An OTA traffic dynamic expansion method and system

Abstract

The application discloses an OTA traffic dynamic expansion method and system, relates to the field of communication, and realizes active monitoring of states such as remaining amount of a package, terminal activity and network switching by deploying an Applet module in a user SIM card, and generates quantitative state data. When a preset trigger condition is met, an expansion request is initiated to an operator cloud platform. The platform combines user historical behaviors, location information and network load, performs traffic peak value prediction and expansion level determination based on a multi-task graph reasoning model. After evaluating expansion benefits through a cause-effect reasoning module, OTA instructions are generated and delivered. The SIM card temporarily authorizes and records feedback accordingly, and realizes accurate, moderate and intelligent expansion control.

Description

TECHNICAL FIELD

[0001] The present specification relates to the field of communications, and more specifically, to an OTA traffic dynamic expansion method and system. BACKGROUND

[0002] With the rapid development of mobile Internet business, users' demand for instantaneity and flexibility of traffic resources is growing. Traditional traffic packages usually use a fixed quota preset mechanism. Once the user approaches the traffic limit, the user needs to manually purchase additional traffic through the client, which causes experience interruption, response lag, and resource waste. Some operators support issuing configuration instructions through the OTA (Over-The-Air) channel to adjust the package, but the related mechanism relies on user-initiated requests or customer service manual intervention, and lacks automatic perception and prediction ability for the actual state of the user.

[0003] On the other hand, existing traffic expansion mechanisms are usually triggered based on static strategies or rules, such as triggering expansion when the remaining traffic is below a fixed threshold, but such mechanisms do not consider user behavior characteristics, network load status, or scenario context information, which can lead to inaccurate expansion and unreasonable authorization. In addition, for new users or data sparse users, the system has difficulty in accurately modeling their expansion needs, resulting in technical bottlenecks such as cold start inadaptability and large prediction bias.

[0004] In recent years, although some machine learning models have been introduced on the cloud side to predict traffic behavior, there are still the following shortcomings: (1) Most models do not have the ability to jointly model spatial mobile patterns and time trends, making it difficult to capture the behavior evolution rules of users in different areas and time periods; (2) The expansion strategy development process does not fully utilize user experience feedback after expansion, making it impossible to achieve experience-driven strategy optimization; (3) There is a lack of quantitative evaluation mechanism for whether it is "worth expanding" after expansion, which can lead to redundant use of resources or ineffective expansion that does not actually improve user experience.

[0005] Therefore, there is an urgent need for a traffic dynamic expansion method that can realize end-to-cloud collaborative perception, support scenario understanding and adaptive control, and take into account the adaptation ability of new users and the evaluation of experience benefits, in order to improve operational efficiency and user satisfaction. SUMMARY

[0006] A series of simplified concepts are introduced in the summary section, which will be further described in detail in the specific embodiments section. The summary section of the present application does not mean to attempt to limit the key features and necessary technical features of the claimed technical solutions, nor does it mean to attempt to determine the protection scope of the claimed technical solutions.

[0007] In a first aspect, the present application proposes an OTA traffic dynamic expansion method, comprising:

[0008] An Applet module for monitoring traffic usage status is deployed in a user SIM card to obtain local package usage information and generate quantization status data, wherein the Applet module performs event counting and window accumulation on high-dynamic data including network switching and session frequency through a lightweight statistical processing method, and forms low-dimensional quantization indicators locally to reduce SIM card computing overhead;

[0009] When the quantization status data meets a preset triggering condition, the Applet module uploads a capacity expansion request to an operator cloud platform through a terminal;

[0010] The operator cloud platform generates a capacity expansion decision based on the capacity expansion request, in combination with historical traffic usage data, user behavior characteristics, geographic location information, and current network load, performs traffic demand prediction and credit policy evaluation;

[0011] An OTA instruction with temporary capacity expansion parameters is generated according to the capacity expansion decision;

[0012] The OTA instruction is delivered to the SIM card through an OTA channel, so that the SIM card updates the internal package configuration according to the instruction and starts temporary traffic authorization;

[0013] During the temporary authorization period, the SIM card records the actual usage and periodically reports to the operator cloud platform to support dynamic adjustment or renewal control.

[0014] In a feasible implementation, the obtaining of the local package usage information and the generation of the quantization status data include:

[0015] The remaining traffic quota is determined by reading the internal package file of the user SIM card;

[0016] Terminal environment switching information is obtained, wherein the terminal environment switching information includes network standard switching records and cell ID change frequency;

[0017] The user data session start frequency is captured;

[0018] The terminal start cycle change information is recorded, wherein the start cycle change information includes changes in active cycle and sleep cycle;

[0019] The quantization status data is generated by fusing the remaining traffic quota, the terminal environment switching information, the user data session start frequency, and the start cycle change information according to pre-defined weights.

[0020] In a feasible implementation, it further includes:

[0021] In a high mobility scenario, the weight of the terminal environment switching information is increased.

[0022] In the fixed hotspot area scenario, the weight of the above data session initiation frequency and the above initiation period change information is increased;

[0023] In the night low activity scenario, the weight of the above remaining traffic quota is increased.

[0024] In an available implementation, the expansion request includes the above quantized state data,

[0025] The operator cloud platform, based on the expansion request, combines historical traffic usage data, user behavior characteristics, geographic location information, and current network load to perform traffic demand prediction and credit policy evaluation, and generates expansion decisions, including:

[0026] The operator cloud platform, based on the quantized state data, historical traffic usage data, user behavior characteristics, geographic location information, and current network load, performs prediction through a dynamic attention-guided multi-task graph reasoning model to generate expansion decisions including user future traffic peak, expansion emergency level, and recommended authorized effective period;

[0027] The dynamic attention-guided multi-task graph reasoning model includes:

[0028] A behavior graph construction module constructs a behavior event graph according to the user behavior characteristics, and injects the geographic location information and the network load as node attributes into the graph structure;

[0029] A graph attention processing module is used to perform graph neural network processing on the behavior event graph to extract a spatial behavior evolution tensor;

[0030] A time series feature extraction module is used to perform multi-channel time convolution on the quantized state data and the historical traffic usage data to extract a time series feature tensor;

[0031] A multi-task decoding module is used to input the spatial behavior evolution tensor and the time series feature tensor after fusion into a multi-task decoding structure to generate expansion decisions including user future traffic peak, expansion emergency level, and recommended authorized effective period.

[0032] In an available implementation, the graph constructed by the behavior graph construction module is a heterogeneous graph structure, the node types of the heterogeneous graph structure include user behavior events, geographic location nodes, time nodes, network load level nodes, and package level nodes, and the edges of the heterogeneous graph structure represent behavior transitions and scenario causal relationships;

[0033] The behavior graph construction module includes a dynamic node generation unit that generates corresponding temporary nodes in real time and injects them into the heterogeneous graph structure when a user behavior mutation event is detected.

[0034] In a feasible implementation, the graph attention processing module adopts a spatio-temporal graph convolution network structure to simultaneously capture the temporal evolution rule and spatial dependency of the behavior graph.

[0035] The behavior offset extracted by the graph attention processing module is also used to drive an adaptive authorization strategy. When a high offset scenario is detected, the expansion parameter is automatically adjusted, including extending the authorization validity period or increasing the expansion amount.

[0036] In a feasible implementation, the graph attention processing module includes a main processing branch for extracting a current user behavior graph embedding representation and an auxiliary branch for extracting a same-type user group behavior graph embedding. The feature difference between the main processing branch and the auxiliary branch represents a behavior offset, which is used to improve the determination result of the expansion emergency level.

[0037] In a feasible implementation, the multi-task decoding module integrates a reinforcement learning mechanism based on an Actor-Critic architecture.

[0038] The Actor network is used to input the spatial behavior evolution tensor and the temporal feature tensor into the multi-task decoding structure after fusion, to generate a recommended authorization validity expansion decision including a user future traffic peak, an expansion emergency level, and an expansion decision.

[0039] The Critic network is used to receive the current state feature and the expansion decision to estimate its state value function.

[0040] The cloud platform constructs a reward signal based on the behavior feedback information of the user after expansion, including whether the expansion data is exhausted, the business usage frequency, the user subjective interaction behavior, and the abnormal usage indicators, to simultaneously optimize the parameters of the Actor network and the Critic network.

[0041] In a feasible implementation, the Actor-Critic architecture integrates a meta-learning mechanism. In the scenario of a new user or a cold start user, the cloud platform performs graph comparison learning based on the structural features of the user behavior graph and a set of historical behavior graphs to calculate a similarity score, and selects a plurality of optimal matching historical users as a source strategy candidate set. Strategy distillation and local parameter freezing mechanisms are used to transfer the candidate source strategy to the current Actor network.

[0042] A dynamic behavior offset detection module is introduced during the fusion migration process. The migration adaptability is evaluated according to the strategy behavior change rate of the new user. When the adaptability is lower than a preset threshold, a strategy fine-tuning and freezing control process is automatically triggered.

[0043] In an implementable embodiment, the multi-task graph reasoning model further comprises a causal reasoning decision inhibition module, and the specific steps of the causal reasoning decision inhibition module comprise:

[0044] A causal graph structure is established based on user historical expansion behaviors, post-expansion traffic usage and experience indicators, and the causal graph structure comprises expansion behavior processing variables, experience indicator result variables, covariate and mediator variable nodes;

[0045] Expansion-non-expansion samples and corresponding multi-dimensional experience feedback data are extracted from a user group for causal reasoning model training;

[0046] Based on the current user state, the causal reasoning model is used to predict the expected scores of user experience under the execution of expansion and non-expansion respectively, and the experience improvement value between the two is calculated, wherein the experience improvement value comprises a double-layer causal evaluation system result, and the double-layer causal evaluation system result comprises an immediate experience improvement result of the user after expansion based on immediate experience indicators, and a long-term value improvement result of the long-term value evaluation of expansion based on user retention rate and package upgrade tendency indicators;

[0047] The experience improvement value is compared with a preset intervention benefit threshold;

[0048] When the experience improvement value is lower than the preset intervention benefit threshold, the generation of the current expansion instruction is automatically inhibited, or the expansion amount and the authorized validity period are reduced.

[0049] In a second aspect, the present application provides an OTA traffic dynamic expansion system, comprising:

[0050] An acquisition unit is configured to deploy an Applet module for monitoring traffic usage state in a user SIM card to acquire local package usage information and generate quantitative state data;

[0051] An uploading unit is configured to upload an expansion request to an operator cloud platform through a terminal by the Applet module when the quantitative state data meets a preset triggering condition;

[0052] A first generation unit is configured to perform traffic demand prediction and credit strategy evaluation based on the expansion request, historical traffic usage data, user behavior characteristics, geographic location information and current network load by the operator cloud platform to generate an expansion decision;

[0053] A second generation unit is configured to generate an OTA instruction with temporary expansion parameters according to the expansion decision;

[0054] A starting unit is configured to issue the OTA instruction to the SIM card through an OTA channel to enable the SIM card to update the internal package configuration according to the instruction and start temporary traffic authorization.

[0055] A monitoring unit is configured to record actual usage in the temporary authorization period, and periodically report to the operator cloud platform to support dynamic adjustment or renewal control.

[0056] In summary, the application actively monitors multiple dimensions of state, including remaining quota, data session frequency, terminal activity period, network environment switching, and generates a quantitative state vector, to achieve end-side prediction and avoid relying entirely on user manual triggering, significantly improving the automation level and timeliness of traffic services. The operator cloud platform integrates user historical traffic behavior, current state, geographic location, and network load information to build a dynamic attention-guided multi-task graph reasoning model, combining graph neural networks and temporal convolution network structures to predict user future traffic demand and emergency level from spatial evolution and temporal trend, and generate more refined and scenario-aware expansion strategies. Before generating the expansion instruction, the causal reasoning module simulates the user experience results under the condition of "expansion or not", and if expansion cannot significantly improve the experience, the expansion or authorization level is automatically suppressed, solving the problem of expansion behavior and user perception disconnection in traditional methods and improving resource utilization efficiency. The multi-task decoding module introduces a reinforcement learning mechanism based on the Actor-Critic architecture, using user actual usage behavior after expansion (such as whether to exhaust the authorized quota, subjective satisfaction behavior, abnormal traffic patterns, etc.) as a reward signal to update the model strategy, forming a "perception-prediction-feedback-optimization" closed-loop process, significantly improving the adaptive ability and long-term service quality of the strategy. By building a user behavior graph comparison learning mechanism, similar historical user strategy networks can be quickly matched in new user scenarios, combined with strategy distillation and local freezing migration, solving the cold start problem of traditional models when data is insufficient, and expanding the model's application range. During the expansion authorization period, the SIM card continuously records traffic usage and periodically returns to the cloud platform to determine whether to terminate, renew, or adjust the quota, achieving dynamic resource control and avoiding over-authorization. In summary, the application combines end-side active perception, cloud-side deep prediction, multi-modal information fusion, and causal decision inhibition to significantly improve the system's intelligence, expansion efficiency, and user experience guarantee capability.

[0057] The OTA traffic dynamic expansion method proposed in the application, other advantages, objectives and features of the application will be partially embodied through the following description, and some will be understood by those skilled in the art through research and practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0058] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments with reference made to the accompanying drawings. The drawings are for purposes of illustration only and are not intended to be limiting in

[0059] Figure 1 A flow chart of an OTA traffic dynamic expansion method provided by an embodiment of the present application;

[0060] Figure 2 A flow chart of a method for obtaining local package usage information and generating quantization state data provided by an embodiment of the present application;

[0061] Figure 3 A flow chart of specific steps of a causal reasoning decision inhibition module provided by an embodiment of the present application;

[0062] Figure 4 A structural schematic diagram of an OTA traffic dynamic expansion system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0063] The terms "first", "second", "third", "fourth" and the like in the description and in the claims of the present application, if any, are used for distinguishing between similar objects talking about the embodiments and do not necessarily have a particular chronological, sequential or hierarchical order. It is to be understood that the data so designated are interchangeable under appropriate circumstances so that the embodiments of the application described herein are practiced in the embodiments in other than the illustrated or described order. Additionally, the terms "comprise", "comprising", "include", "including", and the like, are meant to be inclusive in a manner that the processes, methods, articles, or apparatuses that include a series of steps or units are not to be limited to those strictly enumerated in the specification and claims of the present application but are meant to include any steps or units that are related thereto or that are mutatis mutandis to the process, method, article, or apparatus specifically enumerated in the specification and claims of the present application. The following detailed description of the embodiments of the present application will be made with reference to the accompanying drawings, which are provided for the purpose of illustration and description only and are not intended to be limiting.

[0064] Reference will now be made to Figure 1 A flow chart of an OTA traffic dynamic expansion method provided by an embodiment of the present application, which can specifically include:

[0065] S110, deploying an Applet module for monitoring traffic usage status in a user SIM card to obtain local package usage information and generate quantization status data, wherein the Applet module performs event counting and window accumulation on high dynamic data including network switching and session frequency through a lightweight statistical processing method, and forms low-dimensional quantization indicators locally to reduce SIM card computing overhead;

[0066] S120, when the quantization status data meets a preset triggering condition, the Applet module uploads a capacity expansion request to an operator cloud platform through a terminal;

[0067] S130, the operator cloud platform generates a capacity expansion decision based on the capacity expansion request, in combination with historical traffic usage data, user behavior characteristics, geographic location information and current network load, performs traffic demand prediction and credit policy evaluation;

[0068] S140, generating an OTA instruction with temporary capacity expansion parameters according to the capacity expansion decision;

[0069] S150, issuing the OTA instruction to the SIM card through an OTA channel, so that the SIM card updates the internal package configuration according to the instruction and starts temporary traffic authorization;

[0070] S160, during the temporary authorization period, the SIM card records the actual usage and periodically reports to the operator cloud platform to support dynamic adjustment or renewal control.

[0071] For example, the following is a specific embodiment of the present application, which aims to provide a traffic dynamic expansion method based on OTA, combining user SIM card side status sensing and intelligent prediction mechanism of the operator cloud platform, to realize on-demand, real-time and trusted authorization control of user package traffic, specifically including the following processes:

[0072] First, a Applet module with active state awareness capability is pre-deployed in the user's SIM card, which has the ability to periodically or event-drivenly read the local package state information. Specifically, the Applet can extract the current package remaining traffic quota from the file structure inside the SIM card, identify the current package period remaining time, and simultaneously listen to the terminal side network mode switching record, cell ID switching frequency, user data session start frequency, and terminal active and dormant period change behavior. The above multi-dimensional state data is fused according to the preset weight to generate "quantitative state data", which comprehensively reflects the user's current traffic use pressure and context network environment characteristics in numerical form. The Applet module performs event counting and window accumulation on high-dynamic data including network switching and session frequency through a lightweight statistical processing method, and forms low-dimensional quantitative indicators locally, thereby effectively reducing the calculation and storage overhead of the SIM card while ensuring monitoring accuracy.

[0073] When the above quantitative state data meets one or more trigger conditions (for example: remaining traffic is lower than a threshold, data usage frequency abnormally increases, network environment frequently switches, etc.), the Applet will actively construct a capacity expansion request message and report it to the operator cloud platform through the terminal communication link. The capacity expansion request carries the generated quantitative state data and user identification information, which is used to support the subsequent decision-making process of the cloud platform.

[0074] After receiving the capacity expansion request, the operator cloud platform starts the intelligent prediction and strategy evaluation process. The platform extracts the traffic consumption pattern, typical active period, and usage preference from the user's historical traffic usage records, and combines the quantitative state data carried in the capacity expansion request, as well as the user's current geographic location information and access network load level, as input features into a dynamic attention-guided multi-task graph reasoning model.

[0075] The model first constructs a heterogeneous graph structure according to the user's behavior event stream through the behavior graph construction module. The graph contains user behavior nodes, geographic location nodes, network load level nodes, and package level nodes, etc. The nodes are connected by behavior transition paths and context causal relationships, and real-time environmental attributes are injected. The graph attention processing module performs graph neural network reasoning on the graph to generate a spatial behavior evolution tensor; at the same time, the time series feature extraction module uses a multi-channel time convolution structure to extract a time series trend feature tensor from the current state data and historical traffic usage sequence. The multi-task decoding module fuses the above two types of tensors and outputs three types of prediction results: user's future traffic peak prediction value, capacity expansion urgency level prediction value, and recommended temporary authorization validity period.

[0076] If the user's current state behavior deviates from its historical behavior trajectory, the platform can further enable a causal reasoning suppression module that builds a causal graph structure based on the user's historical capacity expansion and experience feedback indicators (such as download success rate, webpage response delay, video stutter rate, etc.), and estimates the counterfactual value of experience improvement for both "expansion" and "non-expansion" paths under the current state. If this value is lower than the set intervention benefit threshold, the platform will suppress the current expansion request or automatically downgrade the authorization level.

[0077] After the platform finally confirms to allow expansion, an OTA instruction containing temporary expansion parameters (such as additional traffic quota, limited time window, usage area, etc.) will be generated based on the above prediction results. The OTA instruction is sent to the user terminal through the OTA channel (such as SMS, BIP or HTTP channel), and the terminal guides the SIM card to receive the instruction. After receiving the OTA instruction, the Applet module first performs digital signature verification and authorization ID verification to ensure the legality of the instruction; if the verification is passed, the SIM card automatically updates its internal package configuration according to the instruction content, and activates the temporary expansion state.

[0078] During the temporary expansion period, the Applet module continuously monitors the actual usage behavior of the authorized traffic, constructs a usage feedback vector, and the feedback content includes the usage proportion of the expansion quota, the time period distribution, whether there is resource waste or sudden outbreak, network environment stability, and whether the user triggers the request again during the authorized period. The Applet module reports the feedback vector to the cloud platform according to the strategy set period (such as 30 minutes or every 10% usage proportion). The platform executes dynamic adjustment logic based on the feedback, including: judging whether to extend the expansion validity period; whether to terminate the expansion state in advance; whether to include the current behavior into the reward signal of the reinforcement learning model for optimizing subsequent strategies; whether to update the "expansion credit level" and behavior label in the user portrait.

[0079] In the case of new users or cold start users, to improve the accuracy of the prediction model when the initial data is insufficient, the platform further introduces a reinforcement learning mechanism based on the Actor-Critic architecture, and integrates a meta-learning module. Specifically, the platform identifies a set of historical users similar to the behavior graph structure of the new user through graph contrastive learning mechanism, forms a candidate set of transfer strategies, and migrates the strategy network to the initial parameters of the Actor network of the current user through strategy distillation and local freezing. The platform also evaluates the strategy adaptability based on the user behavior change rate, and can automatically trigger fine-tuning or freezing control mechanism when the transferred strategy is not effective, to ensure the rapid and stable convergence of the strategy for new users.

[0080] In summary, the application actively monitors multiple dimensions of state including remaining balance, data session frequency, terminal activity period, network environment switching, and generates a quantitative state vector, to achieve an end-side foreknowledge of expansion demand, avoiding complete reliance on user manual triggering, and significantly improving the automation level and timeliness of traffic services. The operator cloud platform integrates user historical traffic behavior, current state, geographic location, and network load information to build a dynamic attention-guided multi-task graph reasoning model, combining graph neural networks and time convolution network structures to predict user future traffic demand and emergency level from both spatial evolution and temporal trends, and to generate more refined and scenario-aware expansion strategies. Before generating the expansion instruction, the causal reasoning module simulates the counterfactual scenario of user experience under expansion or not, and if expansion cannot significantly improve experience, it automatically suppresses expansion or reduces the authorized level, addressing the problem of expansion behavior and user perception disconnection in traditional methods, and improving resource utilization efficiency. The multi-task decoding module introduces a reinforcement learning mechanism based on the Actor-Critic architecture, using user actual usage behavior after expansion (such as whether to exhaust authorized quota, subjective satisfaction behavior, abnormal traffic patterns, etc.) as a reward signal to update the model's strategy, forming a closed-loop process of "perception-prediction-feedback-optimization", significantly improving the adaptive ability and long-term service quality of the strategy. By building a user behavior graph comparison learning mechanism, the strategy network of similar historical users can be quickly matched in new user scenarios, combined with strategy distillation and local freezing migration, to solve the cold start problem of traditional models when data is insufficient, expanding the model's application scope. During the expansion authorization period, the SIM card continuously records traffic usage and periodically returns to the cloud platform to determine whether to terminate, renew, or adjust the quota, achieving dynamic resource regulation and avoiding over-authorization problems. In summary, the application combines end-side active perception, cloud-side deep prediction, multi-modal information fusion, and causal decision inhibition to significantly improve the system's intelligence, expansion efficiency, and user experience guarantee capability.

[0081] In a feasible implementation manner, as shown in Figure 2 The step S110 of acquiring local package usage information and generating quantitative state data includes:

[0082] S1101, read the internal package file of the user SIM card to determine the remaining traffic quota;

[0083] S1102, acquire terminal environment switching information, wherein the terminal environment switching information includes network standard switching records and cell ID change frequency;

[0084] S1103, capture user data session start frequency;

[0085] S1104, record the start period change information of the terminal, wherein the start period change information includes the change of the active period and the sleep period;

[0086] S1105, generate the quantization state data according to the pre-defined weight fusion of the change of the residual flow quota, the terminal environment switching information, the user data session start frequency and the start period change information.

[0087] In a feasible implementation, further comprising:

[0088] In a high mobility scenario, the weight of the terminal environment switching information is increased;

[0089] In a fixed hotspot area scenario, the weight of the data session start frequency and the start period change information is increased;

[0090] In a night low activity scenario, the weight of the residual flow quota is increased.

[0091] Exemplarily, in order to realize the intelligent perception of the user flow usage state and the pre-judgment mechanism of triggering the expansion request, an Applet is deployed in the user SIM card, which is used to obtain the multi-dimensional state indicators on the terminal side, and generate quantization state data reflecting the current usage situation. The process includes the following specific steps:

[0092] Firstly, the Applet module performs the following sub-process after starting on the SIM card side:

[0093] S1101, read the remaining quota information: the Applet module extracts the remaining flow quota, used flow, and current package validity period of the current package by accessing the EF-DATA-CAPACITY file or the exclusive service file defined by the operator in the SIM card, which is used to evaluate the usage progress of the user package.

[0094] S1102, obtain the terminal environment switching information: the Applet module listens to the network environment parameters reported by the terminal, including the network mode switching record (such as the frequency of 5G to 4G drop switching) and the change frequency of cell ID, so as to judge whether the user is in a high-speed mobile or weak coverage environment. The above information can be transmitted to the SIM card by the terminal through a standard interface (such as CAT, BIP).

[0095] S1103, capture the user data session start frequency: the Applet module records the establishment frequency of the data channel on the terminal side, that is, the trigger frequency of events such as PDP context activation and IP layer data flow start, which is used to represent the data request density and behavior intensity of the user.

[0096] S1104, record the terminal start period change information: the module identifies the current high-frequency wake-up, data pull active use situation by continuously observing the change mode of the terminal activation period and sleep period (for example, the shortening of the activation duration or the increase of the wake-up frequency).

[0097] S1105, fusion generation of quantitative state data: the Applet module standardizes the multiple indexes obtained in S1101 to S1104, and generates a set of multi-dimensional quantitative state data according to the pre-defined weight configuration issued by the operator platform. For example, the following feature vector form can be constructed:

[0098] Q=[R_data, F_switch, F_session, R_awake] x W

[0099] Wherein: R_data represents the remaining traffic proportion of the package, F_switch represents the network mode or cell ID switching frequency in unit time, F_session represents the data session start frequency in unit time, R_awake represents the change intensity of the terminal active period, and W is a dynamic weight vector which is adaptively adjusted according to the current scene.

[0100] Further, in order to improve the environmental adaptability and discrimination ability of the state data, a scene adaptive weight adjustment strategy is introduced in the embodiment, which specifically includes:

[0101] In the high mobility scene, if the cell ID changes quickly and the mode frequently switches, it is determined that the user is currently in a traffic mobile scene (such as high-speed rail, subway, etc.), and the system will automatically increase the weight of the "terminal environment switching information" (F_switch) to enhance the expansion response ability to potential interruption risk;

[0102] In the fixed hotspot area scene, if the user GPS position remains stable for a long time and the data session frequency is continuously high, the system will appropriately increase the weights of "data session start frequency" (F_session) and "activation period change" (R_awake) to identify the high-activity but stable-residence user behavior mode;

[0103] In the night low-activity scene, if the current is in the night period, and the terminal activity and session frequency are significantly reduced, the system will focus on the package balance information (R_data), that is, increase its weight, to ensure that the user will not be interrupted due to package depletion before the next possible activation peak.

[0104] The finally generated quantized state data Q will be uploaded to the cloud platform by the terminal as part of the capacity expansion request, supporting subsequent prediction analysis, capacity expansion strategy formulation and OTA instruction generation. This method not only realizes intelligent state aggregation on the terminal side, but also improves the discriminability and generalization ability of the state vector by combining scene understanding strategies, effectively ensuring the initiative, adaptability and resource accuracy of the capacity expansion behavior.

[0105] In a feasible implementation, the capacity expansion request includes the quantized state data,

[0106] The operator cloud platform, based on the capacity expansion request, combines historical traffic usage data, user behavior characteristics, geographic location information and current network load to perform traffic demand prediction and credit policy evaluation, generating a capacity expansion decision, including:

[0107] The operator cloud platform, based on the quantized state data, historical traffic usage data, user behavior characteristics, geographic location information and current network load, performs prediction through a dynamic attention-guided multi-task graph reasoning model, generating a capacity expansion decision including user future traffic peak, capacity expansion emergency level and recommended authorized validity period;

[0108] The dynamic attention-guided multi-task graph reasoning model includes:

[0109] A behavior graph construction module constructs a behavior event graph according to the user behavior characteristics, and injects the geographic location information and the network load as node attributes into the graph structure;

[0110] A graph attention processing module is used to perform graph neural network processing on the behavior event graph to extract a spatial behavior evolution tensor;

[0111] A time series feature extraction module is used to perform multi-channel time convolution on the quantized state data and the historical traffic usage data to extract a time series feature tensor;

[0112] A multi-task decoding module is used to input the spatial behavior evolution tensor and the time series feature tensor after fusion into a multi-task decoding structure to generate a capacity expansion decision including user future traffic peak, capacity expansion emergency level and recommended authorized validity period.

[0113] In a feasible implementation, the graph constructed by the behavior graph construction module is a heterogeneous graph structure, and the node types of the heterogeneous graph structure include user behavior events, geographic location nodes, time nodes, network load level nodes and package level nodes. The edges of the heterogeneous graph structure represent behavior transitions and scene causal relationships;

[0114] The behavior graph construction module includes a dynamic node generation unit that generates corresponding temporary nodes in real time and injects them into the heterogeneous graph structure when a user behavior mutation event is detected.

[0115] In a feasible implementation, the graph attention processing module adopts a spatio-temporal graph convolution network structure to simultaneously capture the temporal evolution rule and spatial dependency of the behavior graph.

[0116] The behavior offset extracted by the graph attention processing module is also used to drive an adaptive authorization strategy. When a high offset scenario is detected, the expansion parameter is automatically adjusted, including extending the authorization validity period or increasing the expansion amount.

[0117] In a feasible implementation, the multi-task decoding module integrates a reinforcement learning mechanism based on an Actor-Critic architecture.

[0118] The Actor network is used to input the fused spatial behavior evolution tensor and temporal feature tensor into the multi-task decoding structure to generate a recommended authorization validity expansion decision, including user future traffic peak, expansion urgency level, and recommended authorization validity expansion decision.

[0119] The Critic network is used to receive the current state feature and the expansion decision to estimate its state value function.

[0120] The cloud platform constructs a reward signal based on the behavior feedback information of the user after expansion, including whether the expansion data is exhausted, business usage frequency, user subjective interaction behavior, and abnormal usage indicators, to simultaneously optimize the parameters of the Actor network and the Critic network.

[0121] In a feasible implementation, the Actor-Critic architecture integrates a meta-learning mechanism. In the scenario of new users or cold start users, the cloud platform performs graph comparison learning based on the structural features of the user behavior graph and historical behavior graphs to calculate a similarity score, and selects multiple optimal matching historical users as a source strategy candidate set. Strategy distillation and local parameter freezing mechanisms are used to transfer the candidate source strategy to the current Actor network.

[0122] A dynamic behavior offset detection module is introduced during the fusion migration process. The migration adaptability is evaluated according to the strategy behavior change rate of the new user. When the adaptability is lower than a preset threshold, the strategy re-tuning and freezing control process is automatically triggered.

[0123] For example, the behavior graph construction module organizes the multi-source behavior of the user and the scene elements into a heterogeneous graph . Among them, It simultaneously contains multiple types of nodes such as user behavior events, geographic location nodes, time nodes, network load level nodes, and package level nodes. represents the behavior transition and the causal relationship between the scene; is the relationship type adjacency matrix of the graph. For each node attribute vector (including event time, location encoding, load / tariff level, etc.), forming the node feature tensor To improve the ability to characterize sudden behavior, the module has a dynamic node generation unit: when the behavior statistics in the continuous window show abnormal drift (for example, the z-score of the session start frequency, location migration rate, or cell ID switching rate compared to the sliding mean exceeds the threshold ), a temporary node is generated immediately and added with in / out edges according to the detected trigger relationship, with the assigned attributes recording the type of sudden event, the time of occurrence, and the context summary, so that the heterogeneous graph can be updated online.

[0124] The graph attention processing module uses a spatio-temporal graph convolution network (ST-GCN) to jointly model spatial dependence and temporal evolution. Let the input after time expansion be In the first layer, the spatial convolution first propagates each relationship using the symmetric normalized adjacency to get:

[0125] ,

[0126] Then do one-dimensional convolution in the time dimension (kernel length ) to capture sequence dependence:

[0127]

[0128] After stacking layers, the embedding of the current target user subgraph is obtained through the readout operator. At the same time, to improve robustness, the auxiliary representation of the same type of user group (obtained from the matched similar user subgraph through the isomorphic network) is introduced. The difference vector between the two is calculated by .

[0129] Describe the "behavior deviation", which is used by the downstream strategy for adaptive authorization. When a high deviation scenario is detected (for example , the system adjusts the authorization validity period, capacity level, and peak budget according to

[0130]

[0131] , where is the adjustment coefficient.

[0132] The multi-task decoding module integrates an Actor-Critic based reinforcement learning mechanism, which combines the spatial behavior evolution tensor and the temporal feature tensor into a state wherein The quantized state time series can be extracted by TCN / 1D-CNN, The context (such as network load, geographical partition, package constraints). The actor outputs the action as: .

[0133] The critic estimates or The reward is constructed after the actual authorization is issued and the feedback is collected as: .

[0134] wherein, exhausted indicates whether the expansion is reasonably consumed in time, freq is the service usage frequency improvement degree, feedback is the mapping value of user subjective interaction score / complaint reverse signal, anomaly_drop is the abnormal behavior rate or quality drop degree, overgrant is the punishment for waste of over-provisioning; is the weight obtained by experience or offline parameter tuning. The advantage function is used for policy gradient update The critic is trained with TD error.

[0135] To cope with new users / cold start, the actor-critic architecture integrates a meta-learning mechanism. The cloud side first compares the graph of the new user with the graph set of historical users The similarity of the graph comparison learning is:

[0136] and the candidate source policy parameters are weighted and fused to obtain the initialization , ,

[0137] In the distillation stage, the student policy is subjected to a KL constraint with a temperature as: .

[0138] And some layers are parameter frozen to stabilize the migration. During the migration, a dynamic behavior offset detection module is introduced to evaluate the fitness (the smoothing term is ), when automatically trigger the re-fine-tuning of small-step learning rate and the unlocking / locking process of the frozen strategy, until the fitness is restored.

[0139] is the number of nodes, is the number of time steps, is the initial feature dimension; is the adjacency matrix of the relationship , is a degree matrix of the graph, is an identity matrix; is a learnable weight of the first layer pair relationship is a nonlinear activation function; is a nonlinear activation function; are respectively the current user and the graph embedding of the same group, is a behavior offset; is a threshold and intensity of adaptive authorization control; is an Actor parameter, is a Critic parameter; is a reward weight; is a similarity temperature coefficient, is a source policy fusion weight; is a migration fitness index, is a threshold for triggering fine-tuning.

[0140] Through the above mechanism, the system can realize dynamic, personalized and interpretable decision-making of expansion parameters (authorization validity period, expansion level and peak budget) by taking the robust behavior spatiotemporal representation provided by ST-GCN as the core, combining adaptive authorization driven by offset and closed-loop optimization of Actor-Critic, and maintaining good convergence and business effect under cold start and scene mutation.

[0141] In a feasible implementation, the multi-task graph reasoning model further includes a causal reasoning decision inhibition module, as shown in Figure 3 The specific steps of the causal reasoning decision inhibition module include:

[0142] S210, a causal graph structure is established based on user historical expansion behavior, traffic usage and experience indicators after expansion, wherein the causal graph structure includes expansion behavior processing variables, experience indicator result variables, covariate and intermediate variable nodes;

[0143] S220, expansion-non-expansion samples and corresponding multi-dimensional experience feedback data are extracted from the user group for causal reasoning model training;

[0144] S230, based on the current user state, the user experience expected score under the execution of expansion and non-expansion is respectively predicted by the causal reasoning model, and the experience improvement value between the two is calculated, wherein the experience improvement value includes a double-layer causal evaluation system result, and the double-layer causal evaluation system result includes an immediate experience improvement result of the user after expansion based on immediate experience indicators, and a long-term value improvement result of the long-term value evaluation of expansion based on user retention rate and package upgrade tendency indicators;

[0145] S240, the experience improvement value is compared with a preset intervention benefit threshold;​

[0146] S250. When the experience improvement value is lower than the above-mentioned preset intervention benefit threshold, the generation of the current expansion command is automatically suppressed, or the expansion quota and authorization validity period are reduced.

[0147] For example, the multi-task graph inference model also includes a causal inference decision suppression module. This module is used to further determine whether to issue an expansion command after generating an expansion decision by modeling the causal relationship between expansion behavior and user experience, thereby avoiding resource redundancy, invalid expansion, or expansion with negative effects.

[0148] First, the cloud platform constructs a cause-effect graph structure based on users' historical expansion records. This cause-effect graph uses whether a user received an expansion operation as the processing variable. The user experience metrics after the expansion (such as average download speed, service completion rate, user satisfaction score, etc.) are used as the outcome variables. The causal graph is constructed by introducing variables related to user behavior, environment, and package level as covariates (C), and some mediator variables (M) that may be located between expansion and experience. .

[0149] Secondly, representative control samples (expansion / non-expansion) are extracted from the historical user base, and their multi-dimensional experience feedback data is labeled (such as whether users quickly exhaust their quotas after expansion, network congestion response, changes in complaint rate, login frequency, etc.) to construct a causal inference model for training. Models such as Dragonnet (two-branch causal inference network), TARNet (counterfactual contrast network), and GNN-Causal (graph causal inference structure) can be used to model the potential outcomes.

[0150] Then, for the user currently initiating the expansion request, the model inputs its current state feature vector. Calculate the following two results respectively:

[0151] Predicted experience score under expanded capacity:

[0152]

[0153] Predicted experience score without scaling:

[0154]

[0155] Under the two-tiered causal evaluation system, short-term and long-term experience improvements are calculated:

[0156] 1. Short-term benefit improvement (instant experience)

[0157]

[0158] wherein, mainly determined by instant experience indicators, such as download rate, latency, packet loss rate, etc.

[0159] 2. Long-term value improvement

[0160]

[0161] wherein, determined by long-term value indicators, such as user retention rate, package upgrade tendency, long-term satisfaction, etc.

[0162] 3. Comprehensive experience improvement value

[0163] The system will weight and integrate the short-term and long-term improvement results:

[0164]

[0165] wherein, is an adjustable weight, representing the proportion of attention to instant experience and long-term value.

[0166] Subsequently, the improvement value is compared with a preset intervention benefit threshold

[0167] If , it means that the expansion behavior will significantly improve user experience, allowing the current expansion strategy to continue to be executed;

[0168] If , the suppression mechanism is triggered, and one of the following two control methods can be executed: directly suppressing the generation of expansion instructions (i.e. not issuing OTA expansion instructions); or adjusting the authorization parameters of the expansion instruction, such as: reducing the expansion amount, shortening the temporary authorization validity period, or downgrading the expansion level.

[0169] Through the above-mentioned manner, the embodiment not only can predict the expansion demand based on user behavior, but also further judges the actual necessity of expansion through causal analysis, so that the system avoids problems such as over-expansion or inefficient authorization, and improves the intelligence and economy of the expansion decision system.

[0170] As shown in Figure 4 , the present application proposes an OTA traffic dynamic expansion system 10, comprising:

[0171] An acquisition unit 11 is configured to deploy an Applet module for monitoring traffic usage state in a user SIM card, to acquire local package usage information and generate quantitative state data;

[0172] ​The uploading unit 12 is configured to upload a capacity expansion request to the operator cloud platform by the Applet module through the terminal when the quantized state data meets a preset triggering condition.

[0173] The first generating unit 13 is configured to perform traffic demand prediction and credit policy evaluation based on the capacity expansion request, in combination with historical traffic usage data, user behavior characteristics, geographic location information and current network load, and generate a capacity expansion decision by the operator cloud platform.

[0174] The second generating unit 14 is configured to generate an OTA instruction with temporary capacity expansion parameters according to the capacity expansion decision.

[0175] The starting unit 15 is configured to issue the OTA instruction to the SIM card through an OTA channel, so that the SIM card updates the internal package configuration according to the instruction and starts temporary traffic authorization.

[0176] The monitoring unit 16 is configured to record the actual usage of the SIM card within the temporary authorization period and periodically report to the operator cloud platform to support dynamic adjustment or renewal control.

[0177] It can be understood that the system of the second aspect can perform any of the methods.

[0178] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for dynamic capacity expansion based on OTA traffic, characterized in that, include: An applet module for monitoring data usage status is deployed in the user's SIM card to obtain local data plan usage information and generate quantitative status data. The applet module performs event counting and window accumulation on highly dynamic data, including network switching and session frequency, through a lightweight statistical processing method, and forms low-dimensional quantitative indicators locally to reduce the SIM card's computing overhead. When the quantized state data meets the preset triggering conditions, the Applet module uploads a capacity expansion request to the operator's cloud platform through the terminal; Based on the expansion request, the operator cloud platform combines historical traffic usage data, user behavior characteristics, geographic location information, and current network load to perform traffic demand prediction and credit policy evaluation, and generate expansion decisions. OTA instructions with temporary expansion parameters are generated based on the expansion decision; The OTA command is sent to the SIM card through the OTA channel, so that the SIM card updates its internal data plan configuration and initiates temporary data authorization according to the command; During the temporary authorization period, the SIM card records the actual usage and periodically reports it to the operator's cloud platform to support dynamic adjustment or renewal control.

2. The method for dynamic capacity expansion based on OTA traffic according to claim 1, characterized in that, The process of obtaining local package usage information and generating quantified status data includes: Read the user's SIM card's internal data plan file to determine the remaining data allowance; Obtain terminal environment switching information, wherein the terminal environment switching information includes network standard switching records and cell ID change frequency; Capture user data session startup frequency; Record the startup cycle change information of the terminal, wherein the startup cycle change information includes changes in the activation cycle and the sleep cycle; Quantitative status data is generated by fusing the remaining traffic quota, terminal environment switching information, user data session startup frequency, and startup cycle change information according to predefined weights.

3. The method for dynamic capacity expansion based on OTA traffic according to claim 2, characterized in that, Also includes: In high mobility scenarios, the weight of the terminal environment switching information is increased; In fixed hotspot area scenarios, the weights of the data session initiation frequency and the initiation cycle change information are increased; In low-activity scenarios at night, increase the weight of the remaining traffic quota.

4. The method for dynamic capacity expansion based on OTA traffic according to claim 2, characterized in that, The expansion request includes the quantized status data. Based on the capacity expansion request, the operator's cloud platform, combined with historical traffic usage data, user behavior characteristics, geographical location information, and current network load, performs traffic demand prediction and credit policy evaluation to generate a capacity expansion decision, including: The operator cloud platform, based on the quantified status data, historical traffic usage data, user behavior characteristics, geographic location information, and current network load, uses a dynamic attention-guided multi-task graph inference model to predict and generate expansion decisions, including future peak traffic for users, expansion urgency level, and recommended authorization validity period. The dynamic attention-guided multi-task graph inference model includes: Behavior graph construction module: Constructs a behavior event graph based on the user behavior characteristics, and injects the geographic location information and network load as node attributes into the graph structure; The graph attention processing module is used to perform graph neural network processing on the behavior event graph to extract the spatial behavior evolution tensor. The temporal feature extraction module is used to perform multi-channel temporal convolution on the quantized state data and the historical traffic usage data to extract temporal feature tensors; The multi-task decoding module is used to fuse the spatial behavior evolution tensor and the temporal feature tensor and input them into the multi-task decoding structure to generate expansion decisions including future user traffic peaks, expansion urgency levels, and recommended authorization validity periods.

5. The method for dynamic capacity expansion based on OTA traffic according to claim 4, characterized in that, The graph constructed by the behavior graph construction module is a heterogeneous graph structure. The node types of the heterogeneous graph structure include user behavior events, geographical location nodes, time nodes, network load level nodes, and package level nodes. The edges of the heterogeneous graph structure represent behavior transitions and scene causal relationships. The behavior graph construction module includes a dynamic node generation unit, which is used to generate corresponding temporary nodes in real time and inject them into the heterogeneous graph structure when a sudden change in user behavior is detected.

6. The method for dynamic capacity expansion based on OTA traffic according to claim 4, characterized in that, The graph attention processing module adopts a spatiotemporal graph convolutional network structure to simultaneously capture the temporal evolution and spatial dependencies of the behavioral graph. The behavioral offset extracted by the graph attention processing module is also used to drive the adaptive authorization strategy. When a high offset scenario is detected, the expansion parameters are automatically adjusted, including extending the authorization validity period or increasing the expansion amount.

7. The method for dynamic capacity expansion based on OTA traffic according to claim 4, characterized in that, The multi-task decoding module integrates a reinforcement learning mechanism based on the Actor-Critic architecture; The Actor network is used to fuse the spatial behavior evolution tensor and the temporal feature tensor and input them into the multi-task decoding structure to generate expansion decisions including future user traffic peaks, expansion urgency levels, and recommended authorization validity periods. The Critic network is used to receive the current state features and the scaling decision, and to estimate its state value function; The cloud platform constructs a reward signal based on user behavior feedback information after expansion. The reward signal includes whether the expanded data has been exhausted, the frequency of business usage, user subjective interaction behavior, and abnormal usage indicators, so as to simultaneously optimize the parameters of the Actor network and the Critic network.

8. The method for dynamic capacity expansion based on OTA traffic according to claim 7, characterized in that, The Actor-Critic architecture integrates a meta-learning mechanism. In scenarios involving new users or cold-start users, the cloud platform performs graph comparison learning based on the structural features of the user behavior graph and the historical behavior graph set to calculate the similarity score. It selects multiple optimal matching historical users as the source policy candidate set and uses a policy distillation and local parameter freezing mechanism to fuse and transfer the candidate source policies to the current Actor network. During the integration and migration process, a dynamic behavior offset detection module is introduced to assess the migration adaptability based on the rate of change in the new user's policy behavior. When the adaptability is lower than a preset threshold, the policy fine-tuning and freeze control process is automatically triggered.

9. The method for dynamic capacity expansion based on OTA traffic according to claim 4, characterized in that, The multi-task graph inference model further includes a causal inference decision suppression module, the specific steps of which include: A cause-effect graph structure is established based on users' historical expansion behavior, traffic usage after expansion, and experience indicators. The cause-effect graph structure includes expansion behavior processing variables, experience indicator result variables, covariates, and mediator variable nodes. Extract expanded and unexpanded samples and corresponding multi-dimensional experience feedback data from the user group for use in training the causal inference model; Based on the current user status, the expected user experience score is predicted by the causal inference model under the conditions of performing capacity expansion and not performing capacity expansion, and the experience improvement value between the two is calculated. The experience improvement value includes the results of a two-layer causal evaluation system. The results of the two-layer causal evaluation system include the immediate experience improvement result after the user capacity expansion determined based on the immediate experience index, and the long-term value improvement result of the capacity expansion determined based on the user retention rate and package upgrade tendency index. The experience enhancement value is compared with a preset intervention benefit threshold; When the experience improvement value is lower than the preset intervention benefit threshold, the generation of the current expansion command is automatically suppressed, or the expansion quota and authorization validity period are reduced.

10. A dynamic capacity expansion system based on OTA traffic, characterized in that, include: The acquisition unit is used to deploy an applet module in the user's SIM card to monitor data usage status, in order to obtain local data plan usage information and generate quantitative status data. The upload unit is used so that when the quantized state data meets the preset trigger conditions, the Applet module uploads the expansion request to the operator's cloud platform through the terminal; The first generation unit is used by the operator cloud platform to perform traffic demand prediction and credit policy evaluation based on the expansion request, combined with historical traffic usage data, user behavior characteristics, geographical location information and current network load, and generate expansion decisions. The second generation unit is used to generate OTA instructions with temporary expansion parameters based on the expansion decision. The startup unit is used to send the OTA command to the SIM card through the OTA channel, so that the SIM card updates the internal package configuration according to the command and starts temporary data authorization; The monitoring unit is used to record the actual usage of the SIM card during the temporary authorization period and periodically report it to the operator's cloud platform to support dynamic adjustment or renewal control.

Images