Internet of Things flow demand flood peak intelligent response method based on AI

Abstract

The invention discloses an AI-based Internet of Things flow demand flood peak intelligent response method, and relates to the field of Internet of Things flow management, and the method comprises the steps: collecting historical use data and related service information of a plurality of flow pools to form multi-dimensional feature information, and inputting the information into an AI flow prediction model to obtain a flow demand prediction result of each flow pool; the prediction utilization rate is calculated based on the current used flow and the prediction result, when the utilization rate reaches the preset threshold value, the supplement operation is triggered, the cross-pool supplement scheme is generated, the flow pool supplement operation is executed through the supplement instruction, and the problems that supplement is not timely in the flow peak period, and the scheduling efficiency is low are solved.

Description

Technical Field

[0001] The present invention relates to the field of Internet of Things (IoT) traffic management, and more particularly, to an intelligent response method for IoT traffic demand peaks based on AI. Background Art

[0002] With the rapid growth of the number of IoT devices, various sensors, intelligent terminals, and data collection nodes are widely deployed in fields such as transportation, smart cities, and industrial control. The network traffic generated exhibits complex characteristics of multi-services, multi-scenarios, and multi-time series. During major events or when sudden services are launched intensively, the traffic demand often surges rapidly within a short period, posing great challenges to the capacity planning and resource scheduling of the traffic pool. Such peak-period traffic not only has the characteristics of wide geographical distribution and diverse service types, but also because it is difficult for prediction models to accurately capture sudden changes, traditional static reservation or experience-based replenishment strategies are difficult to meet the demand in a timely manner. On the one hand, it may lead to a decline or interruption in service quality due to insufficient capacity, and on the other hand, frequent replenishment will result in resource waste and cost out of control. These problems urgently need to improve the proactive prediction and response efficiency to traffic peaks through multi-dimensional feature analysis and intelligent prediction means. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide an intelligent response method for IoT traffic demand peaks based on AI to solve the problems mentioned in the background art.

[0004] To achieve the above object, the present invention adopts the following technical solutions: An intelligent response method for IoT traffic demand peaks based on AI, comprising: Collecting historical usage data and service information for multiple traffic pools to form multi-dimensional feature information; Inputting the multi-dimensional feature information into an AI traffic prediction model to respectively obtain traffic demand prediction results for the multiple traffic pools; Calculating the predicted utilization rate of each traffic pool respectively based on the current used traffic and traffic demand prediction result of each traffic pool; When the utilization rate of the traffic pool reaches a preset threshold T trigger a replenishment operation for the traffic pool and generate a cross-pool replenishment plan, where the cross-pool replenishment plan includes the allocation of replenishment packages for each traffic pool that needs to replenish traffic; Issuing the allocation of the replenishment packages in the cross-pool replenishment plan as a replenishment instruction to execute the traffic pool replenishment operation.

[0005] In some embodiments, the multi-dimensional feature information includes time features, service features, customer features, scenario features, and traffic pool features, and the traffic pool features include the unit price of replenishment packages, historical replenishment delay rate, and remaining budget.

[0006] In some embodiments, the AI traffic prediction model is a multi-layer long short-term memory network model, which is used to generate a prediction sequence for each traffic pool. It is calculated according to the following formula: ; In the formula, i is the traffic pool number, t is the current time point, k is the prediction step, n is the number of feature dimensions, is the p-th dimensional feature value, is the corresponding weight.

[0007] In some embodiments, the predicted utilization rate of each traffic pool is calculated according to the following formula: ; In the formula, is the used traffic of the i-th traffic pool before time t, m is the number of prediction time steps, is the total capacity of the i-th traffic pool, j is the index of the prediction time step.

[0008] In some embodiments, the method adopts a convex optimization algorithm to generate a cross-pool replenishment plan by solving the following linear programming problem: ; In the formula, is the replenishment amount allocated to the i-th traffic pool, is the unit price of the replenishment package for the i-th traffic pool, is the global replenishment budget, N is the total number of traffic pools.

[0009] In some embodiments, the AI traffic prediction model is trained through the following steps: Using historical traffic data and corresponding multi-dimensional feature information, minimizing the mean square error between the predicted value and the actual value as the training objective, and optimizing the corresponding weights .

[0010] In some embodiments, the preset threshold T is set manually 。

[0011] In some embodiments, the preset threshold T is set by the following method: Using reinforcement learning with the predicted utilization rate, replenishment cost and default penalty of the traffic pool as the state input, and autonomously outputting the preset threshold used for the global optimization scheduling.

[0012] In some embodiments, the design and training process of the reinforcement learning model includes: The state is defined as a state vector containing the predicted utilization rate of each traffic pool, the unit price of supplementary packages, and the default penalty; The action is defined as adjusting the preset threshold T value; The reward function is designed as the negative value of the weighted sum of the supplementary cost and the default penalty; The training process is carried out in a simulation environment, and the threshold strategy is optimized through multiple iterations to minimize the long-term cumulative cost.

[0013] In some embodiments, after the supplementary operation is executed, the actual effect of the supplement is recorded, including the change in the utilization rate after the supplement; the recorded supplementary effect data is used to provide feedback information for subsequent traffic prediction and supplementary strategy optimization.

[0014] The advantages of the present invention over the prior art are as follows: First, the historical usage data and related business information of multiple traffic pools are subjected to multi-dimensional feature extraction, and the data is input into a multi-layer long short-term memory network model with an attention mechanism to generate the traffic demand prediction sequence of each traffic pool; based on the currently used traffic and the prediction results, the predicted utilization rate is calculated. When the utilization rate reaches the preset threshold, the system will automatically trigger the traffic pool supplement and generate a cross-pool supplement plan, which is immediately executed after the supplement instruction is issued, significantly solving the problems of insufficient capacity and lagging supplement during the traffic peak period, effectively improving the real-time performance and accuracy of scheduling, and avoiding service interruption or over-supplementation caused by prediction deviation. Further combined with an auxiliary mechanism, the optimal allocation of cross-pool supplement is completed under the global budget and cost constraints through a convex optimization algorithm; the triggering threshold is dynamically adjusted by reinforcement learning, enabling the system to autonomously optimize the response strategy according to the supplementary cost and the default penalty; after the supplementary operation, the change in the utilization rate is recorded in real time and fed back to the prediction and scheduling model to form a closed-loop optimization of prediction and response, continuously enhancing the adaptive ability and economy of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is the overall flowchart of the present invention; Figure 2 is the detailed flowchart of data collection of the present invention; Figure 3 is the flowchart of model prediction of the present invention; Figure 4 is the flowchart of supplement triggering and plan generation of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] The following describes the specific embodiments of the present invention with reference to the accompanying drawings.

[0017] As shown in Figure 1, the present invention provides an intelligent response method for the peak of Internet of Things (IoT) traffic demand based on artificial intelligence technology, aiming to efficiently respond to the sudden peak of traffic demand in the IoT system through intelligent traffic prediction and resource scheduling, ensuring the reasonable allocation of traffic resources and the stability of system operation. The following is a description of the detailed implementation process of the present invention, covering multiple links such as data collection, model prediction, utilization rate calculation, replenishment trigger, solution generation, and instruction execution, and a detailed description of each step is provided.

[0018] In the IoT system, a traffic pool refers to a shared traffic resource unit allocated to multiple devices or users. Each traffic pool has a certain total capacity, and the devices will gradually consume the traffic in it when using network services. To achieve accurate traffic demand prediction, it is first necessary to comprehensively collect historical usage data and relevant business information of multiple traffic pools. After being processed, these data form multi-dimensional feature information, which serves as the basis for subsequent analysis.

[0019] As shown in Figure 2, specifically, the collected data includes several key aspects. The time feature reflects the periodicity and regularity of traffic usage, such as date, specific time points (hours, minutes), day of the week, and whether it is a holiday. These information can help capture seasonal or periodic changes in traffic demand. The business feature involves the specific attributes of the business, such as business type (video stream, data transmission, etc.), business volume size, and business priority. There may be significant differences in the traffic demand patterns of different businesses. The customer feature focuses on the characteristics of the customer groups using the traffic pool, such as customer type (personal user or enterprise user), customer scale (number of devices), and customer activity (usage frequency). These factors will affect the speed and scale of traffic consumption. The scenario feature is related to the specific environment of traffic usage, including geographical location (city or rural), network environment (4G or 5G), and device type (sensor or smart terminal). These conditions may have a direct impact on traffic demand. In addition, the traffic pool feature is a parameter closely related to the operation and management of the traffic pool, including the unit price of the supplementary package (the unit cost of each supplementary traffic), the historical supplementary delay rate (the proportion of the response time of the supplementary operation), and the remaining budget (the available balance of funds for replenishment). These information provides an important basis for resource allocation.

[0020] Through the systematic collection and preprocessing of the above data, such as normalization, missing value filling, or outlier removal, structured multi-dimensional feature information is finally formed. These feature information are represented in vector form, providing input data for subsequent AI model prediction.

[0021] As shown in Figure 3, in order to predict the future traffic demand of each traffic pool, this method adopts an advanced AI traffic prediction model, specifically a multi-layer long short-term memory network (LSTM) with an attention mechanism. The LSTM model is particularly suitable for processing time series data and can effectively capture the long-term dependencies in traffic usage patterns. The attention mechanism enhances the model's ability to focus on key information by dynamically adjusting the feature weights, thereby improving the prediction accuracy.

[0022] For each traffic pool i , the model will generate a traffic demand prediction sequence for its future k time steps, denoted as . Its calculation process can be expressed by the following formula: ; In this formula, i represents the number of the traffic pool, t is the current time point, k is the prediction step (e.g., predicting the demand for the next 1 hour or 1 day), n is the total number of feature dimensions, is the value of the p-th dimensional feature at time t , and is the weight of the corresponding feature, which is automatically learned by the attention mechanism based on historical data. The core idea of this formula is to perform a weighted sum of multi-dimensional feature information to generate the predicted value of future traffic demand. The introduction of the weight enables the model to dynamically identify which features (such as holidays or business types) have a greater impact on traffic demand, thereby improving the pertinence and accuracy of the prediction.

[0023] In practical applications, the architecture of the model includes multiple LSTM layers. Each layer is responsible for extracting features with different time spans, and finally, the information is integrated through the attention layer to output the prediction sequence. The model includes an input layer, which is responsible for receiving multi-dimensional feature information. These features may include time, business type, customer attributes, scenarios, and data related to the traffic pool. The input layer organizes and transmits this information to the subsequent layers, providing the basic data required for the model to make predictions; The middle multi-layer LSTM layers are composed of multiple stacked LSTM units. LSTM is an improved recurrent neural network (RNN). Through gating mechanisms such as forget gates, input gates, and output gates, it can effectively remember long-term information and process time series data. The multi-layer structure enables the model to learn more complex patterns and features. Usually, the number of layers is between 2 and 4, depending on the data complexity and performance requirements. The attention mechanism layer dynamically adjusts the weights of the features, enabling the model to focus on the features that have the greatest impact on the prediction result. Specifically, the attention mechanism may calculate the weight of each feature through a neural network (as shown in formula As shown in the figure, the pertinence and accuracy of the prediction are enhanced. The model generates the final traffic demand prediction result at the output layer. In addition, to improve the generalization ability, the model may also include a Dropout layer (optional), which prevents overfitting by randomly discarding some neurons during training.

[0024] During the process of training the model, historical traffic data and corresponding feature information are required, and the model parameters are optimized by minimizing the error (such as mean squared error) between the predicted value and the actual value. During the training process, the attention mechanism will gradually adjust the values of, usually the range of these weights is between [0, 1], and the sum is ensured to be 1 through normalization processing.

[0025] Therefore, the prediction process can be summarized as follows: 1. The system first extracts a set of key features (such as historical traffic values, service type codes, time tags, etc.) of each traffic pool at time t.

[0026] 2. The feature sequences of several past moments are sequentially input into the LSTM network. The LSTM automatically learns the change rules of these features in the time dimension and outputs a "hidden state" vector at the last step, representing the temporal context of the current traffic trend.

[0027] 3. Use a simple fully connected layer to map this hidden state into multiple groups of weight vectors , and each group of weights corresponds to the importance distribution of each feature in a future moment (such as t+1, t+2... t+k).

[0028] 4. Multiply each group of weight vectors with the feature vector at the current moment and sum them with weights to obtain the traffic demand prediction value corresponding to the future moment.

[0029] 5. If multiple-step prediction is required, generate all future-step weight vectors at once and calculate them sequentially; it is also possible to use a rolling method, feedback the just-predicted value to the next step, and then repeat the above process.

[0030] In this way, the LSTM is responsible for grasping which features and what temporal changes can most affect future traffic. The fully connected layer then dynamically gives the feature weights at different prediction moments, and finally uses weighted summation to obtain the specific traffic prediction for each step.

[0031] As shown in Figure 4, after obtaining the traffic demand prediction results of each traffic pool, the predicted utilization rate of each traffic pool needs to be calculated by combining the currently used traffic to evaluate its future resource usage. The predicted utilization rate is calculated as follows: ; Among them, represents the traffic consumed by the i th traffic pool as of the current time t; m is the predicted number of time steps, that is, the prediction range considered in the future; is the i th traffic pool's predicted traffic demand at the j th future time step; is the i th traffic pool's total capacity. The formula calculates the percentage of the sum of the currently used traffic and the predicted demands within the future m time steps in the total capacity, reflecting the resource tightness of the traffic pool in the future period.

[0032] The reason for designing the formula like this is that it can comprehensively consider the current traffic usage status and future potential demands, providing a forward-looking utilization rate indicator. For example, if a certain traffic pool has a high currently used traffic and the predicted future demands are expected to continue to increase, then its predicted utilization rate may quickly approach or exceed the total capacity, thus indicating that intervention measures need to be taken. The value of the parameter m needs to be flexibly adjusted according to the business scenario. For example, for a system with high real-time requirements, m can be set to a smaller value (such as the predicted number of steps within 1 hour), while for long-term planning, m can be appropriately extended (such as 1 day or 1 week).

[0033] When the predicted utilization rate of a certain traffic pool reaches or exceeds the preset threshold T , the system will trigger the replenishment operation of this traffic pool. This threshold T is a key parameter used to balance the resource utilization efficiency and replenishment cost. If T is set too low, it may lead to frequent replenishment, increasing the operation cost; if T is too high, there may be a risk of service interruption due to insufficient traffic.

[0034] To intelligently determine the optimal threshold T , this method introduces reinforcement learning technology. The reinforcement learning model takes the predicted utilization rate, replenishment cost, and default penalty of the traffic pool as state inputs, and through interaction and learning with the environment, gradually optimizes the output of the threshold required for global scheduling. Specifically, the model defines the current state as a vector, including the of each traffic pool, the unit price of the replenishment package, and the default penalty that may be caused by insufficient traffic (such as the compensation cost for service interruption). The action of the model is to adjust TFor the value, the goal is to find a threshold strategy with the lowest long-term cumulative cost through trial and error. The reward function can be designed as the negative value of the weighted sum of the replenishment cost and the default penalty, and the weights are determined according to business priorities.

[0035] During the training process, the reinforcement learning model will simulate different traffic demand scenarios and try different T values, and update the strategy according to the effect of the replenishment operation (such as whether the traffic shortage is avoided). Initially, T it can be set to 80% or 90% according to experience. After multiple iterations, the model will converge to a dynamic threshold that adapts to specific business needs, usually fluctuating between 70% and 95%, depending on the capacity and usage pattern of the traffic pool. Of course, in other embodiments, it can also be preset manually.

[0036] Once the replenishment operation is triggered, the system needs to generate a cross-pool replenishment plan to decide how to allocate traffic from the global resources to meet the demand. This method uses a convex optimization algorithm to achieve this goal by solving a linear programming problem. The optimization goal is to minimize the replenishment cost, and the mathematical expression is: ; At the same time, the following constraint conditions need to be met: ; and: ; where, is the replenishment traffic volume allocated to the i th traffic pool, is the unit price of the replenishment package for this traffic pool, is the global replenishment budget, N is the total number of traffic pools. It should be noted that the constraint conditions here assume that the replenishment traffic will increase the total capacity of the traffic pool, thereby reducing the predicted utilization rate to be lower than the threshold T .

[0037] The significance of this linear programming problem is to allocate the replenishment traffic at the lowest cost within the limited budget to ensure that the utilization rate of all traffic pools is controlled within a safe range in the future m time steps. Since the objective function and the constraint conditions are linear, mature optimization tools (such as the simplex method or the interior point method) can be used to solve it efficiently to obtain the optimal replenishment volume .

[0038] After generating the cross-pool replenishment plan, the system will convert the allocation of replenishment packages into specific replenishment instructions and send them to the corresponding traffic pools for execution. These instructions may include directly purchasing new traffic packages and allocating them to the target traffic pool, or in some scenarios, transferring traffic from traffic pools with lower utilization to the traffic pools in need of replenishment. During the execution process, the system will record the actual effects of replenishment, such as the change in utilization rate after replenishment, in order to provide feedback for subsequent prediction and optimization.

[0039] To ensure the effective implementation of the method, some key parameters need to be reasonably set and the AI model needs to be fully trained. Usually, k can be set from 1 hour to 7 days, and m can be a multiple of k. For example, if k is 1 day, then m can be 3 to 7, depending on the business requirements for real-time and forward-looking. The feature dimension n is determined by the collected data, usually including time, business, customer, scenario, and traffic pool features, and the total number may be between dozens and hundreds. The unit price of the replenishment package is determined according to the market price, such as ranging from 10 yuan / GB to 20 yuan / GB, while the global replenishment budget is set by the financial situation of the enterprise, such as 500GB or 1000GB per month.

[0040] The training of the AI traffic prediction model requires a large amount of historical data, and the training goal is to minimize the prediction error. After training, the model can continuously update its parameters through online learning to adapt to the changes in traffic patterns. The training of the reinforcement learning model requires building a simulation environment to simulate the state changes of the traffic pool and the execution of replenishment operations, and optimizing the threshold strategy through multiple iterations.

[0041] To more intuitively illustrate the implementation process of the method, the following is a simplified application example. Assume that the system manages 3 traffic pools, the current time t = 0, and the parameters of each traffic pool are as follows: Traffic Pool 1: Used traffic 100GB, total capacity 200GB, unit price of replenishment package 10 yuan / GB Traffic Pool 2: Used traffic 150GB, total capacity 300GB, unit price of replenishment package 8 yuan / GB Traffic Pool 3: Used traffic 200GB, total capacity 500GB, unit price of replenishment package 12 yuan / GB Assume that the prediction step k = 1, the number of time steps m = 1, and the AI model predicts the traffic demand for the next 1 time step as: Traffic Pool 1: 50GB Traffic Pool 2: 100GB Traffic Pool 3: 150GB According to the prediction utilization formula, the calculation results are as follows: Traffic Pool 1: (100 + 50) / 200 × 100% = 75% Traffic Pool 2: (150 + 100) / 300 × 100% = 83.33% Traffic Pool 3: (200 + 150) / 500 × 100% = 70% Assume the threshold T = 80%. Since the predicted utilization rate of Traffic Pool 2, 83.33%, exceeds the threshold, it needs to be replenished. Let the global budget = 500GB. Solve the linear programming problem with the following constraints: ; Calculated as: ; Therefore, at least 12.5GB needs to be replenished to Traffic Pool 2. Assume 15GB is replenished, and the cost is 15 × 8 = 120 yuan. At this time, the new total capacity of Traffic Pool 2 is 315GB, and the predicted utilization rate is 250 / 315 × 100% ≈ 79.37%, which is lower than 80% and meets the requirements. The utilization rates of other traffic pools are already lower than the threshold and do not need to be replenished.

[0042] In this way, the system can optimize the allocation of traffic resources at a low cost and ensure service stability.

[0043] The present invention realizes intelligent response to IoT traffic demands by combining AI prediction, reinforcement learning, and convex optimization techniques. In the future, according to actual application scenarios, feature selection, model architecture, or replenishment strategies can be further optimized to improve prediction accuracy and resource utilization efficiency. For example, more real-time data can be introduced, or more complex deep learning models can be adopted to handle complex traffic patterns.

[0044] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. An AI-based intelligent response method for the peak of Internet of Things traffic demand, characterized in that, Including: Collecting historical usage data and business information for multiple traffic pools to form multi-dimensional feature information; Inputting the multi-dimensional feature information into an AI traffic prediction model to respectively obtain traffic demand prediction results for the multiple traffic pools; Based on the currently used traffic and the traffic demand prediction results of each traffic pool, calculating the predicted utilization rate of each traffic pool respectively; When the utilization rate of the traffic pool reaches a preset threshold T a replenishment operation of the traffic pool is triggered, and a cross-pool replenishment plan is generated, where the cross-pool replenishment plan includes the allocation of replenishment packages for each traffic pool that needs to replenish traffic; Issuing the supplement package allocation case in the cross-pool supplement plan as a supplement instruction to perform the traffic pool supplement operation.

2. The AI-based intelligent response method for the peak of IoT traffic demand according to claim 1, wherein The multi-dimensional feature information includes time features, business features, customer features, scenario features, and traffic pool features, and the traffic pool features include supplement package unit price, historical supplement delay rate, and remaining budget.

3. The AI-based intelligent response method for the peak flood of Internet of Things traffic requirements according to claim 1, wherein The AI traffic prediction model is a multi-layer long short-term memory network model, which is used to generate a prediction sequence for each traffic pool , which is calculated according to the following formula: ； where \(i\) is the number of the flow pool, \(t\) is the current time point, \(k\) is the prediction step, \(n\) is the number of feature dimensions, is the \(p\)-th dimensional feature value, is the corresponding weight.

4. The AI-based IoT traffic demand peak intelligent response method according to claim 3, wherein Predicted utilization rate of each traffic pool Calculated according to the following formula: ； In the formula, is the used flow rate of the i-th flow pool before time t, m is the number of prediction time steps, is the total capacity of the i-th flow pool, j is the index of the prediction time step.

5. The AI-based intelligent response method for the peak of Internet of Things traffic demand according to claim 4, wherein The method adopts a convex optimization algorithm to generate a cross-pool supplement plan by solving the following linear programming problem: ； wherein, is the supplement amount allocated to the i-th traffic pool, is the unit price of the supplementary package for the i-th traffic pool, is the global supplementary budget, N is the total number of traffic pools.

6. The AI-based intelligent response method for the peak flood of Internet of Things traffic requirements according to claim 3, characterized in that, The AI traffic prediction model is trained through the following steps: Using historical traffic data and corresponding multi-dimensional feature information, minimizing the mean squared error between the predicted value and the actual value is used as the training objective to optimize the corresponding weights .

7. The AI-based intelligent response method for the peak flood of Internet of Things traffic requirements according to claim 1, wherein, Preset threshold T Set artificially 。 8. The AI-based intelligent response method for the peak flood of IoT traffic requirements according to claim 1, characterized in that Preset threshold T It is set by the following method: Using reinforcement learning, with the predicted utilization rate, replenishment cost, and default penalty of the traffic pool as state inputs, the preset threshold is autonomously output T .

9. The AI-based intelligent response method for the peak of Internet of Things traffic demand according to claim 8, wherein The design and training process of the reinforcement learning model includes: The state is defined as a state vector containing the predicted utilization rate of each traffic pool, the supplement package unit price, and the default penalty; The action is defined as adjusting a preset threshold value T value; The reward function is designed as the negative value of the weighted sum of the supplement cost and the default penalty; The training process is carried out in a simulated environment, and the threshold strategy is optimized through multiple iterations to minimize the long-term cumulative cost.

10. The AI-based intelligent response method for the peak of IoT traffic demand according to claim 1, wherein After performing the supplement operation, record the actual effect of the supplement, including the change in utilization rate after the supplement; use the recorded supplement effect data to provide feedback information for subsequent traffic prediction and supplement strategy optimization.

Images