Method for determining policy for son and electronic device using same

Abstract

A method for determining a policy for a SON is provided. The method may comprise the steps of: receiving PM data of a communication network; generating feature data representing the PM data by using a deep learning-based encoder; determining an outlier score of the feature data by using a deep learning-based outlier detection model; and selecting one of an artificial intelligence (AI)-based policy and a non-AI-based policy for performing SON in the communication network according to the outlier score.

Description

Method for determining policies for SON and electronic devices using the same

[0001] The present disclosure relates to a base station and an electronic device for managing the base station. More specifically, the present disclosure relates to a method for determining a policy for implementing a Self-Organizing Network (SON).

[0002] The introduction of 5th Generation Mobile Networks (5G) has led to increased congestion in communication networks, significantly increasing operational complexity and operational expenses (OPEX). Consequently, the introduction of advanced SON (Single-Operational Network) capabilities is essential to address these challenges. SON is primarily used for automatic configuration and optimization of access networks, and can be utilized for various purposes, including improved call quality, load balancing, and reduced power consumption.

[0003] For effective network operation, SON functions need to be performed based on artificial intelligence (AI). However, outlier data generated in complex and rapidly changing network environments can lead to inaccurate AI decisions, potentially degrading network performance.

[0004] According to one aspect of the present disclosure, a method for determining a policy for SON may be provided. The method may include receiving PM data from a communication network. The method may include generating feature data representing the PM data using a deep learning-based encoder. The method may include determining an outlier score of the feature data using a deep learning-based outlier detection model. The method may include selecting one of an artificial intelligence-based policy and a non-AI-based policy for performing SON in the communication network based on the outlier score.

[0005] According to one aspect of the present disclosure, an electronic device for determining a policy for SON may be provided. The electronic device may include a memory storing one or more instructions and one or more processors for executing the one or more instructions stored in the memory. The one or more processors may be configured to receive PM data of a communication network by executing the one or more instructions. The one or more processors may be configured to generate feature data representing the PM data using a deep learning-based encoder by executing the one or more instructions. The one or more processors may be configured to determine an outlier score of the feature data using a deep learning-based outlier detection model by executing the one or more instructions. The one or more processors may be configured to select one of an artificial intelligence-based policy and a non-artificial intelligence-based policy for performing SON in the communication network based on the outlier score by executing the one or more instructions.

[0006] According to one aspect of the present disclosure, a computer-readable recording medium having recorded thereon a program for causing an electronic device to execute any one of the aforementioned and later-described methods for determining a policy for a SON may be provided.

[0007] FIG. 1 is a drawing illustrating a base station and an electronic device according to one embodiment.

[0008] FIG. 2 is a diagram illustrating operations of a base station and an electronic device according to one embodiment.

[0009] FIG. 3 is a diagram illustrating the operations of an AI server, an outlier detection module, and a policy switch module according to one embodiment.

[0010] FIG. 4 is a diagram illustrating feature data of a feature space according to one embodiment.

[0011] Figure 5 is a drawing illustrating an encoder according to one embodiment.

[0012] FIG. 6 is a diagram illustrating an operation for generating outlier learning data and normal learning data according to one embodiment.

[0013] Figure 7 is a diagram illustrating learning of an outlier detection model according to one embodiment.

[0014] FIG. 8 is a diagram illustrating an operation for determining an outlier score according to one embodiment.

[0015] FIG. 9 is a flowchart of a method for determining a policy for SON according to one embodiment.

[0016] Figures 10 and 11 are flowcharts of a method for performing SON according to embodiments.

[0017] FIG. 12 is a flowchart of a method for updating an outlier detection model and an AI-based SON model according to one embodiment.

[0018] Figures 13a and 13b are block diagrams of electronic devices according to embodiments.

[0019] Hereinafter, terms used in this specification will be briefly described, and the present disclosure will be described in detail. In this disclosure, the expression “at least one of a, b, or c” can refer to “a,” “b,” “c,” “a and b,” “a and c,” “b and c,” “all of a, b, and c,” or variations thereof.

[0020] The terms used in this disclosure are selected from widely used, common terms, taking into account the functions of the disclosure. However, these terms may vary depending on the intentions of those skilled in the art, precedents, the emergence of new technologies, etc. Furthermore, in certain cases, terms may be arbitrarily selected by the applicant, in which case their meanings will be described in detail in the relevant description. Therefore, the terms used in this disclosure should not be defined simply as names, but rather based on the meanings of the terms and the overall content of the disclosure.

[0021] Singular expressions may include plural expressions unless the context clearly indicates otherwise. Terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art described herein. Furthermore, terms containing ordinal numbers, such as "first" or "second," used herein may be used to describe various components, but such components should not be limited by such terms. Such terms are used solely to distinguish one component from another.

[0022] When a part of the specification is said to "include" a component, unless otherwise specifically stated, this does not exclude other components but rather implies the inclusion of other components. Furthermore, terms such as "part" and "module" used in the specification refer to a unit that processes at least one function or operation, which may be implemented in hardware, software, or a combination of hardware and software.

[0023] In this disclosure, "outlier" may refer to data that falls outside the range of observed data. For example, if the performance data and configuration parameters of a base station's communication network fall outside the observed range, they may be considered outliers. For example, if the status of the communication network, such as call quality, load, or power consumption, changes rapidly, outliers may occur in the performance data and configuration parameters of the communication network. For example, if the number of terminals (User Equipment (UE)) within the communication network rapidly increases, outliers may occur in the performance data and configuration parameters of the communication network.

[0024] In this disclosure, "inlier" may refer to data falling within the observed data range. For example, if the performance data and configuration parameters of a base station's communication network fall within the observed range, they may be considered inlier.

[0025] In the present disclosure, an 'outlier score' may be a numerical indicator indicating the degree or probability that data is an outlier.

[0026] In the present disclosure, the term 'outlier detection technique' may refer to a specific algorithm, method, or approach for outlier detection (or anomality detection). For example, the outlier detection technique may include, but is not limited to, at least one of local outlier factor (LOF), isolation forests, extended isolation forest (EIF), k-nearest neighbors (k-NN), Z-score, Robust Covariance, One-Class Support Vector Machine (SVM), or autoencoder reconstruction.

[0027] In the present disclosure, 'outlier learning data' may mean learning data labeled as an outlier among learning data.

[0028] In the present disclosure, 'normal learning data' may mean learning data labeled as normal among learning data.

[0029] In the present disclosure, 'Performance Management (PM) data' may refer to data indicating the performance of a communication network of a base station. For example, the PM data may include, but is not limited to, at least one of a Channel Quality Indicator (CQI) histogram, a Signal to Noise Ratio (SINR) histogram, a Rank Indicator (RI) histogram, a Modulation Coding Scheme (MCS) histogram, or a Block Error Rate (BLER) histogram.

[0030] In the present disclosure, 'Configuration Management (CM) data' may mean data indicating configuration parameters of a base station regarding a communication network. For example, the CM data may include, but is not limited to, at least one of a base station data retransmission period (evolved NodeB (eNB) data retransmission time), a base station response timeout time (eNB response timeout time), a terminal data retransmission period (UE data retransmission time), a terminal response timeout time (UE response timeout time), or a Transmission Time Interval (TTI) bundling of a Transport Block Set (TBS).

[0031] In the present disclosure, a 'Key Performance Indicator (KPI)' may refer to data that serves as an important indicator for performance evaluation and service management of a communication network. For example, the KPI may include, but is not limited to, at least one of a call drop rate (Radio Resource Control (RRC) Connection Drop Rate), a downlink (DL) Internet Protocol (IP) Throughput, an uplink (UL) Internet Protocol (IP) Throughput, or a VoLTE (Voice over LTE) Quality Defect Rate.

[0032] Below, with reference to the attached drawings, embodiments of the present disclosure are described in detail so that those skilled in the art can easily implement the present disclosure. However, the present disclosure may be implemented in various different forms and is not limited to the embodiments described herein.

[0033] The present disclosure will be described in detail with reference to the attached drawings below.

[0034] FIG. 1 is a drawing illustrating a base station (10) and an electronic device (12) according to one embodiment.

[0035] In one embodiment, the base station (10) may be a wireless station for communication and relay with a terminal (UE) such as a mobile phone. For example, the base station (10) may provide an access network to a user. For example, the base station (10) may be, but is not limited to, an eNB, which is a base station used in a Long-Term Evolution (LTE) network, or a Next Generation NodeB (gNB), which is a base station used in a 5G network.

[0036] The base station (10) may be configured to perform SON to optimize and manage the communication network (11). The base station (10) may be configured to perform various SON functions for self-configuration, self-optimization, and self-healing.

[0037] In one embodiment, the electronic device (12) may be a system for managing and monitoring a base station (10). The electronic device (12) may perform management, monitoring, etc. of a communication network (11) on a management plane. For example, the electronic device (12) may include an EMS (Element Management System) server.

[0038] The base station (10) can perform SON using an AI-based SON model or a non-AI-based SON model.

[0039] The AI-based SON model can be modeled to perform SON by automatically adjusting and optimizing the configuration parameters of the communication network (11) according to the status of the communication network (11). Here, the status of the communication network (11) can be determined from PM data, CM data, and KPI of the communication network (11). The AI-based SON model can be modeled based on machine learning or deep learning.

[0040] In contrast, non-AI-based SON models can be modeled to perform SON without using AI, such as machine learning or deep learning. Non-AI-based SON models can be designed to reliably manage communication networks even in situations where outliers occur. For example, non-AI-based SON models can be modeled to perform SON using conventional methods, rule-based methods, or manual user intervention.

[0041] When the base station (10) performs SON using an AI-based SON model, whether the PM data, CM data, and KPI of the communication network (11) show outliers may affect the performance and reliability of the SON. For example, when the PM data, CM data, and KPI of the communication network (11) do not fall within the range of the learning data of the AI-based SON model, the performance and reliability of the SON may deteriorate. Accordingly, when the PM data, CM data, and KPI of the communication network (11) show outliers, it may be desirable for the base station (10) to perform SON using a non-AI-based SON model.

[0042] Below, examples of a method for determining a policy for performing SON by considering outliers in a communication network (11) are described.

[0043] FIG. 2 is a drawing explaining the operations of a base station (100) and an electronic device (200) according to one embodiment.

[0044] In one embodiment, the base station (100) may include an Operations Administration Maintenance (OAM) module (110), a SON agent module (120), and a wireless communication network device (130). The electronic device (200) may include a management module (210), an AI server (220), and a SON management module (230).

[0045] The modules (110, 120) illustrated in FIG. 2 may be configurations that are actually implemented by at least one processor included in the base station (100) executing a program or command stored in a memory included in the base station (100). In addition, the modules (210, 230, 231, 232) illustrated in FIG. 2 may be configurations that are actually implemented by at least one processor included in the electronic device (200) executing a program or command stored in a memory included in the electronic device (200). Therefore, the operations described below as being performed by the modules (110, 120) of the base station (100) may actually be performed by at least one processor included in the base station (100). In addition, the operations described below as being performed by the modules (210, 230, 231, 232) of the electronic device (200) may actually be performed by at least one processor included in the electronic device (200).

[0046] In operation S21, the OAM module (110) can receive PM data, CM data, and KPI of the communication network of the base station (100) from the wireless communication network device (130).

[0047] In one embodiment, a wireless communication network device (130) may include a radio unit (RU) (131), a scheduler (132), and a modem (133). The RU (131) may be a device that processes a wireless signal. The scheduler (132) may be a device that schedules an uplink and a downlink of a communication network. The modem (133) may be a device that performs modulation and demodulation between digital and analog signals.

[0048] In one embodiment, the wireless communication network device (130) can collect terminal information such as signal strength, transmission speed, and connection status of each terminal. The wireless communication network device (130) can generate PM data, CM data, and KPI of the communication network by statistically analyzing and analyzing the collected terminal information.

[0049] In operation S22, the OAM module (110) can transmit PM data, CM data, and KPI of the communication network of the base station (100) to the management module (210).

[0050] In one embodiment, the OAM module (110) may transmit PM data, CM data, and KPI of the communication network to the management module (210) at predetermined intervals. Alternatively, the OAM module (110) may transmit PM data, CM data, and KPI of the communication network to the management module (210) in response to a request from the management module (210).

[0051] In operation S23, the management module (210) can transmit PM data, CM data, and KPI received from the OAM module (110) to the AI ​​server (220).

[0052] In one embodiment, the AI-based SON model (222) may be an AI-based model for the base station (100) to perform SON. The AI-based SON model (222) may be a model based on machine learning or deep learning. The outlier detection model (223) may be a deep learning-based model for detecting outliers in a communication network.

[0053] In one embodiment, the AI ​​server (220) can collect PM data, CM data, and KPI of a communication network. The AI ​​server (220) can store the collected PM data, CM data, and KPI in a database (221).

[0054] In one embodiment, the AI ​​server (220) can train an AI-based SON model (222). The AI ​​server (220) can train the AI-based SON model (222) using PM data, CM data, and KPI collected in the database (221) as training data.

[0055] In one embodiment, the AI ​​server (220) can train an outlier detection model (223). The AI ​​server (220) can train the outlier detection model (223) using PM data, CM data, and KPI collected in the database (221) as training data.

[0056] The AI ​​server (220) can use the same training data for training the AI-based SON model (222) and the outlier detection model (223). Accordingly, consistent outlier detection and management can be achieved between the AI-based SON model (222) and the outlier detection model (223).

[0057] The AI ​​server (220) may decide to train the AI-based SON model (222) and the outlier detection model (223) when initial generation or update of the AI-based SON model (222) and the outlier detection model (223) is required. For example, the update of the AI-based SON model (222) and the outlier detection model (223) may include at least one of regular updates, updates at the request of the base station (100), updates at the request of an administrator, or updates at the request of PM data, CM data, and KPIs.

[0058] The AI ​​server (220) may include resources for training an AI-based SON model (222) and an outlier detection model (223). For example, the AI ​​server (220) may include, but is not limited to, a Graphics Processing Unit (GPU), a Neural Processing Unit (NPU), a Tensor Processing Unit (TPU), a Visual Processing Unit (VPU), or a hardware accelerator.

[0059] In some embodiments, the AI ​​server (220) can receive PM data, CM data, and KPI directly from the OAM module (110).

[0060] In some embodiments, the AI ​​server (220) may receive PM data, CM data, and KPI generated by imitating the communication network of the base station (100) from the simulator. In this case, the AI ​​server (220) may use the imitated PM data, CM data, and KPI as training data to train the AI-based SON model (222) and the outlier detection model (223).

[0061] In some embodiments, the AI ​​server (220) may be an external device of the electronic device (200). In this case, the AI ​​server (220) may be connected to the electronic device (200) via a wired network such as Ethernet or Infiniband, a wireless network such as Wireless Fidelity (Wi-Fi), or a Universal Serial Bus (USB).

[0062] In operation S24, the AI ​​server (220) can transmit the PM data and the outlier detection model (223) received from the management module (210) to the SON management module (230).

[0063] In one embodiment, the SON management module (230) may include a policy switch module (231) and an outlier detection module (232).

[0064] The outlier detection module (232) can determine whether PM data is an outlier using the outlier detection model (223). When the AI ​​server (220) updates the outlier detection model (223), the outlier detection module (232) can determine whether PM data is an outlier using the updated outlier detection model (223).

[0065] If the outlier detection module (232) determines that the PM data is an outlier, the policy switch module (231) may select a non-AI-based policy as the policy for performing SON. Alternatively, if the outlier detection module (232) determines that the PM data is normal, the policy switch module (231) may select an AI-based policy as the policy for performing SON.

[0066] The SON management module (230) can receive PM data and determine whether it is an outlier when a policy decision is required to perform SON. For example, when a regular policy decision, a policy decision at the request of the base station (100), a policy decision at the request of an administrator, or a policy decision based on changes in PM data, CM data, and KPIs is required, the SON management module (230) can receive PM data and determine whether it is an outlier.

[0067] In some embodiments, the SON management module (230) may receive PM data directly from the management module (210). Alternatively, the SON management module (230) may receive PM data directly from the OAM module (110).

[0068] In operation S25, the SON management module (230) can transmit the selected policy to the management module (210).

[0069] In operation S26, the management module (210) can transmit the selected policy to the OAM module (110).

[0070] If the policy switch module (231) selects a non-AI-based policy as the policy for performing SON, the management module (210) can transmit the non-AI-based policy to the OAM module (110). Alternatively, if the policy switch module (231) selects an AI-based policy as the policy for performing SON, the management module (210) can transmit the AI-based policy to the OAM module (110).

[0071] The management module (210) can transmit an AI-based SON model (222) to the OAM module (110). When the policy switch module (231) selects an AI-based policy as a policy for performing SON, when the AI-based SON model (222) is initially created, when the AI-based SON model (222) is updated, when the base station (100) requests the AI-based SON model (222), or when there is an administrator request, the management module (210) can transmit the AI-based SON model (222) to the OAM module (110).

[0072] In some embodiments, the management module (210) may receive an AI-based SON model (222) from the AI ​​server (220) and transmit it to the OAM module (110). Alternatively, the management module (210) may receive an AI-based SON model (222) from the AI ​​server (220) via the SON management module (230) and transmit it to the OAM module (110).

[0073] In some embodiments, the OAM module (110) can receive an AI-based SON model (222) directly from an AI server (220).

[0074] In operation S27, the OAM module (110) can transmit the policy and AI-based SON model (222) received from the management module (210) to the SON agent module (120).

[0075] The SON agent module (120) may perform SON using a non-AI-based SON model when receiving a non-AI-based policy. Alternatively, the SON agent module (120) may perform SON using an AI-based SON model (222) when receiving an AI-based policy.

[0076] In operation S28, the SON agent module (120) can transmit the setting parameters determined from the SON to the wireless communication network device (130).

[0077] By having the wireless communication network device (130) operate according to the set parameters, the communication network of the base station (100) can be optimized and managed.

[0078] In one embodiment, operations (S21 to S28) may be performed on an application for SON. The management module (210) may determine an application for SON on the management plane. For example, the application may be, but is not limited to, energy saving (ES), load balancing (LB), or a scheduler.

[0079] FIG. 3 is a diagram illustrating the operations of an AI server (310), an outlier detection module (320), and a policy switch module (330) according to one embodiment.

[0080] The AI ​​server (310) can store PM data, CM data, and KPI of the communication network of the base station in a database (DB). The AI ​​server (310) can use the PM data, CM data, and KPI of the communication network collected in the database (DB) as learning data to create or update an outlier detection model and an AI-based SON model.

[0081] In operation S31, the AI ​​server (310) can train the encoder.

[0082] The AI ​​server (310) can create or update an encoder by training the encoder using PM data, CM data, and KPI of the communication network collected in the database (DB) as learning data.

[0083] An encoder can be trained to generate feature data representing PM data by compressing it. PM data in a communication network can have high dimensions, and the encoder can facilitate data processing by reducing the dimensionality of the PM data. The encoder can be trained to position the PM data features corresponding to the CM data and KPIs closer together in the feature space as the similarity between the CM data and the KPIs increases.

[0084] The encoder may be a deep learning-based model. For example, the encoder may be, but is not limited to, a convolutional neural network (CNN)-based encoder, a recurrent neural network (RNN)-based encoder, a multilayer perceptron (MLP)-based encoder, or an autoencoder.

[0085] In operation S32, the AI ​​server (310) can generate outlier learning data and normal learning data.

[0086] The AI ​​server (310) can generate outlier learning data and normal learning data based on the distribution of feature data in the feature space.

[0087] In one embodiment, the AI ​​server (310) can cluster feature data and generate outlier learning data and normal value learning data based on the clusters of the feature data.

[0088] In operation S33, the AI ​​server (310) can calculate an outlier score for each of the outlier learning data and the normal learning data.

[0089] The AI ​​server (310) can calculate an outlier score for each of the outlier learning data and the normal learning data by using two or more outlier detection techniques.

[0090] In operation S34, the AI ​​server (310) can train an outlier detection model.

[0091] The AI ​​server (310) can create or update an outlier detection model by training the outlier detection model using outlier learning data and normal learning data.

[0092] The AI ​​server (310) can train an outlier detection model to infer outlier scores of each of the outlier learning data and the normal learning data calculated in operation S33, and output a final outlier score indicating whether each of the outlier learning data and the normal learning data is an outlier from the inferred outlier scores.

[0093] The AI ​​server (310) may generate an outlier threshold value that serves as a comparison criterion with the final outlier score for outlier detection. The AI ​​server (310) may generate an outlier threshold value such that the final outlier score output by the outlier detection model from the outlier learning data is greater than or equal to the outlier threshold value, and the final outlier score output by the outlier detection model from the normal learning data is less than the outlier threshold value.

[0094] In operation S35, the AI ​​server (310) can train an AI-based SON model.

[0095] The AI ​​server (310) can create or update an AI-based SON model by training the AI-based SON model using normal learning data.

[0096] The same normalized training data used in the outlier detection model can be used for training the AI-based SON model. This allows for consistent outlier detection and management between the AI-based SON model and the outlier detection model.

[0097] The AI ​​server (310) can store the encoder, the outlier detection model, the AI-based SON model, and the outlier threshold value in a database (DB). Furthermore, the AI ​​server (310) can transmit the encoder, the outlier detection model, and the outlier threshold value to the outlier detection module (320). Furthermore, the AI ​​server (310) can transmit the AI-based SON model to the base station.

[0098] The outlier detection module (320) can receive PM data of the communication network of the base station. The outlier detection module (320) can determine whether the received PM data of the communication network is an outlier.

[0099] In operation S36, the outlier detection module (320) can generate feature data.

[0100] The outlier detection module (320) can generate feature data representing PM data using an encoder received from the AI ​​server (310).

[0101] In operation S31, the encoder may be trained such that the feature data of PM data corresponding to the CM data and the KPI are positioned closer to each other in the feature space as the similarity between the CM data and the KPI are higher. Accordingly, in operation S36 according to an embodiment, feature data may be generated such that the feature data of the PM data are positioned closer to each other in the feature space as the similarity between the CM data and the KPI are higher. In operation S36 according to an embodiment, feature data may be generated such that the feature data of the PM data are positioned closer to each other in the feature space as the similarity between the KPI corresponding to the PM data is higher. In operation S36 according to an embodiment, feature data may be generated such that the feature data of the PM data are positioned closer to each other in the feature space as the similarity between the CM data and the KPI corresponding to the PM data are higher.

[0102] In operation S37, the outlier detection module (320) can determine an outlier score.

[0103] The outlier detection module (320) can determine an outlier score of feature data using an outlier detection model received from the AI ​​server (310).

[0104] In operation S34, the outlier detection model can be modeled such that the location of the feature data in the feature space affects the outlier score. The location of the feature data reflects the similarity between the CM data and the KPI corresponding to the PM data. Accordingly, in operation S37, the outlier detection module (320) can determine the outlier score of the feature data by reflecting the relationship among the PM data, CM data, and KPI through the outlier detection model.

[0105] In operation S38, the outlier detection module (320) can determine whether the feature data is an outlier.

[0106] If the outlier score determined in operation S37 is less than the outlier threshold, the outlier detection module (320) may determine that the feature data is normal. Alternatively, if the outlier score determined in operation S37 is greater than or equal to the outlier threshold, the outlier detection module (320) may determine that the feature data is an outlier.

[0107] If the outlier detection module (320) determines in operation S38 that the feature data is normal, the policy switch module (330) may select an AI-based policy as the policy for performing SON in operation S39. Alternatively, if the outlier detection module (320) determines in operation S38 that the feature data is an outlier, the policy switch module (330) may select a non-AI-based policy as the policy for performing SON in operation S40.

[0108] FIG. 4 is a diagram illustrating feature data (400, 410) of a feature space according to one embodiment.

[0109] In optimizing a base station's communications network, PM data, CM data, and KPIs are interrelated parameters. CM data can be adjusted based on PM data and KPIs, and PM data and KPIs can change based on the adjusted CM data. For example, PM data such as the CQI histogram and SINR histogram can change as the base station data retransmission cycle is adjusted. As PM data changes, KPIs such as the call drop rate, downlink transmission rate, and uplink transmission rate can change accordingly.

[0110] In one embodiment, the PM data's feature data may be located closer to each other in the feature space as the similarity between the corresponding CM data and the corresponding KPI increases. Furthermore, whether the PM data is an outlier can be determined based on the position of the feature data in the feature space. Accordingly, whether the PM data is an outlier can be determined by reflecting the relationship between the corresponding CM data and the KPI.

[0111] FIG. 4, according to one embodiment, illustrates exemplary feature data (400, 410) of an arbitrary two-dimensional feature space. The first feature data (400) represents first PM data (401), and the second feature data (410) represents second PM data (411).

[0112] In one embodiment, the first PM data (401) and the first KPI (403) may represent the performance of a communication network set with the first CM data (402). Accordingly, the first PM data (401) may correspond to the first CM data (402) and the first KPI (403).

[0113] In one embodiment, the second PM data (411) and the second KPI (413) may represent the performance of a communication network set by the second CM data (412). Accordingly, the second PM data (411) may correspond to the second CM data (412) and the second KPI (413).

[0114] In one embodiment, the similarity within the first CM data (402) may be higher than the similarity between the first CM data (402) and the second CM data (412). In one embodiment, the similarity within the second CM data (412) may be higher than the similarity between the first CM data (402) and the second CM data (412). For example, the first CM data (402) may be the same type of CM data, the second CM data (412) may be the same type of CM data, and the first CM data (402) and the second CM data (412) may be different types of CM data. For example, the first CM data (402) may have similar values, and the second CM data (412) may have similar values, while there may be a relatively large difference between the values ​​of the first CM data (402) and the second CM data (412).

[0115] In one embodiment, the similarity within the first KPI (403) may be higher than the similarity between the first KPI (403) and the second KPI (413). In one embodiment, the similarity within the second KPI (413) may be higher than the similarity between the first KPI (403) and the second KPI (413). For example, the first KPI (403) may be a KPI of the same type, the second KPI (413) may be a KPI of the same type, and the first KPI (403) and the second KPI (413) may be different types of KPIs. For example, the first KPI (403) may have similar values, and the second KPI (413) may have similar values, while there may be a relatively large difference between the values ​​of the first KPI (403) and the second KPI (413).

[0116] In one embodiment, as the similarity within the first CM data (402) and the similarity within the second CM data (412) are higher than the similarity between the first CM data (402) and the second CM data (412), the first feature data (400) may be located densely in the feature space, the second feature data (410) may be located densely in the feature space, and the first feature data (400) and the second feature data (410) may be located apart from each other in the feature space.

[0117] In one embodiment, as the similarity within the first KPI (403) and the similarity within the second KPI (413) are higher than the similarity between the first KPI (403) and the second KPI (413), the first feature data (400) may be densely located in the feature space, the second feature data (410) may be densely located in the feature space, and the first feature data (400) and the second feature data (410) may be spaced apart from each other in the feature space.

[0118] FIG. 5 is a drawing illustrating an encoder (500) according to one embodiment.

[0119] In one embodiment, the encoder (500) may include a Variational Autoencoder (VAE, 510) and a Contrastive Learning Model (Contrastive Learning Model, 520).

[0120] The VAE (510) may be configured to compress PM data. The contrastive learning model (520) may be a contrastive learning-based model that uses CM data and KPIs corresponding to the PM data as labels. The contrastive learning model (520) may generate feature data such that the higher the label similarity, the closer the feature data are to each other in the feature space.

[0121] In some embodiments, the VAE (510) may be replaced with another module capable of compressing PM data. For example, the VAE (510) may be replaced with a CNN-based encoder, an RNN-based encoder, an MLP-based encoder, an autoencoder, a Principal Component Analysis (PCA)-based encoder, or a Linear Discriminant Analysis (LDA)-based encoder, but is not limited to the examples listed.

[0122] In some embodiments, the VAE (510) may be omitted, i.e., the encoder (500) may only include a contrastive learning model (520).

[0123] FIG. 6 is a diagram illustrating an operation for generating outlier learning data and normal learning data according to one embodiment.

[0124] The AI ​​server can generate outlier training data and normal training data from PM data, CM data, and KPIs collected in the database. More specifically, the AI ​​server can train an encoder using the collected PM data, CM data, and KPIs, generate feature data for the collected PM data using the encoder, and generate outlier training data and normal training data from the feature data.

[0125] In operation S61, the AI ​​server can cluster feature data.

[0126] The AI ​​server can cluster feature data based on the distribution of feature data in the feature space. For example, the AI ​​server can group feature data into clusters (601, 602, 603, 604) based on the distribution of feature data in the feature space. Various clustering techniques can be used for clustering. For example, K-Means clustering, Mean-Shift clustering, hierarchical clustering, or Gaussian Mixture Model (GMM) clustering can be used, but are not limited thereto.

[0127] In operation S62, the AI ​​server can generate outlier learning data and normal learning data.

[0128] The AI ​​server can generate outlier learning data and normal learning data based on the distribution of feature data in the feature space. For example, the AI ​​server can generate normal learning data in the feature space within a predetermined distance from the center point of each of the clusters (601, 602, 603, 604), and generate outlier learning data in the feature space outside the predetermined distance. For example, the AI ​​server can generate normal learning data in the feature space within a 3-sigma range of each of the clusters (601, 602, 603, 604), and generate outlier learning data in the feature space outside the 3-sigma range.

[0129] The graph on the right side of Figure 6 illustrates exemplary examples of outlier training data and normal training data generated by the AI ​​server. In the graph on the right side of Figure 6, dotted circles represent outlier training data, and solid circles represent normal training data.

[0130] The AI ​​server can label the generated normal learning data as normal and label the generated outlier learning data as outlier.

[0131] In some embodiments, the AI ​​server can generate outlier training data and normal training data using a generative model. Based on the generative model, the AI ​​server learns the distribution of feature data in the feature space, thereby generating outlier training data and normal training data.

[0132] FIG. 7 is a diagram illustrating learning of an outlier detection model (700) according to one embodiment.

[0133] In one embodiment, the outlier detection model (700) may include two or more outlier score estimation models (710, 720, 730) and a final outlier score estimation model (740). In one embodiment, FIG. 7 illustrates an outlier detection model (700) including three outlier score estimation models (710, 720, 730).

[0134] In one embodiment, the outlier training data may be labeled as outliers, and the normal training data may be labeled as normal. Accordingly, the AI ​​server can determine whether the training data is outlier training data or normal training data based on the labels.

[0135] The AI ​​server can train an outlier detection model (700) to generate a final outlier score from each of the outlier training data and the normal training data.

[0136] The AI ​​server may use different outlier detection techniques to calculate outlier scores for each of the outlier training data and the normal training data. In one embodiment, the AI ​​server may use a first outlier detection technique to calculate first outlier scores, a second outlier detection technique to calculate second outlier scores, and a third outlier detection technique to calculate third outlier scores for each of the outlier training data and the normal training data.

[0137] The AI ​​server can train outlier score estimation models (710, 720, 730) to infer the produced outlier scores. In one embodiment, the AI ​​server can train each of the outlier score estimation models (710, 720, 730) such that the first outlier score estimation model (710) infers first outlier scores from outlier training data and normal training data, the second outlier score estimation model (720) infers second outlier scores from outlier training data and normal training data, and the third outlier score estimation model (730) infers third outlier scores from outlier training data and normal training data.

[0138] The AI ​​server can determine an outlier threshold. The outlier threshold can be a value that serves as a comparison criterion for the final outlier score in order to identify outliers. For example, if the final outlier score estimation model (740) is trained to output a probability ranging from 0 to 1 as the final outlier score, the AI ​​server can set the outlier threshold to 0.5.

[0139] The AI ​​server can train the final score estimation model (740) to output a final outlier score. The AI ​​server can train the final outlier score estimation model (740) to generate a final outlier score from the outlier scores inferred by the outlier score estimation models (710, 720, 730).

[0140] The AI ​​server can train the final score estimation model (740) to output a final outlier score that is less than the outlier threshold for outlier learning data and to output a final outlier score that is greater than or equal to the outlier threshold for normal learning data.

[0141] In one embodiment, the models (700, 710, 720, 730, 740) of FIG. 7 may be deep learning-based models. For example, the models (700, 710, 720, 730, 740) may be CNN-based models, RNN-based models, Dense Neural Network (DNN)-based models, or Long Short-Term Memory (LSTM)-based models, but are not limited to the listed examples.

[0142] Since the outlier detection model (700) is configured to determine a final outlier score by reflecting outlier scores produced by two or more outlier detection techniques, various types of outliers can be detected, and the robustness of outlier detection can be improved.

[0143] FIG. 8 is a diagram illustrating an operation for determining an outlier score according to one embodiment.

[0144] The outlier detection module can receive PM data from the base station's communication network and determine an outlier score for the received PM data. To determine the outlier score, the outlier detection module can receive an encoder (810), an outlier detection model (820), and an outlier threshold value from the AI ​​server.

[0145] The outlier detection module can generate feature data from PM data using an encoder (810). The feature data can be positioned closer to each other in the feature space as the similarity between the CM data corresponding to the PM data or the similarity between the KPI corresponding to the PM data increases.

[0146] The outlier detection module can generate outlier scores from feature data using an outlier detection model (820). If an outlier score lower than the outlier threshold is generated, the PM data can be determined as normal. If an outlier score greater than or equal to the outlier threshold is generated, the PM data can be determined as an outlier.

[0147] FIG. 9 is a flowchart of a method for determining a policy for SON according to one embodiment.

[0148] In step S901, the electronic device can receive PM data from a communication network.

[0149] An electronic device may receive PM data of a communication network from a base station. In one embodiment, the PM data may include data indicating the performance of the communication network of the base station.

[0150] In step S902, the electronic device can generate feature data representing PM data using a deep learning-based encoder.

[0151] In one embodiment, the electronic device may generate feature data such that the feature data of the PM data are positioned closer to each other in the feature space as the similarity between the CM data corresponding to the PM data increases. In other words, the closer the distance between the feature data in the feature space, the higher the similarity between the CM data corresponding to the PM data.

[0152] In one embodiment, the electronic device may generate feature data such that the closer the feature data of the PM data are to each other in the feature space, the higher the similarity between the KPIs corresponding to the PM data. In other words, the closer the distance between the feature data in the feature space, the higher the similarity between the KPIs corresponding to the PM data.

[0153] In one embodiment, the electronic device may generate feature data such that the feature data of the PM data are positioned closer to each other in the feature space as the similarity between the CM data and the KPI corresponding to the PM data increases. In other words, the closer the distance between the feature data in the feature space, the higher the similarity between the CM data and the KPI corresponding to the PM data.

[0154] In step S903, the electronic device can determine an outlier score of the feature data using a deep learning-based outlier detection model.

[0155] In one embodiment, the outlier detection model can be modeled such that the location of feature data in the feature space affects the outlier score. In one embodiment, the location of the feature data can reflect the similarity between the PM data, the corresponding CM data, and the KPI. Accordingly, the electronic device can determine the outlier score of the feature data by reflecting the relationship among the PM data, the CM data, and the KPI through the outlier detection model.

[0156] In step S904, the electronic device may select one of an AI-based policy and a non-AI-based policy for performing SON in the communication network based on the outlier score.

[0157] In one embodiment, the electronic device may select an AI-based policy as the policy for performing SON in the communication network if the outlier score is less than a predetermined outlier threshold. In another embodiment, the electronic device may select a non-AI-based policy as the policy for performing SON in the communication network if the outlier score is greater than or equal to the predetermined outlier threshold.

[0158] In one embodiment, an outlier score below a predetermined outlier threshold may indicate that the PM data is normal. In this case, the base station can optimize the communication network by performing SON using an AI-based SON model.

[0159] In one embodiment, an outlier score greater than or equal to a predetermined outlier threshold may indicate that the PM data is an outlier. In this case, the base station can perform SON using a non-AI-based SON model, thereby reliably managing the network even in situations where outliers occur.

[0160] Figure 10 is a flowchart of a method for performing SON according to one embodiment.

[0161] In step S1001, the base station can transmit PM data to the electronic device.

[0162] PM data can be determined by the performance of the communication network of a base station set as CM data. The base station can receive PM data from a wireless communication network device and transmit it to an electronic device.

[0163] In one embodiment, the base station may transmit PM data of the communication network to the electronic device periodically, at the request of the electronic device, at the request of an administrator, or when a significant change is detected in the PM data, CM data, and KPI of the communication network.

[0164] In step S1002, the base station can receive one of an AI-based policy and a non-AI-based policy from the electronic device.

[0165] The electronic device can determine either an AI-based policy or a non-AI-based policy based on PM data transmitted by the base station. The base station can receive the determined policy from the electronic device.

[0166] In step S1003, the base station can perform SON according to the received policy.

[0167] If an AI-based policy is received, the base station can perform SON using an AI-based SON model. Alternatively, if a non-AI-based policy is received, the base station can perform SON using a non-AI-based SON model.

[0168] Receiving an AI-based policy may indicate that the PM data transmitted by the base station to the electronic device is normal. In this case, the base station can optimize the network by performing SON using an AI-based SON model. Conversely, receiving a non-AI-based policy may indicate that the PM data transmitted by the base station to the electronic device is an outlier. In this case, the base station can perform SON using a non-AI-based SON model, allowing it to reliably manage the network even in situations where outliers occur.

[0169] Figure 11 is a flowchart of a method for performing SON according to one embodiment.

[0170] In step S1101, the base station can transmit PM data to the electronic device.

[0171] Step S1101 may be applied to the descriptions of step S901 of FIG. 9 and step S1001 of FIG. 10.

[0172] In step S1102, the electronic device can generate feature data representing PM data using a deep learning-based encoder.

[0173] In one embodiment, the encoder may be an encoder generated or updated by the electronic device. In another embodiment, the encoder may be an encoder generated or updated by an AI server external to the electronic device.

[0174] The description of step S902 of FIG. 9 may be applied to step S1102.

[0175] In step S1103, the electronic device can determine an outlier score of the feature data using a deep learning-based outlier detection model.

[0176] In one embodiment, the outlier detection model may be a model generated or updated by the electronic device. In another embodiment, the outlier detection model may be a model generated or updated by an AI server external to the electronic device.

[0177] The description of step S903 of FIG. 9 may be applied to step S1103.

[0178] In step S1104, the electronic device may select one of an artificial intelligence-based policy and a non-artificial intelligence-based policy for performing SON in the communication network based on the outlier score.

[0179] In one embodiment, the electronic device may select an AI-based policy as the policy for performing SON in the communication network if the outlier score is less than a predetermined outlier threshold. In another embodiment, the electronic device may select a non-AI-based policy as the policy for performing SON in the communication network if the outlier score is greater than or equal to the predetermined outlier threshold.

[0180] The description of step S904 of FIG. 9 may be applied to step S1104.

[0181] In step S1105, the base station can perform SON according to the selected policy.

[0182] In one embodiment, the base station may receive a selected policy from an electronic device. If an AI-based policy is received, the base station may perform SON using an AI-based SON model. Alternatively, if a non-AI-based policy is received, the base station may perform SON using a non-AI-based SON model.

[0183] In one embodiment, the AI-based SON model may be a model generated or updated by the electronic device. In another embodiment, the AI-based SON model may be a model generated or updated by an AI server external to the electronic device.

[0184] The description of step S1003 of FIG. 10 may be applied to step S1105.

[0185] FIG. 12 is a flowchart of a method for updating an outlier detection model and an AI-based SON model according to one embodiment.

[0186] In step S1201, the AI ​​server can receive PM data, CM data, and KPI of the communication network from the base station.

[0187] The AI ​​server can repeatedly receive PM data, CM data, and KPI of the communication network from the base station, and store the received PM data, CM data, and KPI in a database.

[0188] In step S1202, the AI ​​server can determine whether an update of the outlier detection model and the AI-based SON model is required.

[0189] The AI ​​server may decide to perform an update of the outlier detection model and the AI-based SON model when any of the following is required: regular updates, updates at the request of the base station (100), updates at the request of the administrator, or updates at the request of the PM data, CM data, and KPI changes.

[0190] If the decision is not made to perform an update, the AI ​​server may proceed to step S1201. In this case, the AI ​​server may collect PM data, CM data, and KPIs from the communication network until the update is performed.

[0191] If it is decided to perform an update, the AI ​​server may proceed to step S1203.

[0192] In step S1203, the AI ​​server can update the outlier detection model and the AI-based SON model using the PM data, CM data, and KPI.

[0193] The AI ​​server can use PM data, CM data, and KPI collected in the database as learning data to update the outlier detection model and AI-based SON model.

[0194] FIG. 13A is a block diagram of an electronic device (1300) according to one embodiment.

[0195] The electronic device (1300) may include a processor (1310), a memory (1320), and a communication interface (1330). The processor (1310), the memory (1320), and the communication interface (1330) may communicate with each other via a bus.

[0196] The processor (1310) can control the overall operations of the electronic device (1300). For example, the processor (1310) can control the overall operations of the electronic device (1300) to determine a policy for SON by executing one or more instructions of a program stored in the memory (1320). There can be one or more processors (1310).

[0197] The processor (1310) may be configured as at least one of, but is not limited to, a CPU, a microprocessor, a GPU, an ASIC (Application Specific Integrated Circuits), a DSP (Digital Signal Processors), a DSPD (Digital Signal Processing Devices), a PLD (Programmable Logic Devices), an FPGA (Field Programmable Gate Arrays), an application processor, an NPU, or an artificial intelligence processor designed with a hardware structure specialized for processing artificial intelligence models.

[0198] The memory (1320) may store instructions, data structures, and program codes that can be read by the processor (1310). Operations performed by the processor (1310) may be implemented by executing instructions or codes of a program stored in the memory (1320).

[0199] The memory (1320) may include a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (e.g., SD or XD memory, etc.), and may include a non-volatile memory including at least one of a ROM (Read-Only Memory), an EEPROM (Electrically Erasable Programmable Read-Only Memory), a PROM (Programmable Read-Only Memory), a magnetic memory, a magnetic disk, or an optical disk, and a volatile memory such as a RAM (Random Access Memory) or an SRAM (Static Random Access Memory).

[0200] The memory (1320) may store one or more instructions and / or programs that cause the electronic device (1300) to operate to determine a policy for the SON.

[0201] The communication interface (1330) can perform data communication with other electronic devices under the control of the processor (1310).

[0202] The communication interface (1330) may include a communication circuit that can perform data communication between the electronic device (1300) and another electronic device using at least one of data communication methods including, for example, wired LAN, wireless LAN, Wi-Fi, Bluetooth, ZigBee, Wi-Fi Direct (WFD), infrared Data Association (IrDA), Bluetooth Low Energy (BLE), Near Field Communication (NFC), Wireless Broadband Internet (Wibro), World Interoperability for Microwave Access (WiMAX), Shared Wireless Access Protocol (SWAP), Wireless Gigabit Alliances (WiGig), or RF communication.

[0203] The communication interface (1330) can perform data communication with the base station. For example, the communication interface (1330) can receive PM data, CM data, and KPI from the base station, and transmit a determined policy and AI-based SON model to the base station.

[0204] FIG. 13b is a block diagram of an electronic device (1300) according to one embodiment.

[0205] In explaining Fig. 13b, any content overlapping with that explained in Fig. 13a is omitted for brevity.

[0206] In one embodiment, the memory (1320) may store instructions and / or programs for implementing the functions of the management module (1321), the outlier detection module (1322), and the policy switch module (1323). In addition, the memory (1320) may further store an outlier detection model (1324), an AI-based SON model (1325), and a database (1326).

[0207] The management module (1321), the outlier detection module (1322), and the policy switch module (1323) can be executed by the processor (1310). Since the descriptions related to the operations of each of the aforementioned modules have already been described in the descriptions of the previous drawings, a repeated description will be omitted.

[0208] Meanwhile, embodiments of the present disclosure may also be implemented in the form of a recording medium containing computer-executable instructions, such as program modules, executed by a computer. Computer-readable media may be any available media that can be accessed by a computer, and include both volatile and nonvolatile media, removable and non-removable media. Furthermore, computer-readable media may include computer storage media and communication media. Computer storage media include both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Communication media may typically include computer-readable instructions, data structures, or other data in a modulated data signal, such as program modules.

[0209] Additionally, a computer-readable storage medium may be provided in the form of a non-transitory storage medium. Here, the term "non-transitory storage medium" simply means a tangible device that does not contain signals (e.g., electromagnetic waves). This term does not distinguish between cases where data is permanently stored in the storage medium and cases where data is temporarily stored. For example, a "non-transitory storage medium" may include a buffer in which data is temporarily stored.

[0210] According to one embodiment, the method according to various embodiments disclosed in the present document may be provided as included in a computer program product. The computer program product may be traded as a product between a seller and a buyer. The computer program product may be distributed in the form of a machine-readable storage medium (e.g., compact disc read-only memory (CD-ROM)), or may be distributed online (e.g., downloaded or uploaded) through an application store or directly between two user devices (e.g., smartphones). In the case of online distribution, at least a portion of the computer program product (e.g., a downloadable app) may be temporarily stored or temporarily generated in a machine-readable storage medium, such as the memory of a manufacturer's server, an application store's server, or an intermediary server.

[0211] The above description of the present disclosure is provided for illustrative purposes only, and those skilled in the art will readily appreciate that modifications to other specific forms can be made without altering the technical spirit or essential features of the present disclosure. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, components described as being single may be implemented in a distributed manner, and similarly, components described as being distributed may be implemented in a combined manner.

[0212] The scope of the present disclosure is indicated by the claims described below rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present disclosure.

Claims

1. In determining the policy for SON, Step of receiving PM data from a communication network (S901); A step (S902) of generating feature data representing the PM data using a deep learning-based encoder; Step (S903) of determining an outlier score of the above feature data using a deep learning-based outlier detection model; and A method comprising a step (S904) of selecting one of an artificial intelligence-based policy and a non-artificial intelligence-based policy for performing SON in the communication network according to the above outlier score.

2. In paragraph 1, The step of selecting one of the above AI-based and non-AI-based policies is: a step of selecting the AI-based policy if the above outlier score is less than a predetermined outlier threshold; and A method comprising the step of selecting the non-AI based policy if the above outlier score is greater than or equal to the predetermined outlier threshold.

3. In paragraph 1 or 2, The step of generating feature data representing the above PM data is: A method comprising the step of generating the feature data such that the feature data of the PM data is located closer to each other in the feature space as the similarity between the CM data of the communication network corresponding to the PM data is higher using the deep learning-based encoder.

4. In any one of paragraphs 1 to 3, The step of generating feature data representing the above PM data is: A method comprising the step of generating the feature data such that the feature data of the PM data are located closer to each other in the feature space as the similarity of the KPI of the communication network corresponding to the PM data is higher using the deep learning-based encoder.

5. In any one of paragraphs 1 to 4, The above outlier detection model is a method learned to infer outlier scores of input data calculated by two or more outlier detection techniques and output a final outlier score indicating whether the input data is an outlier from the inferred outlier scores.

6. In any one of paragraphs 1 to 5, The step of generating feature data representing the above PM data is: A method comprising the step of generating feature data representing the PM data using the encoder updated based on the PM data, CM data, and KPI collected from the communication network.

7. In paragraph 6, The above updated encoder is, A method in which the similarity of the collected CM data and the similarity of the collected KPI are higher, so that the feature data of the collected PM data corresponding to the collected CM data and the collected KPI are learned to be located closer to each other in the feature space.

8. In any one of paragraphs 1 to 7, The step of determining the outlier score of the above feature data is: A method comprising the step of determining an outlier score of the feature data using an updated outlier detection model based on PM data collected from the communication network.

9. In paragraph 8, The above updated outlier detection model is, It is trained to infer outlier scores of each of the outlier learning data and the normal learning data calculated by two or more outlier detection techniques, and output a final outlier score indicating whether each of the outlier learning data and the normal learning data is an outlier from the inferred outlier scores. A method wherein the above outlier learning data and the above normal learning data are generated based on a distribution in a feature space of feature data representing the collected PM data.

10. In any one of paragraphs 1 to 9, The above AI-based policy includes a policy for performing SON on the communication network using an AI-based SON model, The above artificial intelligence-based SON model is learned based on the same training data as the deep learning model for determining the above outlier score.

11. In an electronic device (1300) that determines a policy for SON, A memory (1320) storing one or more instructions; and comprising one or more processors (1310) for executing one or more instructions stored in the memory; The one or more processors (1310) execute the one or more instructions, Receive PM data from the communication network, Generate feature data representing the above PM data using a deep learning-based encoder, Using a deep learning-based outlier detection model, the outlier score of the above feature data is determined, An electronic device configured to select one of an AI-based policy and a non-AI-based policy for performing SON in the communication network based on the above outlier score.

12. In paragraph 11, The one or more processors (1310) execute the one or more instructions, If the above outlier score is less than a pre-determined outlier threshold, the AI-based policy is selected, An electronic device configured to select the non-AI based policy if the above outlier score is greater than the predetermined outlier threshold.

13. In paragraph 11 or 12, The one or more processors (1310) execute the one or more instructions, An electronic device configured to generate feature data such that the feature data of the PM data is located closer to each other in a feature space as the similarity between the CM data of the communication network corresponding to the PM data is higher using the deep learning-based encoder.

14. In any one of paragraphs 11 to 13, The one or more processors (1310) execute the one or more instructions, An electronic device configured to generate feature data such that the feature data of the PM data are located closer to each other in a feature space as the similarity of the KPI of the communication network corresponding to the PM data is higher using the deep learning-based encoder.

15. A computer-readable recording medium having recorded thereon a program for executing the method of any one of clauses 1 to 10 on a computer.

Images