service node, such as an authentication service, gateway service, or storage service, whose identifier can be included in the regional index according to a unified numbering rule.

[0066] In this embodiment of the invention, by compressing the communication relationship between terminals from the original connection or flow granularity to a directional communication quantity-time slice structure, the subsequent dependency matrix construction can obtain a stable proportion structure and directional evidence based on window statistics.

[0067] In one embodiment, for example, when a video conference join action occurs, directional communication from the terminal to the authentication / signaling service will appear first in the action window and account for a relatively concentrated proportion; this phenomenon will be reflected in the sum of directional communication volume within adjacent time slices. The continuous emergence and cumulative improvement of [these] provide direct input for subsequent dependency building.

[0068] S2: Based on the terminal's short-cycle bandwidth sequence, perform burst execution action window identification processing on the terminal to obtain action window markers.

[0069] It should be noted that step S2 is used to identify the start-up phase time window in action-triggered sudden scenarios and output the action window marker. Action window markers are used to limit the scope of dependency statistics in subsequent steps, trigger forecast upscaling, and trigger hard reservations at critical nodes.

[0070] Specifically, the bandwidth subsequence of the terminal's most recent K time slices is obtained based on the sliding window cache. A relative surge ratio is calculated on the bandwidth subsequence to obtain the surge ratio. A continuous rise count operation is performed on the bandwidth subsequence to obtain the continuous rise count. A startup feature score is generated based on the surge ratio and the continuous rise count. A threshold persistence determination process is performed on the startup feature score to obtain an action window flag. Specifically, when the startup feature score exceeds the threshold and persists for at least L time slices, the action window flag is set to 1; otherwise, the action window flag is set to 0.

[0071] To construct a window baseline and calculate the relative burst ratio, the terminal short-cycle bandwidth sequence is obtained. As input, a window mean benchmark is constructed based on the most recent K time slices, and the relative burst ratio is calculated. For example, calculate using the following formula:

[0072] ;

[0073] in, The value represents the burst ratio; K represents the window length; ϵ is a constant to prevent the denominator from being zero; Sum the bandwidth within the window.

[0074] In this embodiment of the invention, by transcribing the current bandwidth level into a ratio of the relative window average, the ramp-up before a sudden start can be significantly presented in a proportional form, thereby reducing the impact of the difference in absolute bandwidth between different terminals on the determination.

[0075] In some implementations, the window mean can also be replaced by the window mean or weighted mean; those skilled in the art can select a suitable window reference form according to the terminal bandwidth fluctuation characteristics.

[0076] For calculating continuous ascent characteristics to characterize the initiation climb pattern, within the window... Perform continuous ascending counting to obtain the continuous ascending count. .in, It can be defined as satisfying The number of consecutive time slices, or defined as the ratio of the number of rises within a window to the length of the window.

[0077] In this embodiment of the invention, action-triggered bursts typically exhibit a gradual increase in bandwidth over several consecutive time slices rather than a single-point spike. This continuous increase characteristic can suppress the interference of occasional spikes and transient noise on the determination of the action window.

[0078] For generating action window markers and applying persistent constraints, based on the burst ratio With continuous rising count Generate start-up decision quantity The determination quantity is then subjected to threshold and persistence determination to output an action window marker. .

[0079] In one embodiment, it can be generated and determined according to the following rules: when and And if it continues for at least L time slices, then set Set to 1, otherwise set to 1. The value is 0. Among them... These are preset threshold and duration parameters.

[0080] In this embodiment of the invention, continuous constraints are used to avoid false triggering caused by short-term jitter, and the action window is limited to the sudden start-up period rather than the sudden peak period, so that subsequent prediction and allocation can enter the protection state in advance.

[0081] It should be noted that the above thresholds and continuous parameters are not limited to a fixed value. Those skilled in the art can set them based on different access rates, different service types and different numbers of terminals. In some implementations, different threshold groups can also be set for different terminal types to adapt to the different startup modes of terminals such as cameras, gateways, and mobile phones.

[0082] S3: Construct a regional communication dependency matrix based on the aggregated inter-terminal directional communication traffic, and generate a set of key nodes and a set of key edges based on the regional communication dependency matrix.

[0083] It should be noted that step S3 is used to construct a communication dependency matrix among several operating devices connected to CPE terminals within the area. And derive the set of key nodes from it. With the set of key edges The dependency matrix is ​​used to express the strength of directional dependencies between terminals, while critical nodes and critical edges are used to determine hard-reserved objects and priority protection objects in subsequent steps.

[0084] Specifically, constructing a regional communication dependency matrix based on the aggregated inter-terminal directional communication traffic includes: obtaining the directional communication traffic windows of the most recent W time slices based on the aggregated inter-terminal directional communication traffic; performing window accumulation and normalization processing on each pair of terminals to calculate the dependency evidence quantity; and outputting the dependency evidence quantity as input for generating the elements of the regional communication dependency matrix to construct the regional communication dependency matrix.

[0085] The construction of the regional communication dependency matrix specifically includes: determining the set of time slices during the action window based on the action window marker, and performing time slice localization on the directional communication within the time slice set for the first time exceeding a starting threshold; based on the time slice localization result for the first time exceeding the starting threshold, performing statistical processing on the sequence relationship of the terminal pair consisting of terminal i and terminal j to generate a sequence factor; wherein, the sequence factor is used to characterize the frequency proportion of communication from terminal i to terminal j occurring before other directional communication in multiple action window events; performing synthesis processing based on the dependency evidence quantity and the sequence factor to generate dependency strength, and arranging the dependency strength of each terminal pair according to the regional terminal index to form the regional communication dependency matrix.

[0086] For constructing a structure that relies on the amount of evidence to express the proportion of directions, within the most recent W time-slice windows, for each pair of terminals... Perform window accumulation and normalization to obtain the amount of dependent evidence. For example, calculate using the following formula:

[0087] ;

[0088] in, This represents the amount of evidence indicating the dependency of terminal i on terminal j. Accumulate the directional communication volume within the window; W represents the total outbound traffic of terminal i within the window; W is the window length.

[0089] In this embodiment of the invention, the directional concentration is expressed by the outbound proportion. When the outbound communication of terminal i is consistently concentrated towards terminal j, It will increase significantly, and can distinguish between sporadic communication and stable pointing relationships.

[0090] In one embodiment, for example, a home gateway or authentication service is often accessed centrally by multiple terminals within an action window, at which point multiple... It will maintain a high level in the corresponding direction.

[0091] To enhance directionality by extracting sequential relationship factors within the action window, based on action window labeling... Determine the action window time slice set, and within the set, for each pair of directional communication sequences Perform location processing on the first significant occurrence time to form a sequential relationship factor. .

[0092] The moment of first significant occurrence can be defined as the time when the action window satisfies the following conditions: The earliest time period film, The initial threshold; the sequential relationship factor. It can be defined as occurring within multiple action window events. Earlier Frequency of occurrence or A function representing the average advance of the occurrence time relative to the start of the action window.

[0093] In this embodiment of the invention, the dependency relationship in action-triggered burst scenarios is not only reflected in the proportion of communication volume, but also in the temporal sequence of the startup phase, such as authentication / handshake before media streaming, or gateway establishing a channel before camera uploading. By extracting the sequence factor within the action window, the dilution effect of background communication during non-startup periods on dependency determination can be suppressed.

[0094] In some implementations, the construction form of the sequential relationship factor is not limited. It can be a frequency ratio, a time difference decay function, or a combination of both. Those skilled in the art can choose to implement it based on the statistical granularity available to CPE.

[0095] For synthesized dependency strength and forming a regional communication dependency matrix, based on the amount of dependency evidence. Factors related to sequence Generate dependency strength For example, calculate using the following formula:

[0096] ;

[0097] in, This indicates the strength of the dependency of terminal i on terminal j; This indicates that the values ​​are truncated to the interval [0,1]. The values ​​of each terminal pair within region r are... Arranged according to a unified index, forming a regional communication dependency matrix. .

[0098] In this embodiment of the invention, the proportion evidence and the time sequence evidence are merged into a single computable structure, enabling subsequent steps to derive key nodes at the matrix level and directly participate in bandwidth allocation.

[0099] It should be noted that the above-mentioned dependence strength The calculation method is only one optional synthesis method. In another possible implementation, a weighted superposition form can also be introduced, for example... ,in These are the weighting coefficients; this change does not alter the fundamental idea of ​​using the dependency matrix to express the strength of directional dependencies.

[0100] Specifically, generating a set of key nodes and a set of key edges based on the regional communication dependency matrix includes: performing column summation on each terminal based on the regional communication dependency matrix to calculate the degree of dependency; performing threshold filtering on the degree of dependency to add terminal indices that meet the key node determination criteria to the key node set; performing threshold filtering on the dependency strength in the regional communication dependency matrix to add directed edges that meet the key edge determination criteria to the key edge set; and outputting the set of key nodes and the set of key edges for demand prediction and bandwidth allocation.

[0101] For deriving the set of key nodes and the set of key edges from the dependency matrix, based on the region communication dependency matrix... Perform column summation on each terminal index i to calculate the degree of dependency. For example, calculate using the following formula:

[0102] ;

[0103] in, This indicates the degree to which terminal i is dependent; This represents the strength of the dependency of terminal j on terminal i. Perform threshold filtering to meet the requirements. The terminal index is added to the key node set. .

[0104] At the same time, for matrix elements Perform threshold filtering to meet the requirements. Directed edge Add to key edge set .

[0105] In this embodiment of the invention, the degree of dependence reflects the degree to which the node supports the business actions of others. When multiple terminals stably point to the same node in the action window, the column sum of the node will increase significantly, thus being identified as a key node; the key edge is used to further locate the key dependency link.

[0106] In one embodiment, for example, when multiple terminals access the authentication service first during the membership initiation phase, the authentication service's corresponding node... Significantly increased and included The directed edges from the corresponding terminals to the authentication service will be included. This is used for subsequent hard reserve and priority protection.

[0107] S4: Generate the next time slice terminal demand prediction value based on the action window marker and the terminal short-cycle bandwidth sequence.

[0108] It should be noted that step S4 is used to generate the terminal demand forecast value for the next time slice. This allows for actionable predictive inputs on short-cycle demand changes before resource allocation. This is achieved by referencing the output of step S1. and reference the action window marker output in step S2. The prediction for the start-up phase has been revised upwards.

[0109] Specifically, a basic forecast value is set for each terminal based on the terminal short-cycle bandwidth sequence, and an exponential smoothing update process is performed on the basic forecast value; an upward adjustment or maintenance process is performed on the basic forecast value based on the action window flag to obtain the demand forecast value.

[0110] To maintain the base forecast and perform exponential smoothing updates, a base forecast is maintained for each terminal. And perform exponential smoothing updates as follows:

[0111] ;

[0112] in, This represents the base forecast value for time slice t; This refers to the actual bandwidth rate. This is the base forecast value for the previous time slice; This is the smoothing coefficient.

[0113] In this embodiment of the invention, the latest observations are given higher weights to improve response speed, while historical terms are retained to suppress single-point fluctuations, so that the predicted sequence has stability and interpretability in a short period.

[0114] In some implementations, λ can be a fixed value; in another possible implementation, λ can also be set in stages according to terminal type or bandwidth fluctuation level.

[0115] For predictive upscaling triggered by action window markers and outputting demand prediction values, based on action window markers... Basic forecast value Perform an upward or hold-up adjustment to generate the demand forecast for the next time slice. For example, output as follows:

[0116] ;

[0117] in, This indicates the forecast value for demand in the next time slot; Adjust the coefficient to start the window; Mark the action window.

[0118] In this embodiment of the invention, when the terminal is in the action window, the probability of the terminal entering a higher load in the next time slice is significantly increased. By adjusting the coefficient to amplify the prediction in advance, a more sufficient upper bound for the start-up demand can be provided for the subsequent allocation stage, thereby reducing allocation lag.

[0119] In one embodiment, for example, during the live broadcast startup phase, the bandwidth has not yet reached its peak but is already in a continuous upward trend. If it is 1, then according to Enlarging the image more closely reflects the actual needs of the next time segment, which is beneficial for allocating resources and ensuring availability in advance.

[0120] It should be noted that, The value of is not limited, and those skilled in the art can set different upward adjustment ranges according to different service activation characteristics; in some implementations, it is also possible to Apply no more than An upper bound constraint is imposed to prevent extreme amplification, but this constraint is only an optional limitation at the engineering implementation level.

[0121] S5: Generate terminal bandwidth allocation results based on the demand forecast, the allocable bandwidth limit, the set of key nodes, and the set of key edges, and convert the terminal bandwidth allocation results into executable rate limiting parameters or queue guarantee parameters to take effect in the corresponding time slice.

[0122] It should be noted that step S5 is used to generate terminal bandwidth allocation results by integrating demand forecasting, capacity constraints, and dependency structure. It is then issued as an executable rate limiting parameter or queue guarantee parameter so that it takes effect in the corresponding time slice.

[0123] Specifically, generating terminal bandwidth allocation results based on the demand forecast, the allocable bandwidth limit, the set of key nodes, and the set of key edges includes: calculating hard-reserved quotas for key terminals based on the demand forecast, the allocable bandwidth limit, and the action window markers to generate hard-reserved quotas; calculating the remaining capacity based on the allocable bandwidth limit and the hard-reserved quotas; and performing proportional allocation processing on non-key terminals based on the remaining capacity and the demand forecast to generate terminal bandwidth allocation results.

[0124] To determine the set of key terminals and establish key mapping relationships, based on the set of key nodes... With the set of key edges Determine the set of key terminals under the c-th CPE Non-critical terminal set .

[0125] Wherein, if the region index corresponding to terminal d belongs to , or terminal d participate If any endpoint of a key directed edge is included, then it is assigned to... This mapping relationship is used to ensure that the effects of subsequent hard reservations and proportional allocations on key objects are clear and traceable.

[0126] It should be noted that the rules for determining the set of key terminals are not limited to the one form described above; in some implementations, they may also be based solely on... In another possible implementation, to determine the key terminal set, it is also possible to... Endpoints with higher edge weights are given higher priority, but this does not change the basic idea of ​​deriving key objects based on the dependency matrix and using them for allocation.

[0127] For calculating hard-reserved quotas for key terminals within the action window, based on demand forecasts... Allocable bandwidth limit and action window markers Hard-reserved quota for computing on key terminals For example, calculate using the following formula:

[0128] ;

[0129] in, This represents the hard-reserved quota calculated for terminal d in time slice t; For hard-reserve ratio; This represents the maximum allocatable bandwidth. This represents the demand forecast.

[0130] In this embodiment of the invention, by prioritizing the locking of a portion of the bandwidth share of critical terminals within the startup window, critical dependent links are prevented from being interrupted due to contention at the start of a burst, thereby avoiding the failure of overall business operations caused by the restriction of critical nodes.

[0131] In one embodiment, when multiple terminals simultaneously initiate the membership application process, the authentication service node is highly relied upon and thus included. The corresponding terminal obtains a hard reservation within the action window. This ensures that the authentication link remains available before congestion occurs.

[0132] To calculate the remaining capacity and allocate it proportionally to non-critical terminals, after calculating the hard reserve quota for each terminal, the remaining capacity is calculated first. For example, calculate using the following formula:

[0133] ;

[0134] in, Indicates the remaining capacity; Total hard reserve quota.

[0135] Subsequently, based on the remaining capacity, bandwidth is allocated to non-critical terminals according to the predicted demand ratio, generating terminal bandwidth allocation results. For example, for Calculate using the following formula:

[0136] ;

[0137] in, This represents the allocation result of terminal d in time slice t; ϵ is a zero-prevention constant; This represents the sum of predicted demand from non-critical terminals.

[0138] In this embodiment of the invention, while ensuring the hard reservation of key terminals, the remaining capacity is allocated according to the demand ratio, making the allocation result interpretable and feasible, and ensuring that the total allocation does not exceed [a certain limit]. .

[0139] It should be noted that those skilled in the art can also use equivalent weighting methods to achieve the allocation, such as using static weights as the proportional factor.

[0140] Specifically, the terminal bandwidth allocation result is converted into executable rate limiting parameters or queue guarantee parameters to take effect in the corresponding time slice. This includes: marking the trigger ratio of the statistical region based on the action window and performing regional throttling and scaling processing on the ratio allocation of non-critical terminals; converting the terminal bandwidth allocation result into token bucket rate or queue minimum guarantee parameters and issuing them so that the parameters take effect in the next time slice.

[0141] The process involves converting the allocation results into executable parameters and issuing them to make them effective, thus affecting the terminal bandwidth allocation results. Convert the rate limiting parameters or queue guarantee parameters that can be executed by the CPE, and make them effective in the next time slice.

[0142] In one embodiment, each terminal can be corresponding to The parameters are mapped to the token bucket shaping rate, burst bucket size, or queue minimum guaranteed rate; then the parameters are sent to the CPE's flow control module or queue scheduling module to complete the update.

[0143] In this embodiment of the invention, the rate target output by the algorithm is implemented as control parameters that can be executed by the network device, so that the dynamic allocation result can take effect quickly at the time slice boundary.

[0144] In another possible implementation, if the CPE supports setting queues separately for terminals and critical edges, the directional communication corresponding to the critical edge can be further mapped to independent queues for protection.

[0145] Reference Figure 2 , Figure 2 This is a schematic diagram of the CPE bandwidth dynamic allocation system based on user behavior prediction, according to an embodiment of the present invention.

[0146] like Figure 2 As shown, the CPE bandwidth dynamic allocation system based on user behavior prediction proposed in this embodiment of the invention includes:

[0147] The acquisition module 10 is used to acquire the time slice communication volume of each terminal under each CPE according to the preset time slice length and generate the terminal short-cycle bandwidth sequence, acquire the allocable bandwidth limit of each CPE, and generate the inter-terminal directional communication volume summary based on the time slice directional communication volume statistics.

[0148] The identification module 20 is used to identify the burst execution action window of the terminal based on the short-period bandwidth sequence of the terminal, so as to obtain the action window mark;

[0149] The construction module 30 is used to construct a regional communication dependency matrix based on the summary of inter-terminal directional communication traffic, and to generate a set of key nodes and a set of key edges based on the regional communication dependency matrix.

[0150] The generation module 40 is used to generate a predicted value of the terminal demand for the next time slice based on the action window marker and the terminal short-cycle bandwidth sequence.

[0151] The allocation module 50 is used to generate terminal bandwidth allocation results based on the demand forecast value, the allocable bandwidth limit, the set of key nodes and the set of key edges, and convert the terminal bandwidth allocation results into executable rate limiting parameters or queue guarantee parameters to take effect in the corresponding time slice.

[0152] Other embodiments or specific implementations of the CPE bandwidth dynamic allocation system based on user behavior prediction of the present invention can be referred to the above-described method embodiments, and will not be repeated here.

[0153] It is understood that in the description of this specification, references to terms such as "one embodiment," "another embodiment," "other embodiments," or "first embodiment to Nth embodiment," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0154] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0155] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for dynamic allocation of CPE bandwidth based on user behavior prediction, characterized in that, The method includes the following steps: The time-slice communication volume of each terminal under each CPE is obtained according to the preset time-slice length, and the terminal short-cycle bandwidth sequence is generated. The allocable bandwidth limit of each CPE is obtained, and the inter-terminal directional communication volume is summarized based on the time-slice directional communication volume statistics. Based on the terminal's short-period bandwidth sequence, a burst execution action window identification process is performed on the terminal to obtain an action window marker; specifically including: The bandwidth subsequence of the terminal for the most recent time slices is obtained based on the sliding window cache, and the relative burst ratio is calculated on the bandwidth subsequence to obtain the burst ratio; Perform a continuously rising count operation on the bandwidth subsequence to obtain a continuously rising count; A startup feature score is generated based on the surge ratio and the continuous rise count, and a threshold persistence determination process is performed on the startup feature score to obtain an action window marker. Specifically, when the activation feature score exceeds the threshold and continues for at least a preset number of time slices, the action window is marked as 1; otherwise, the action window is marked as 0. A regional communication dependency matrix is ​​constructed based on the aggregated inter-terminal directional communication traffic, and a set of key nodes and a set of key edges are generated based on the regional communication dependency matrix. Based on the action window marker and the terminal short-cycle bandwidth sequence, generate the predicted terminal demand value for the next time slice; Based on the demand forecast, the allocable bandwidth limit, the set of key nodes, and the set of key edges, a terminal bandwidth allocation result is generated, and the terminal bandwidth allocation result is converted into executable rate limiting parameters or queue guarantee parameters to take effect in the corresponding time slice.

2. The CPE bandwidth dynamic allocation method based on user behavior prediction as described in claim 1, characterized in that, The system acquires the time-slice communication volume of each terminal under each CPE according to the preset time-slice length and generates a short-cycle bandwidth sequence for the terminal. It also acquires the allocatable bandwidth limit for each CPE and generates a summary of inter-terminal directional communication volume based on the time-slice directional communication volume statistics. Specifically, this includes: For each terminal of each CPE, the cumulative number of bytes in each time slice is obtained, and the time slice rate is converted on the cumulative number of bytes to generate a short-cycle bandwidth sequence for the terminal. A sliding window buffer write process is performed on the short-cycle bandwidth sequence of the terminal to form bandwidth sequence segments of the most recent time slices; The allocable bandwidth limit of the CPE is obtained and time-slice aligned storage is performed. Statistical summarization processing is performed on the directional communication traffic between terminals within the time slice to generate a summary of directional communication traffic between terminals.

3. The CPE bandwidth dynamic allocation method based on user behavior prediction as described in claim 1, characterized in that, A regional communication dependency matrix is ​​constructed based on the aggregated inter-terminal directional communication traffic, specifically including: Based on the aggregation of directional communication between terminals, obtain the directional communication windows of the most recent time slices, and perform window accumulation and normalization processing on each pair of terminals to calculate the amount of dependency evidence. The quantity of dependent evidence is output as input to generate the elements of the regional communication dependency matrix, and the regional communication dependency matrix is ​​constructed.

4. The CPE bandwidth dynamic allocation method based on user behavior prediction as described in claim 3, characterized in that, Constructing the regional communication dependency matrix includes: The time slice set during the action window is determined based on the action window marker, and the directional communication within the time slice set is used to perform time slice positioning for the first time exceeding the starting threshold. Based on the time slice positioning result of the first time exceeding the starting threshold, the terminal pair consisting of terminal i and terminal j is subjected to statistical processing of the order relationship to generate the order factor; The sequence factor is used to characterize the frequency proportion of communication from terminal i to terminal j before communication in other directions in multiple action window events. Based on the amount of dependency evidence and the prior factors, a synthesis process is performed to generate dependency strength. The dependency strengths of each terminal pair are then arranged according to the regional terminal index to form a regional communication dependency matrix.

5. The CPE bandwidth dynamic allocation method based on user behavior prediction as described in claim 4, characterized in that, The generation of a set of key nodes and a set of key edges based on the aforementioned regional communication dependency matrix specifically includes: Based on the regional communication dependency matrix, column summation is performed on each terminal to calculate the degree of dependency. Threshold filtering is then performed on the degree of dependency to add the terminal index that meets the key node determination criteria to the key node set. A threshold filtering process is performed on the dependency strength in the regional communication dependency matrix to add directed edges that meet the key edge determination criteria to the key edge set. Output the set of key nodes and the set of key edges for demand prediction and bandwidth allocation.

6. The CPE bandwidth dynamic allocation method based on user behavior prediction as described in claim 1, characterized in that, Based on the action window marker and the terminal short-cycle bandwidth sequence, a predicted value for the terminal demand in the next time slice is generated, specifically including: A basic prediction value is set for each terminal based on the terminal short-cycle bandwidth sequence, and the basic prediction value is then subjected to exponential smoothing update processing. Based on the action window markers, the base forecast value is either adjusted upwards or maintained to obtain the demand forecast value.

7. The CPE bandwidth dynamic allocation method based on user behavior prediction as described in claim 1, characterized in that, The terminal bandwidth allocation result is generated based on the demand forecast, the allocable bandwidth limit, the key node set, and the key edge set, specifically including: Based on the demand forecast, the allocable bandwidth limit, and the action window marker, hard reservation quota is calculated for key terminals to generate hard reservation quota. The remaining capacity is calculated based on the allocable bandwidth limit and the hard reserve quota. For non-critical terminals, a proportional allocation process is performed based on the remaining capacity and the demand forecast value to generate terminal bandwidth allocation results.

8. The CPE bandwidth dynamic allocation method based on user behavior prediction as described in claim 7, characterized in that, The terminal bandwidth allocation result is converted into executable rate limiting parameters or queue guarantee parameters to take effect in the corresponding time slice, specifically including: Based on the action window, the trigger ratio of the statistical area is marked, and the ratio allocation of non-critical terminals is processed by the area throttling scaling. The terminal bandwidth allocation result is converted into a token bucket rate or a queue minimum guarantee parameter and sent out so that the parameter takes effect in the next time slice.

9. A CPE bandwidth dynamic allocation system based on user behavior prediction, characterized in that, The system includes: The acquisition module is used to acquire the time slice communication volume of each terminal under each CPE according to the preset time slice length and generate the terminal short-cycle bandwidth sequence, acquire the allocable bandwidth limit of each CPE, and generate the inter-terminal directional communication volume summary based on the time slice directional communication volume statistics. The identification module is used to identify the burst execution action window of the terminal based on the terminal's short-period bandwidth sequence, so as to obtain the action window marker; specifically, it includes: The bandwidth subsequence of the terminal for the most recent time slices is obtained based on the sliding window cache, and the relative burst ratio is calculated on the bandwidth subsequence to obtain the burst ratio; Perform a continuously rising count operation on the bandwidth subsequence to obtain a continuously rising count; A startup feature score is generated based on the surge ratio and the continuous rise count, and a threshold persistence determination process is performed on the startup feature score to obtain an action window marker. Specifically, when the activation feature score exceeds the threshold and continues for at least a preset number of time slices, the action window is marked as 1; otherwise, the action window is marked as 0. The construction module is used to construct a regional communication dependency matrix based on the summary of inter-terminal directional communication traffic, and to generate a set of key nodes and a set of key edges based on the regional communication dependency matrix. The generation module is used to generate a predicted value of terminal demand for the next time slice based on the action window marker and the terminal short-cycle bandwidth sequence. The allocation module is used to generate terminal bandwidth allocation results based on the demand forecast value, the allocable bandwidth limit, the set of key nodes, and the set of key edges, and to convert the terminal bandwidth allocation results into executable rate limiting parameters or queue guarantee parameters to take effect in the corresponding time slice.

Images