Traffic prediction device and traffic prediction method using SHAP algorithm-based clustering

Abstract

The present specification relates to a traffic prediction device using artificial intelligence. A traffic prediction device using clustering according to one embodiment includes a memory and a processor. The processor may be configured to acquire traffic data separated by time units from IoT devices deployed in multiple separated regions, extract the contribution to traffic prediction of each of the traffic data using an algorithm, cluster the traffic data based on the contribution, and train a Temporal Convolution Network (TCN) using the clustered traffic data.

Description

Technology Field

[0001] An embodiment relates to a traffic prediction device and method using clustering based on the SHAP (Shapley additive exPlanation) algorithm. Background Technology

[0003] With the widespread adoption of IoT devices, these devices have come to generate and transmit massive amounts of data over the Internet. Consequently, traffic prediction for IoT devices has emerged as an important research topic. Most existing traffic prediction technologies are based on static characteristics and suffer from the problem of failing to properly reflect the diverse and dynamic patterns of IoT traffic.

[0004] Furthermore, existing techniques failed to properly identify the features necessary for traffic prediction and performed forecasts by considering even unnecessary features. This leads to the problem of increased computational load. Prior art literature

[0006] Korean Patent Publication No. 10-1944095 (2019.01.24) “Clustering-based AI Task Matching Method” The problem to be solved

[0007] The embodiment aims to overcome the limitations of existing methods by providing a technology that effectively identifies and predicts unique and diverse patterns of IoT traffic.

[0008] The purpose of the embodiment is to accurately identify only the important features in IoT traffic prediction and to provide a prediction model based thereon. means of solving the problem

[0010] A traffic prediction device using clustering according to one embodiment is a device for predicting traffic of a wireless communication device using a deep learning model, comprising a memory and a processor, wherein the processor may be configured to acquire traffic data separated by time units from IoT devices deployed in multiple separated regions, extract the contribution to traffic prediction of each of the traffic data using an algorithm, cluster the traffic data based on the contribution, and train a Temporal Convolution Network (TCN) using the clustered traffic data.

[0011] In addition, the above algorithm may be a SHAP (Shapley additive explanations) algorithm.

[0012] In addition, the above contribution can be obtained by the processor extracting the index with the highest Shapley value from each of the traffic data.

[0013] In addition, the processor may be configured so that the index clusters identical data into the same group.

[0014] A traffic prediction method using clustering according to one embodiment is a method for predicting traffic of a wireless communication device using a deep learning model, comprising the steps of receiving traffic data for the nth date from IoT devices placed in multiple divided regions from a user, predicting traffic data for the n+1th date of each of the multiple divided regions using a learned deep learning model, and presenting traffic data for the n+1th date of each of the multiple divided regions to the user, wherein the deep learning model may be configured to acquire traffic data from IoT devices placed in multiple divided regions, extract hourly features of the traffic data using a SHAP (Shapley additive explanations) algorithm, extract a contribution to traffic prediction for each of the traffic data based on the hourly features, cluster the traffic data based on the contribution, and learn using the clustered traffic data.

[0015] In addition, the deep learning model may be configured to generate the clustered traffic data by extracting the index with the highest Shapley value from each of the traffic data.

[0016] In addition, the deep learning model can be configured to cluster data with the same index into the same group.

[0017] A traffic prediction method using clustering according to another embodiment, in a method for predicting traffic of a wireless communication device using a deep learning model, may include the steps of: obtaining traffic data for the nth date from IoT devices deployed in multiple divided regions; extracting hourly features of the traffic data using a SHAP (Shapley additive explanations) algorithm; extracting a contribution to traffic prediction for each of the traffic data based on the hourly features; clustering the traffic data based on the contribution; training a TCN (Temporal Convolution Network) using the clustered traffic data; and predicting traffic data for the n+1th date of each of the multiple divided regions using the trained TCN.

[0018] Additionally, the clustering step may include the step of the TCN extracting the index with the highest Shapley value from each of the traffic data and the step of clustering data with the same index into the same group.

[0019] A computer-readable recording medium according to another embodiment tangibly embodies a program of instructions that can be executed by a digital processing device to implement a method for predicting traffic of IoT devices placed in multiple divided regions using a digital processing device, and as a recording medium readable by a digital processing device, a program for executing the above-described method on a computer can be recorded thereon. Effects of the invention

[0021] The embodiment utilizes the SHAP algorithm to perform feature-based clustering of data and constructs a lightweight prediction model based on this. This enables effective prediction without unnecessary data or computations.

[0022] The embodiment enables accurate prediction while reducing the complex computations and long training time required for traffic prediction when using existing models by constructing a lightweight prediction model.

[0023] The embodiment can achieve data collection efficiency and cost reduction by collecting data targeting only important features rather than collecting all data.

[0024] The embodiment provides flexibility to respond to various IoT traffic patterns, thereby enabling easy adaptation to changes and diversity in the IoT environment while maintaining performance. Brief explanation of the drawing

[0026] Figure 1 is a conceptual diagram showing the structure of a TCN model. Figure 2 is a flowchart of a traffic prediction method using clustering of an embodiment. FIG. 3 is a block diagram of a traffic prediction device using clustering according to one embodiment. Figure 4 is a graph showing the performance of the traffic prediction method using clustering of the embodiment. Specific details for implementing the invention

[0027] In describing the embodiments of this specification, detailed descriptions of known technologies related to this specification are omitted if it is determined that such descriptions would unnecessarily obscure the essence of this specification. Furthermore, the terms described below are defined in consideration of their functions within this specification, and these definitions may vary depending on the intentions or practices of the user or operator. Therefore, such definitions should be based on the content throughout this specification. Terms used in the detailed description are intended merely to describe the embodiments of this specification and should never be interpreted restrictively. Unless explicitly stated otherwise, expressions in the singular form include the meaning of the plural form. In this description, expressions such as "include" or "comprise" are intended to refer to certain characteristics, numbers, steps, actions, elements, parts thereof, or combinations thereof, and should not be interpreted as excluding the existence or possibility of one or more other characteristics, numbers, steps, actions, elements, parts thereof, or combinations thereof other than those described.

[0028] Terms containing ordinal numbers, such as “first,” “second,” etc., may be used to describe various components, but said components are not limited by said terms. These terms may be used solely in a nominal sense to distinguish one component from another, and their sequential meaning is determined not by such nomenclature but by the context of the description.

[0029] The term “and / or” is used to include all cases of any combination of the multiple items in question. For example, “and / or B” means including all three cases, such as “and B.”

[0030] When it is stated that one component is "connected" or "joined" to another component, it should be understood that while it may be directly connected or joined to that other component, there may also be other components in between.

[0031] Hereinafter, specific embodiments of the examples will be described with reference to the drawings. The following detailed description is provided to facilitate a comprehensive understanding of the methods, devices, and / or objects described herein. However, this is merely illustrative and the examples are not limited thereto.

[0032] Figure 1 is a conceptual diagram showing the structure of a TCN model.

[0033] The Temporal Convolutional Network (TCN) model, which is the artificial intelligence model used in the embodiment, can solve the problem of long-term dependencies that occurs in sequence modeling, such as recurrent neural networks.

[0034] The TCN model addresses the limitation of existing CNN (Convolutional Neural Network) models, which is the computational load that increases dramatically with sequence length, by using dilated convolution. Dilated convolution can overcome the limitations of 1D convolution by using the hyperparameter 'd'. As a result, TCN can effectively learn information over longer time ranges.

[0035] The SHAP algorithm measures the influence of input features on a machine learning model's predictions. Based on this measured influence, the SHAP algorithm can understand the model's predictions and identify the importance of features. By generating the machine learning model's predictions using various combinations of features, the SHAP algorithm can calculate the contribution of each feature to the prediction.

[0036] In the embodiment, predictions for all feature combinations of the model can be generated using the SHAP algorithm.

[0037] In the example, the Shapley value for a feature can be calculated, which is the average of the difference in prediction results when a specific feature is included or excluded along with other features. In other words, this allows measuring how much a feature contributes to the result. Once the Shapley value for all features is calculated, it can be visualized and analyzed, and it is possible to calculate which feature has the greatest influence on the model's prediction.

[0038] The SHAP algorithm can solve the black box problem, a limitation of machine learning models, and further enable a higher level of analysis.

[0039] In the example, the Milan dataset is used. The Milan dataset was provided at the “Data Challenge” held by Telecom Italia in 2014. The Milan dataset was collected from November 1, 2013, to December 31, 2013, and consists of a total of 10,000 (100x100) cells.

[0040] The hyperparameters used in the example are as shown in Table 1 below.

[0041]

[0042] Figure 2 is a flowchart of a traffic prediction method using clustering of an embodiment.

[0043] Referring to FIG. 2, the traffic prediction method using clustering of the embodiment may include the steps of: obtaining traffic data for the nth date from IoT devices placed in multiple divided regions (S210); extracting hourly features of the traffic data using a SHAP (Shapley additive explanations) algorithm (S220); extracting a contribution to traffic prediction for each of the traffic data based on the hourly features (S230); clustering the traffic data based on the contribution (S240); training a TCN (Temporal Convolution Network) using the clustered traffic data (S250); and predicting traffic data for the n+1th date for each of the multiple divided regions using the trained TCN (S260).

[0044] Steps S210, S220, and S230 are steps in which, for example, a processor divides an arbitrary region into 100x100 and uses the SHAP (Shapley Additive exPlanation) algorithm to measure the contribution of each of the 24 input features to the prediction of internet traffic in each region, i.e., one hour of daily traffic data.

[0045] Afterwards, in step S240, if the index of the most important feature among the 24 features is the same, the processor can assign traffic data with the same index to the same cluster.

[0046] Steps S250 and S260 are steps in which, for example, a processor trains a TCN using 24 clustered data points and predicts the n+1th traffic data of each of the multiple separated regions using the trained TCN.

[0047] In the embodiment, traffic data generated from various IoT devices was collected. The traffic data may primarily include features such as hourly traffic volume, request types, and the number of users. The collected raw data can be transformed into 24 input features through a preprocessing process. This preprocessing process may include noise removal, data normalization, and missing value handling.

[0048] For the 24 input features described above, SHAP can be applied to measure the contribution of each feature to traffic prediction. The SHAP algorithm quantifies the impact of each feature's value on the prediction result, thereby distinguishing between important features and those that are not.

[0049] The clustering in the embodiment means that the processor identifies the index of the most important feature based on the result of the SHAP algorithm, and groups data cells having the same important feature index into the same cluster.

[0050] The clustering process of the embodiment provides high computational efficiency and can manage data with similar traffic patterns within the same cluster.

[0051] Subsequently, the TCN model can be trained using each cluster. The TCN model learns traffic fluctuation patterns over time and can predict future traffic based on historical data.

[0052] In the example, a Random Forest model can be used to predict the maximum traffic volume of the n+1th date based on the data of each cell, for example, the nth date of a 100x100 cell.

[0053] The Random Forest model is analyzed using the SHAP algorithm, and the index with the highest Shapley value among the 24 data points of the nth date is extracted. This is applied to 10,000 cells, and cells with the same index can be assigned to a single cluster.

[0054] The 24 data points of the n-th date represent data for each hour. The fact that the index with the highest Shapley value is the same indicates that traffic during common times has a significant impact on the prediction. In other words, it can be concluded that they likely have similar patterns when the index with the highest Shapley value is the same.

[0055] The training of TCN is primarily evaluated through metrics such as MAPE and RMSE, and if necessary, model parameters can be adjusted.

[0056] FIG. 3 is a block diagram of a traffic prediction device using clustering according to one embodiment.

[0057] Figure 3 is a reconstruction of the traffic prediction method using clustering of the embodiment from a hardware perspective.

[0058] The processor (320) may be configured to acquire traffic data separated by time units from IoT devices deployed in multiple separated regions, extract the contribution to traffic prediction for each of the traffic data using the SHAP (Shapley additive explanations) algorithm, cluster the traffic data based on the contribution, and train a TCN (Temporal Convolution Network) using the clustered traffic data.

[0059] A memory (330) according to one embodiment is controlled by a processor (320) and can store an artificial neural network such as a TCN in volatile memory (random access memory, RAM). For example, non-volatile memory (read only memory, ROM) included in the memory (330) can store traffic data in multiple separated regions.

[0060] A memory (330) according to one embodiment can store the processing result of a volatile memory in a non-volatile memory.

[0061] Embodiments according to the present specification may be implemented by various means, e.g., hardware, firmware, software, or a combination thereof. In the case of implementation by hardware, an embodiment of the present specification may be implemented by one or more ASICs (application specific integrated circuits), DSPs (digital signal processors), DSPDs (digital signal processing devices), PLDs (programmable logic devices), FPGAs (field programmable gate arrays), processor controllers, microcontrollers, microprocessors, etc. In the case of implementation by firmware or software, an embodiment of the present specification may be implemented in the form of modules, procedures, functions, etc., that perform the capabilities or operations described above. Software code may be stored in memory (330) and executed by a processor (320). The memory (330) may be located inside or outside the processor (320) and may exchange data with the processor (320) by various means already known.

[0062] Meanwhile, embodiments of the present invention can be implemented as computer-readable code on a computer-readable recording medium. A computer-readable recording medium includes all types of recording devices in which data that can be read by a computer system is stored.

[0063] Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc. Additionally, computer-readable recording media may be distributed across networked computer systems, allowing computer-readable code to be stored and executed in a distributed manner. Furthermore, functional programs, codes, and code segments for implementing the present invention can be easily inferred by programmers in the art to which the present invention belongs.

[0064] The communication module (340) can download traffic data of any region via wireless communication. The communication module (340) may include one or more modules that enable wireless communication between a traffic prediction device using clustering and an external server. For example, the communication module (340) can connect to a 5G communication system. The communication module (340) can perform wireless communication between the traffic prediction device using clustering and an external server or external device through the 5G communication system.

[0065] Figure 4 is a graph showing the performance of the traffic prediction method using clustering of the embodiment.

[0066] In the example, the performance of the TCN model trained for each cluster is evaluated through separate validation data. Performance metrics include MAPE (Mean Absolute Percentage Error) and RMSE (Root Mean Square Error), which can reflect the model's prediction error.

[0067] The formulas for the two indicators are as shown in Mathematical Formula 1 and Mathematical Formula 2 below.

[0068]

[0069]

[0070] Here, is the i-th predicted value, is the i-th ground truth, and n represents the number of data points.

[0071] Based on the performance evaluation, performance can be optimized by modifying the model's structure or parameters if necessary.

[0072] Referring to Fig. 4, the difference between the predicted value and the actual value of the embodiment can be seen, and through this, the performance of the embodiment can be verified. 10,000 cells were verified as test data. The predicted value of the embodiment has a MAPE value of 0.195 and an RMSE value of 95.6. Fig. 4 shows the test results of 1,030 cells.

[0073] The comparison example is a prediction method trained on an LSTM model, an existing traffic prediction method. The comparison example showed a MAPE value of 0.2862 and an RMSE value of 140.7.

[0074] Referring to the exemplary embodiment and comparative example, traffic prediction through the exemplary embodiment can be utilized for resource allocation and management of IoT infrastructure, thereby enabling the smooth processing of traffic generated from a large number of IoT devices.

[0075] The embodiment can significantly reduce necessary infrastructure and data collection costs by improving prediction accuracy and minimizing data collection.

[0076] In addition, the embodiment can improve the user experience by increasing the availability of IoT services and reducing service latency through accurate traffic prediction.

[0077] The foregoing description of this specification is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical concept or essential features of the embodiments. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive.

[0078] The scope of the embodiments is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and the concept of equivalents thereof should be interpreted as being included within the scope of the embodiments.

Claims

Claim 1 A device for predicting traffic of a wireless communication device using a deep learning model, comprising: a memory; and a processor, wherein the processor is configured to acquire traffic data classified by time units from IoT devices deployed in multiple divided regions, extract a contribution to traffic prediction for each of the traffic data using an algorithm, cluster the traffic data based on the contribution, and train a Temporal Convolution Network (TCN) using the clustered traffic data, wherein the algorithm is a Shapley additive explanations (SHAP) algorithm, and the contribution is a traffic prediction device using clustering obtained by the processor extracting the index with the highest Shapley value from each of the traffic data. Claim 2 delete Claim 3 delete Claim 4 In claim 1, the processor is a traffic prediction device using clustering configured such that the index clusters identical data into the same group. Claim 5 A method for predicting traffic of a wireless communication device using a deep learning model comprises: receiving traffic data for the nth date from IoT devices placed in multiple divided regions from a user; predicting traffic data for the n+1th date of each of the multiple divided regions using a learned deep learning model; and presenting traffic data for the n+1th date of each of the multiple divided regions to the user, wherein the deep learning model is configured to acquire traffic data from IoT devices placed in multiple divided regions, extract hourly features of the traffic data using a SHAP (Shapley additive explanations) algorithm, extract a contribution to traffic prediction for each of the traffic data based on the hourly features, cluster the traffic data based on the contribution, and learn using the clustered traffic data, and wherein the deep learning model is configured to generate the clustered traffic data by extracting the index with the highest Shapley value from each of the traffic data. Claim 6 delete Claim 7 In claim 5, the deep learning model is a traffic prediction method using clustering configured to cluster data with the same index into the same group. Claim 8 A method for predicting traffic of a wireless communication device using a deep learning model comprises: a step of obtaining traffic data for the nth date from IoT devices deployed in multiple divided regions; a step of extracting hourly features of the traffic data using a SHAP (Shapley additive explanations) algorithm; a step of extracting a contribution to traffic prediction for each of the traffic data based on the hourly features; a step of clustering the traffic data based on the contribution; a step of training a TCN (Temporal Convolution Network) using the clustered traffic data; and a step of predicting traffic data for the n+1th date for each of the multiple divided regions using the trained TCN, wherein the clustering step includes a step in which the TCN extracts the index with the highest Shapley value from each of the traffic data. Claim 9 In claim 8, the clustering step further comprises the step of clustering data with the same index into the same group, a traffic prediction method using clustering. Claim 10 A computer-readable recording medium having a program for executing the method of claim 8 on a computer, wherein a program of instructions that can be executed by a digital processing device is tangibly implemented to implement a method for predicting traffic of IoT devices deployed in multiple divided regions using a digital processing device, and the recording medium readable by the digital processing device.

Images