Method and device for supporting artificial intelligence model

Abstract

A method performed by a mobility management device, according to one embodiment of the present disclosure, may comprise the operations of: acquiring a first input sequence based on a past movement path of a terminal; converting, into single number or vector, a plurality of consecutive data points included in the first input sequence, so as to acquire a second input sequence based on the converted number or vector; and training a terminal location prediction model with respect to the second input sequence.

Description

Method and device for supporting artificial intelligence models

[0001] The embodiments of the present disclosure relate to artificial intelligence models, specifically to an apparatus and method for simplifying an input sequence required for learning to predict an artificial intelligence model.

[0002] An artificial intelligence system (or integrated intelligence system) is a computer system that implements human-level intelligence; it is a system in which machines learn and make judgments autonomously, and its recognition rate improves with use. Artificial intelligence technology consists of machine learning (deep learning) technology, which utilizes algorithms to autonomously classify and learn the characteristics of input data, and component technologies that utilize machine learning algorithms to mimic functions such as cognition and judgment of the human brain.

[0003] The elemental technologies may include, for example, linguistic understanding technology that recognizes human language / characters, visual understanding technology that recognizes objects like human vision, reasoning / prediction technology that judges information to logically reason and predict, knowledge representation technology that processes human experience information into knowledge data, and motion control technology that controls autonomous driving of vehicles and the movement of robots.

[0004] An electronic device equipped with an artificial intelligence (AI) system can support the function of learning input data acquired through a communication network and predicting future data through a neural network model based on the learning results when processing input data.

[0005] Furthermore, with the advancement of artificial intelligence systems, there is growing interest in the development of AI-based communication technologies that realize system optimization by internalizing AI support functions into the wireless communication system field. For example, as the need for managing the mobility of mobile communication terminals increases with the rise in the number of base stations, attempts to predict the location of mobile communication terminals using AI prediction models are continuing.

[0006] In the case of a terminal location prediction model, the terminal's future path can be predicted based on the path it has traveled in the past. For example, the path a terminal has traveled in the past can be represented as a sequence in which the base stations the terminal belonged to are arranged in a specific order (e.g., chronological order). In other words, sequence data consisting of base stations arranged in order can be used as input for training the terminal location prediction model.

[0007] In the case of prediction models based on sequence data, as the amount of input data increases, the amount of training and the complexity of the prediction model increase significantly; consequently, training costs increase, and execution may become impossible on devices with limited memory capacity.

[0008] The present disclosure may provide a method and apparatus for generating an artificial intelligence prediction model by simplifying model training for sequence input data.

[0009] A method and apparatus according to one embodiment may provide a method and apparatus for simplifying sequence input data to generate a terminal location prediction model in a wireless communication system.

[0010] A method performed by a mobility management device according to one embodiment of the present disclosure may include: an operation of obtaining a first input sequence based on a past movement path of a terminal; an operation of obtaining a second input sequence based on a converted number or vector by converting a plurality of consecutive data points included in the first input sequence into a single number or vector; and an operation of performing training of a terminal location prediction model on the second input sequence.

[0011] A mobility management device according to one embodiment of the present disclosure comprises: a transceiver; and a controller coupled to the transceiver, wherein the controller may be configured to: acquire a first input sequence based on the past movement path of a terminal, and acquire a second input sequence based on a converted number or vector by converting a plurality of consecutive data points included in the first input sequence into a single number or vector, and to perform learning of a terminal location prediction model on the second input sequence.

[0012] According to one embodiment of the present disclosure, the training cost of an artificial intelligence prediction model based on sequence input data can be reduced and the accuracy improved.

[0013] According to one embodiment of the present disclosure, the training cost of a terminal location prediction model, a network function (NF) load prediction model, and / or an NF anomaly detection model in a wireless communication system is reduced, thereby enabling the use of artificial intelligence models even in devices with limited capabilities, which can improve wireless communication performance in various conditions and environments.

[0014] The effects obtainable from the present disclosure are not limited to those mentioned in the embodiments of the present disclosure, and other unmentioned effects will be clearly understood by those skilled in the art to which the present disclosure pertains from the description below.

[0015] FIG. 1 is a flowchart of data processing according to one embodiment.

[0016] FIG. 2 is a diagram illustrating data preprocessing according to one embodiment.

[0017] FIG. 3 is a diagram illustrating a vector embedding according to one embodiment.

[0018] FIG. 4 shows the structure of a wireless communication system that supports artificial intelligence-based operation according to one embodiment of the present disclosure.

[0019] FIG. 5 illustrates an example of a paging message according to one embodiment.

[0020] FIG. 6 is a diagram illustrating the training of a terminal location prediction model according to one embodiment.

[0021] FIG. 7 is a diagram illustrating an artificial intelligence model according to one embodiment.

[0022] FIG. 8 is a diagram illustrating the accuracy of a terminal location prediction model according to one embodiment.

[0023] FIG. 9 is a diagram illustrating the result model performance of data preprocessing according to one embodiment.

[0024] FIG. 10 illustrates an example of a functional structure of a terminal according to one embodiment of the present disclosure.

[0025] FIG. 11 illustrates an example of the functional structure of a core network object according to one embodiment of the present disclosure.

[0026] The terms used in this disclosure are used merely to describe specific embodiments and are not intended to limit the scope of other embodiments. A singular expression may include a plural expression unless the context clearly indicates otherwise. Terms used herein, including technical or scientific terms, may have the same meaning as generally understood by those skilled in the art described in this disclosure. Terms used in this disclosure that are defined in a general dictionary may be interpreted as having the same or similar meaning as they have in the context of the relevant technology, and are not to be interpreted in an ideal or overly formal sense unless explicitly defined in this disclosure. In some cases, even terms defined in this disclosure are not to be interpreted to exclude the embodiments of this disclosure.

[0027] In the various embodiments of the present disclosure described below, a hardware-based approach is described as an example. However, since the various embodiments of the present disclosure include techniques using both hardware and software, the various embodiments of the present disclosure do not exclude a software-based approach.

[0028] Various embodiments are described in detail below with reference to the accompanying drawings. Furthermore, in describing the embodiments of this disclosure, specific descriptions of related known functions or configurations are omitted if it is determined that such detailed descriptions would unnecessarily obscure the essence of the embodiments. Additionally, terms used below are defined considering their functions in the embodiments, and these may vary depending on the intentions or practices of the user or operator. Therefore, their definitions should be based on the content throughout this specification.

[0029] For the same reason, some components in the attached drawings have been exaggerated, omitted, or schematically depicted. Additionally, the size of each component does not entirely reflect its actual dimensions. Identical or corresponding components in each drawing have been assigned the same reference numbers.

[0030] The advantages and features of the present disclosure and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present disclosure is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to make the present disclosure complete and to fully inform those skilled in the art of the scope of the disclosure, and the present disclosure is defined only by the scope of the claims. Throughout the specification, like reference numerals refer to like components.

[0031] At this time, each block of the process flow diagrams and combinations of the flow diagrams can be executed by computer program instructions. Since these computer program instructions can be loaded into the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing equipment, the instructions executed through the processor of the computer or other programmable data processing equipment create means to perform the functions described in the flow diagram block(s). Since these computer program instructions can also be stored in computer-available or computer-readable memory that can be directed toward the computer or other programmable data processing equipment to implement the function in a specific way, the instructions stored in computer-available or computer-readable memory can also produce a manufactured item containing means of instruction to perform the function described in the flow diagram block(s). Since computer program instructions can be loaded onto a computer or other programmable data processing equipment, instructions that perform a series of operations on the computer or other programmable data processing equipment to create a process executed by the computer can also provide operations for executing the functions described in the flowchart block(s).

[0032] Additionally, each block may represent a module, segment, or part of code containing one or more executable instructions for executing a specified logical function(s). It should also be noted that in some alternative execution examples, the functions mentioned in the blocks may occur out of order. For instance, two blocks described in succession may actually be executed substantially simultaneously, or the blocks may be executed in reverse order according to their corresponding functions.

[0033] In this case, the term “part” as used in various embodiments of the present disclosure refers to a software or hardware component such as a field programmable gate array (FPGA) or an application-specific integrated circuit (ASIC), and the “part” may perform certain roles. However, the “part” is not limited to software or hardware. The “part” may be configured to reside in an addressable storage medium or may be configured to run one or more processors. Accordingly, as an example, the “part” may include components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functions provided within the components and “parts” may be combined into a smaller number of components and “parts” or further separated into additional components and “parts.” In addition, the components and '~parts' may be implemented to play one or more CPUs (central processing units) within the device or secure multimedia card.

[0034] According to one embodiment, an electronic device may perform data preprocessing using at least one of hardware such as a digital signal processor, an FPGA, a GPU, a dedicated AI accelerator, and / or a general-purpose CPU. A digital signal processor (DSP) may refer to a microprocessor that processes digital signals in real time. A digital signal processor may convert analog signals received from wireless communication into a digital form that can be processed by an artificial intelligence algorithm, and may perform filtering and modulation / demodulation. A field programmable gate array (FPGA) is a programmable semiconductor device capable of performing high-speed parallel processing. A graphics processing unit (GPU) is a processor optimized for parallel processing capable of rapidly processing large amounts of data. A dedicated AI accelerator is a dedicated chip optimized for AI computation, including an application-specific integrated circuit (ASIC), an NPU, and a TPU, and may perform data preprocessing and AI model execution. A general-purpose CPU is a general-purpose processor capable of executing complex preprocessing algorithms or performing the role of controlling and coordinating other hardware components.

[0035] The data preprocessing method of the present disclosure may be characterized by grouping or integrating an input sequence into segments.

[0036] FIG. 1 is a flowchart of data processing according to one embodiment. FIG. 2 is a diagram for explaining data preprocessing according to one embodiment.

[0037] In operation 110, the electronic device can collect original data. Referring to FIG. 1, the original data collected by the electronic device in operation 110 may include at least one of a text sequence, time-series data arranged sequentially over time, an audio sequence, a video sequence, a sensor data sequence, and a biological sequence. The collected data can be used as input to an artificial intelligence model after undergoing data preprocessing. Referring to FIG. 2, input sequence S u The data points (S1, S2, ..., Si, ..., Sn) undergo data preprocessing and input sequence e u It can be converted to. Input sequence e u It can be used as input for an artificial intelligence model.

[0038] Input sequence S u can refer to a sequence of data arranged in order over time, and can be referred to as a time-series input sequence. Input sequence S u The data points (S1, S2, ..., Si, ..., Sn) may be recorded sequentially over time. However, this is merely an example, and the input sequence S u The input sequence is not limited to a time-series input sequence. The present disclosure relates to a method for processing an input sequence in artificial intelligence technology.

[0039] Input sequence S uThe number of data points (n) included in (210) may be referred to as the length of the time series input sequence or the input size, and may be variable rather than fixed. The length of the input sequence may be determined based on the collected original data, the amount of learning required for prediction, or, in the case of time series data, the time interval for determining data points. For example, if 12 hours of data are learned and data points are determined from the data at 5-minute intervals, the number of data points (n) may be 144.

[0040] An electronic device according to one embodiment comprises an input sequence S u In the case of (210), the learning cost can be effectively reduced by performing data preprocessing using a data grouping or aggregation technique. In operations 120 to 130 of FIG. 1, the electronic device can perform data preprocessing based on a data grouping or aggregation technique. Specifically, in operation 120, the electronic device can group a number of data points included in a time series input sequence into a single segment.

[0041] Referring to FIG. 2, the electronic device input sequence S u Among the data points of (210), consecutive data points of a can be organized into a single group. The present disclosure refers to a group organized to include consecutive l data points as a segment, and may refer to organizing l data points into a segment as grouping or aggregation. As a result of organizing all data points included in a time series input sequence into segments, the time series input sequence can be expressed as a sequence of segments (220).

[0042] For example, as shown in FIG. 2, when a=2, two consecutive data points can be organized into a single segment. Data point (S1) and data point (S2) are organized into segment 1, data point (S3) and data point (S4) are organized into segment 2, ..., data point (S i-1 ) and data points (S i ) is organized into segment i / a, ..., data point (S n-1 ) and data points (S n ) can be organized into segment n / a. That is, input sequence S u (210) can be represented as a sequence of segments (220).

[0043] In operation 130, the electronic device can perform additional data conversion to determine an input sequence (230) in a form suitable for processing by an artificial intelligence model. The electronic device can perform encoding on the segment sequence (220) to convert it into a form suitable for use as input to an artificial intelligence model.

[0044] In operation 140, the electronic device can perform model learning based on a preprocessed input sequence (230). Input sequence S u The input sequence e obtained by grouping or integrating into segments and converting the segments into a single number u The number of is n / a. By grouping into segments, the length of the input sequence can be reduced to 1 / a. The electronic device inputs the sequence e u Since model training is performed based on this, the training speed can be increased and resource or power consumption can be reduced.

[0045] The present disclosure provides two embodiments as a data conversion method of operation 130.

[0046] Example 1. A method for converting or mapping the values ​​of multiple data points included in a segment into a single number, and restoring the value of each data point and the order between the data points from the mapped number.

[0047] Example 2. A method for compressing input vectors included in a segment into a single vector by adding them. Subsequently, learning can be performed based on the ratio of each component of the compressed vector.

[0048] Examples 1 and 2 will be described below.

[0049] According to Example 1, the electronic device can convert the values ​​of data points included in a segment into a single number.

[0050] According to one embodiment, an electronic device can encode the values ​​of a plurality of data points included in a segment and convert or map them into a single number. Additionally, because the weights applied to each value of the data points differ depending on the order of the data points within the segment (e.g., first, second, ..., a-th), the electronic device can decode the mapped number to restore the value and order of each data point.

[0051] For example, when a=2, data points (S i-1 ) and data points (S i Segment including ) i / 2 is a single number (e i / 2 The encoding for conversion to ) can be expressed by Equation 1. As in Equation 1, data points (S i-1 ) is k 1 Mapped to, and data points (S i ) is k 0 Since it is mapped to, the order information of the data points can be preserved.

[0052] [Mathematical Formula 1]

[0053] e i / 2 = S i-1 × k1 + S i × k 0 = S i-1 × k 1 + S i

[0054] Here, k may represent the number of possible values ​​that Si can take. If Si takes a value corresponding to one of k consecutive integers, e is determined based on Equation 1. i / 2 Segment from value i / 2 S constituting i-1 , S i The unique pair of (S i-1 , S i Can infer or determine ).

[0055] For example, if Si has a value corresponding to one of the integers from 0 to (k-1), S i The value of is an integer less than k. Dividing both sides of Equation 1 by k, (e i / 2 / k) = S i-1 + (S i / k) is obtained, S i Since the value of is smaller than k, S i The value of / k is a number less than 1. Therefore, S i = k × {(e i / 2 / value after the decimal point of k)}, S i-1 = {(e i / 2 It can be determined by the value of the integer part of / k).

[0056] As another example, if Si has a value corresponding to one of the integers from 1 to k, S i The value of is an integer less than or equal to k. Dividing both sides of Equation 1 by k, (e i / 2 / k) = S i-1 + (S i / k) is obtained, S i Since the value of is less than or equal to k, S i The value of / k is a number less than or equal to 1. (e i / 2If the decimal part of / k) is 0, S i = k, S i-1 = (e i / 2 / k) - can be determined as 1. (e i / 2 If the decimal part of / k) is not 0, S i = k × {(e i / 2 / value after the decimal point of k)}, S i-1 = {(e i / 2 It can be determined by the value of the integer part of / k).

[0057] That is, Si has a value corresponding to one of k consecutive integers, and according to Equation 1, the data point (S i-1 ) has k 1 As it is multiplied, the data point (S i ) has k 0 Since it is multiplied, e obtained based on Mathematical Formula 1 Unique pair (S) from the value i-1 , S i Can infer or determine ). Pair(S i-1 , S i The ability to determine ) means, S i-1 , S i It can be understood that each value can be determined. This means that the value and order of each data point can be restored by decoding the numbers obtained based on Mathematical Formula 1.

[0058] For example, when k=4, S can have a value corresponding to one of 0, 1, 2, or 3. When S1=0 and S2=3, Segment1=(S1, S2) can be converted to e1= 0×4 + 3 = 3 by Equation 1. From e1=3, S1=0 and S2=3 can be inferred or determined.

[0059] Mathematical Equation 1 can be generalized to Mathematical Equation 2 for any a. Data point (S i-a+1 ), data points (S i-a+2 ), ..., data points(Si-a+a , that is, S i Segment i / a containing ) is a single number (e i / a The encoding for converting to ) can be expressed as in Equation 2.

[0060] [Mathematical Formula 2]

[0061] e i / a = S i-a+1 × k (a-1) + S i-a+2 × k (a-2) + ... + S i × k (a-a)

[0062] Likewise, data points (S i-a+1 ), data points (S i-a+2 ), ..., data points(S i ) can be mapped to a single number based on mathematical formula 2, and the value and order of each data point can be restored by decoding the mapped number.

[0063] However, Equations 1 and 2 are merely examples for illustrative purposes and do not limit the encoding method of Embodiment 1 of the present disclosure. The encoding method of Embodiment 1 simply refers to an encoding method capable of converting or mapping the values ​​of a plurality of data points included in a segment into a single number, and restoring the value of each data point and the order between the data points from the mapped number. For example, a function such as Equation 3 may be used as the encoding method of Embodiment 1.

[0064] [Mathematical Formula 3]

[0065] e i / 2 = ((S i-1 + S i ) 1 (S i-1 + S i + 1)) / 2 + S i

[0066] [Mathematical Formula 4]

[0067] w = floor((sqrt(8 e i / 2 + 1) - 1) / 2)

[0068] t = (ww + w) / 2

[0069] S i = e i / 2 - t

[0070] S i-1 = w - S i

[0071] Mathematical Equation 4 represents the decoding function for the encoding function of Mathematical Equation 3, and given e i / 2 Regarding S i-1 and S i Since it can be determined, the values ​​and order of the data points constituting the segment can be restored.

[0072] Equations 3 and 4 can be referred to as Cantor pairing functions and can be applied when two data points are included in a single segment (i.e., when a=2). As with Equations 1 and 2, encoding and decoding through the Cantor pairing function can be extended to cases where a=3, 4, ...

[0073] According to Example 2, the electronic device can convert a vector of data points included in a segment into a single vector.

[0074] An electronic device according to one embodiment can compress vector-format data points included in a segment into a single vector by adding them. For example, when k=4 and a=2, the data points can be converted to S1=<1,0,0,0> and S2=<0,1,0,0> by one-hot encoding. When vector compression is performed, Segment1 is determined to be <1,1,0,0>, so the length of the sequence can be reduced by about half. Furthermore, when k=4 and a=4, the result of vector compression for S1=<1,0,0,0>, S2=<0,1,0,0>, S3=<0,0,1,0>, and S4=<0,0,0,1> is determined to be Segment1=<1,1,1,1>, so the length of the sequence can be reduced by about one-fourth. Through vector compression of the data points constituting the segment, the learning overhead can be significantly reduced.

[0075] Vector compression according to one embodiment can determine real number values ​​other than 0 and 1, such as <25,30,0,0>, by additionally reflecting elements such as frequency and / or time into Segment1=<1,1,0,0> compressed into a single vector. As a result, computational efficiency, ease of visualization, interpretability, applicability of various algorithms, and the displayability of continuous relationships in categorical data can be improved. The accuracy of the prediction model can be improved.

[0076] When performing vector compression, it is not possible to determine whether S1=<1,0,0,0> and S2=<0,1,0,0> or S1=<0,1,0,0> and S2=<1,0,0,0> from Segment1=<1,1,0,0>, and thus the order of data points S1 and S2 included in Segment1 cannot be determined. The vector compression method of Example 2 can significantly reduce the learning overhead and improve the accuracy of the prediction model by sacrificing the order information of the data points included in the segment.

[0077] In operation 130 of FIG. 1, the electronic device can perform preprocessing on the segment using the data conversion method of Example 1 and / or Example 2, and can determine or generate a preprocessed input sequence.

[0078] According to one embodiment, the electronic device may further perform data reduction, such as dimensionality reduction and / or sampling. When the encoding of Embodiment 1 is applied in Operation 130, the complexity of the encoding and decoding processes of the data transformation may increase significantly depending on the size of one Segment (i.e., the number of data points contained in one Segment (a)). For example, in Equation 2, the data transformed number (e i / a ) is expressed as a power of the total number of cases (k). Therefore, in operations 120 to 130, the input sequence S u When applying the encoding of Example 1 by integrating (210) into segments, memory usage can be optimized by reducing it into a small-dimensional vector using a dimensionality reduction method such as vector embedding.

[0079] In operation 135, the electronic device can further perform data reduction to reduce the size and complexity of the data and shorten model training and prediction times. Additionally, the electronic device can reduce storage space and memory usage by performing data reduction. For example, in operation 135, the electronic device can further perform vector embedding.

[0080] FIG. 3 is a diagram illustrating vector embedding according to one embodiment. Referring to FIG. 3, an electronic device can reduce a large dataset of more than 10,000 dimensions into a vector of less than 1,000 dimensions based on similarity or relationships between data. Referring to FIG. 3, since the matrix (310) is a high-dimensional vector, it increases training time, model complexity, computational cost, and memory usage, and since only one component is 1 and all other components are 0, it can reduce the efficiency of machine learning algorithms. As the dimensionality is reduced as in the result matrix (320) of vector embedding, storage space can be saved and the speed of matrix operations can be increased. In addition, as all components become real numbers through vector embedding as in the matrix (320), computational efficiency, ease of visualization, interpretability, applicability of various algorithms, and the displayability of continuous relationships of categorical data can be improved.

[0081] Subsequently, in operation 140, the electronic device can perform model training based on the preprocessed input sequence (230). At this time, the preprocessed input sequence (230) may include dimensionality-reduced data.

[0082] Specific embodiments using the data preprocessing method of the present disclosure are provided below. However, the specific embodiments are merely examples for explaining the data preprocessing method and do not limit the present disclosure. The data preprocessing method of the present disclosure can be applied as long as it is for the training of an artificial intelligence model that needs to process sequence inputs.

[0083] Hereinafter, the base station, as the entity performing resource allocation for terminals, may be at least one of an eNode B (eNB), Node B, BS (base station), RAN (radio access network), AN (access network), RAN node, NR NB, gNB, radio access unit, base station controller, or a node on a network. The terminal may include a UE (user equipment), MS (mobile station), cellular phone, smartphone, computer, or a multimedia system capable of performing communication functions. In the various embodiments of the present disclosure, the case where the terminal is a UE is described as an example. Furthermore, although the various embodiments of the present disclosure are described below using systems based on LTE, LTE-A, or NR as examples, the various embodiments of the present disclosure may be applied to other communication systems having similar technical backgrounds or channel types. Additionally, the various embodiments of the present disclosure may be applied to other communication systems with some modifications made at the discretion of a person with skilled technical knowledge, without significantly departing from the scope thereof.

[0084] The present disclosure describes various embodiments using terms used in some communication standards (e.g., 3GPP (3rd Generation Partnership Project)), but this is merely illustrative. Various embodiments of the present disclosure can be easily modified and applied to other communication systems. Below, some terms used in the core network of the present disclosure are defined in advance.

[0085] AMF: Access and Mobility Management Function

[0086] NF: Network Function

[0087] NWDAF: Network Data Analytics Function

[0088] CN: Core Network

[0089] CNF: Containerized Network Function

[0090] PCF: Policy Control Function

[0091] SMF: Session Management Function

[0092] UDM: User Data Management

[0093] UPF: User Plane Function

[0094] The operating principles of embodiments according to the present disclosure will be described in detail below with reference to the attached drawings. The meaning of terms used to describe the embodiments may be determined by the content throughout this specification.

[0095] Terms used in this disclosure to refer to network entities or network functions and objects, terms referring to messages, terms referring to identification information, etc., are illustrative for the sake of convenience of explanation. Accordingly, the present invention is not limited to the terms described below, and other terms referring to objects having equivalent technical meanings may be used.

[0096] For convenience, embodiments according to the present disclosure may be described using terms and names defined in 5G system specifications, but the meaning is not limited by the terms and names and may be applied equally to systems conforming to other specifications.

[0097] FIG. 4 shows the structure of a wireless communication system that supports artificial intelligence-based operation according to one embodiment of the present disclosure.

[0098] FIG. 4 illustrates a base station (410), a first terminal (420), and / or a second terminal (430) as part of the nodes utilizing a wireless channel in a wireless communication system. FIG. 4 illustrates only one base station, but this is merely an example. The wireless communication system of FIG. 4 may include other base stations identical or similar to the base station (410). For example, terminals (420, 430) may move within the coverage of multiple base stations.

[0099] A base station (410) is a network infrastructure that provides wireless access to terminals (420, 430). The base station (410) has coverage defined as a certain geographical area based on the distance at which it can transmit signals. In addition to being a base station, the base station (410) may be referred to as an 'access point (AP)', 'eNodeB (eNB)', 'gNodeB (gNB)', '5G node (5th generation node)', 'wireless point', 'transmission / reception point (TRP)', or other terms having an equivalent technical meaning.

[0100] Each of the first terminal (420) and the second terminal (430) is a device used by a user and can perform communication with the base station (410) via a wireless channel. At least one of the first terminal (420) or the second terminal (430) can be operated without user involvement. For example, at least one of the first terminal (420) or the second terminal (430) may be a device that performs machine type communication (MTC) and may not be carried by the user. Each of the first terminal (420) and the second terminal (430) may be referred to as 'user equipment (UE)', 'mobile station', 'subscriber station', 'customer premises equipment (CPE)', 'remote terminal', 'wireless terminal', 'electronic device', or 'user device', or any other term having an equivalent technical meaning.

[0101] The base station (410), the first terminal (420), and the second terminal (430) can transmit and / or receive wireless signals in a millimeter wave (mmWave) band (e.g., 28 GHz, 30 GHz, 38 GHz, 60 GHz). At this time, to improve channel gain, the base station (410), the first terminal (420), and / or the second terminal (430) can perform beamforming.

[0102] Beamforming may include transmitting beamforming and / or receiving beamforming. That is, the base station (410), the first terminal (420), and / or the second terminal (430) may give directivity to the transmitted signal or the received signal. To give directivity to the received signal, the base station (410) and / or the terminals (420, 430) may select serving beams (412, 413, 421, 431) through a beam search or beam management procedure. After the serving beams (412, 413, 421, 431) are selected, subsequent communication may be performed through a resource that is in a quasi-co-located (QCL) relationship with the resource that transmitted the serving beams (412, 413, 421, 431).

[0103] The base station (410), the first terminal (420), and the second terminal (430) of the present disclosure may each be a transmitting apparatus, a transmitting node, a receiving apparatus, and / or a receiving node. For example, the base station (410) may transmit a radio frequency (RF) signal to the first terminal (420). The base station (410) may receive an RF signal from the first terminal (420). As another example, the first terminal (420) may transmit an RF signal to the base station (410) or the second terminal (430). The first terminal (420) may receive an RF signal from the base station (410) or the second terminal (430).

[0104] The structure of a wireless communication system supporting artificial intelligence-based operation may include various network functions (NF), and FIG. 4 may illustrate a base station (410) and terminals (420, 430) corresponding to some of these. Additionally, network functions may include an access and mobility management function (AMF), a session management function (SMF), unified data management (UDM), a data network (DN) or a local part of the DN capable of local access to the data network, a user plane function (UPF), and / or a (radio) access network ((R)AN).

[0105] Network functions can support the following functions.

[0106] An AMF may be a function or device responsible for both wireless access network connectivity and terminal mobility management. An AMF may provide functions for terminal-level connectivity and mobility management, and by default, one AMF may be connected per terminal.

[0107] (R)AN can transmit and receive data wirelessly with the terminal and transmit and receive terminal user plane data with the UPF. In addition, it can perform access and mobility management of the terminal in conjunction with the AMF.

[0108] A DN (Data Network) can represent, for example, operator services, internet access, or third-party services. A DN can transmit downlink protocol data units (PDUs) to a UPF or receive PDUs transmitted from a terminal (UE) from the UPF. A local part of a DN is a part of a DN that can represent a data network with a short data transmission path where local access is possible. It can be used to refer to a DN where edge application servers are deployed to support edge computing services.

[0109] The PCF can receive information about packet flows from the application server and provide functions for determining policies such as mobility management and session management. Specifically, the PCF can support a unified policy framework for controlling network behavior, provide policy rules so that control plane function(s) (e.g., AMF, SMF, etc.) can enforce policy rules, and support front-end implementations for accessing relevant subscription information within the unified data repository (UDR) for policy decisions.

[0110] SMF provides session management functions, and if a terminal has multiple sessions, each session can be managed by a different SMF.

[0111] UDM can store the terminal's subscription information, the context used by the terminal within the network, or policy data, etc.

[0112] UPF can forward downlink PDUs received from DN to terminals via (R)AN, and can forward uplink PDUs received from terminals to DNs via (R)AN.

[0113] These network functions can use terminal location prediction models for paging, registration area updates, edge computing, and / or handover optimization.

[0114] For example, a wireless communication device for mobility management in a core network (e.g., a mobility management entity (MME) or an AMF) may use a terminal location prediction model in a paging procedure to locate terminals in idle mode. FIG. 5 illustrates an example of a paging message according to one embodiment. Referring to FIG. 5, the MME or AMF (550) may initiate a paging procedure by transmitting a paging message to a base station (510) via an N2 interface. When the base station (510) receives the paging message, the base station (510) may obtain a tracking area identity (TAI) list. The base station (510) may broadcast the paging message to terminals in cells belonging to the tracking area indicated by the TAI list. Since the paging procedure generates a large amount of traffic, an increase in the number of base stations may cause a burden on the terminal mobility management of network functions. When an AMF or MME (550) uses a terminal location prediction model, the paging range can be narrowed based on the predicted location, and network resources can be saved by not sending paging messages to unnecessary base stations or cells. Additionally, the AMF or MME (550) can reduce the paging response time by giving high priority to base stations at the predicted location.

[0115] As another example, a wireless communication device (e.g., MME, AMF) can predict the location of a terminal and provide a tracking area list (TAL) to the terminal based on the prediction result. The AMF or MME can predict the cells the terminal will move to next based on the learning results regarding the terminal's previous movement path. For example, the terminal's previous movement path may include information regarding the terminal's previous connection history, movement patterns, connection time (or dwell time) per base station, and / or paging history. The AMF or MME can determine the TAL associated with the predicted cells. By determining the TAL that reflects the terminal's previous movement path, the AMF or MME can reduce the number of unnecessary registration procedures within the network.

[0116] As another example, a wireless communication device (e.g., a terminal) may use an artificial intelligence model (e.g., a machine learning model) to perform operations such as channel state information prediction, channel state information compression, and positioning.

[0117] A data preprocessing method according to one embodiment can be applied to a terminal location prediction model. In the terminal location prediction model, the path traveled by the terminal in the past can be used as the sequence input to the prediction model. If the time to be predicted becomes long and / or the number of candidate base stations (k) to which the terminal can move increases, a large amount of training time and computational resources are required. When using one-hot encoding, the matrix corresponding to the data point (Si) of the sequence input has a k-dimensional vector. For example, when the number of base stations (k) = 100, Si has a 100-dimensional matrix, with only one component value being 1 and the remaining 99 components having values ​​of 0.

[0118] A data preprocessing method according to one embodiment can be applied to a terminal location prediction model. In the case of a terminal location prediction model, the original input sequence (e.g., input sequence S) u (210)) may be a sequence of data arranged in order of the paths the terminal has traveled in the past. The values ​​of the data points (S1, S2, ..., Si, ..., Sn) included in the original input sequence may be determined based on the identification information of the base stations the terminal has previously connected to. For example, if the number of candidate base stations is 100 and the 100 candidate base stations are referred to as base station 0, base station 1, ..., base station 99, the value of the data point when the terminal connects to base station 0 may be 0, the value of the data point when the terminal connects to base station 1 may be 1, the value of the data point when the terminal connects to base station 2 may be 2, ..., and the value of the data point when the terminal connects to base station 100 may be 100. A sequence in which S1=1, S2=7, S3=5, S4=10 may indicate that the terminal connected to base station 1, base station 7, base station 5, and base station 10 in order.

[0119] In operation 120, the electronic device can group data points into segments in which identification information of base stations that the terminal has previously connected to is arranged in order. For example, when grouping two data points into one segment (i.e., a=2), the data points can be grouped as Segment1=(S1,S2) and Segment2=(S3,S4).

[0120] In operation 130, the electronic device may integrate the values ​​of data points included in the segment into a single number (Example 1) or compress a vector of data points included in the segment into a single vector (Example 2). For example, according to Equation 1 of Example 1, Segment1=(S1,S2) is e1= 1 × 100 1 + 7 × 100 0= can be transformed into 107. From the given a=2, k=100, and e1=107, S1=1 and S2=7 can be inferred. Therefore, from e1=107, a movement path in units of 2 base stations from 'Base Station 1 to Base Station 7' ​​can be inferred or determined. The movement path in units of 2 base stations that can be determined from e1=107 is the unique movement path from 'Base Station 1 to Base Station 7'. Therefore, the input sequence e u (230) is the input sequence S u (210) can represent the same terminal movement path.

[0121] According to the data preprocessing of the present disclosure, input sequence S u By grouping or consolidating into segments and converting the segments into a single number, a new input sequence e u ...can be determined. According to the data preprocessing of the present disclosure, the number of data points to be learned by an artificial intelligence model is input sequence S u It can be reduced to 1 / a. For example, if data points are grouped two at a time into one segment (i.e., a=2), the length of the input sequence can be reduced by half. The electronic device inputs the sequence e u Since model training is performed based on this, the training speed can be increased and resource or power consumption can be reduced.

[0122] Additionally, the electronic device may perform data reduction on the transformed data, as in Operation 135. In the case of a terminal location prediction model, the electronic device can determine the relationship between base station 0, base station 1, ..., base station N. The relationship between base stations is determined using collected data and may be determined based on the distance between base stations. For example, in the collected data, S i =1, S i+1 = 5 (i is an arbitrary integer) or S i =5, S i+1When the pattern of data points with a value of 1 is repeated, the electronic device recognizes the repetitive pattern and can identify the relationship between base station 1 and base station 5 (i.e., proximity of locations) based on the repetitive pattern. In a terminal location prediction model, the electronic device can perform vector embedding on the transformed data based on the relationship between candidate base stations (or the degree of proximity of locations).

[0123] FIG. 6 is a diagram illustrating the training of a terminal location prediction model according to one embodiment. Terminal location prediction may be repeated at regular time intervals. For example, an electronic device may periodically repeat the process of collecting, preprocessing, and training data. However, to prevent the amount of training, complexity, and computation from increasing with repetition, the electronic device [requires] an input sequence e u Based on this, output data corresponding to the dwell time of each base station of the terminal can be formed.

[0124] As described in Example 2, the electronic device can obtain the base station-specific dwell time of the terminal based on an input sequence associated with the terminal location. The output can be expressed as a single vector, and the value of each vector can be expressed as a base station and the dwell time at the base station.

[0125] Referring to FIG. 6, the electronic device can collect terminal location data every P=5 minutes as input to a terminal location prediction model.

[0126] The electronic device can generate output data based on input data. The output data may include the terminal's stay time per base station. The electronic device can obtain the terminal's base station stay time based on the terminal's location sequence input. The electronic device can reduce learning costs by representing the terminal's location over time as a single vector.

[0127] FIG. 7 is a diagram illustrating an artificial intelligence model according to one embodiment. As an artificial intelligence model for processing sequence inputs, a recurrent neural network (RNN), a long short-term memory (LSTM), a gated recurrent unit (GRU), a bidirectional RNN, and a transformer may be used. Referring to FIG. 7, a Stacked GRU (710) may be constructed by stacking multiple GRU layers vertically. The output of each GRU layer may be used as the input to the next GRU layer. While stacking GRU layers increases the expressiveness of the model, increasing the number of units in each GRU layer and the number of GRU layers may increase the computational complexity. Referring to FIG. 7, past location information of a terminal in the first layer may be received as input. Data points (S1, S2, ..., S) of the input sequence are received in each GRU unit of the first layer. i-1 , S i , ..., S n-1 , S n ) can be input in order. The Stacked GRU (710) may include additional hidden layers that are not directly involved in the input and output. Referring to FIG. 7, the Stacked GRU (710) may include two additional hidden layers (e.g., a second layer, a third layer). The output of the last GRU unit (719) of the top layer may be the final output of the Stacked GRU (710).

[0128] An electronic device according to one embodiment comprises data points (S1, S2, ..., S) of an input sequence. i-1 , S i , ..., S n-1 , S n Segments of a items in order (Seg1, ..., Seg i / 2 , ..., Seg n / 2It can be grouped into ). The electronic device transforms the segment to create a new input sequence e u ...can be determined. As a result of performing the data preprocessing of the present disclosure, the length of the input sequence is reduced, so the first layer of the Stacked GRU (710) required for model training does not need to have all n GRUs. Referring to FIG. 7, e i It can be input into the GRU group (713). The GRU group (713) may be composed of a single GRU unit to make calculations simpler. The output of the GRU group (713) can be input into the GRU group (715) of the second layer, and the output of the GRU group (715) can be input into the GRU group (717) of the third layer.

[0129] An electronic device according to one embodiment comprises an input sequence S of size n. u By grouping into segments and transforming, an input sequence e of size n / a u Model training can be performed using. The electronic device has an input sequence S of size n. u By grouping and converting into segments, the number of calculations of the GRU units of the stacked GRU (710) or the number of GRU units themselves can be reduced, and the computational complexity can be mitigated.

[0130] According to one embodiment, the output layer of an artificial intelligence model can generate output data using a fully connected (FC) neural network (720). The FC neural network (720) can receive the output of a stacked GRU (710) (i.e., the output of a GRU unit (719)) as input. The FC neural network (720) can generate output data equal to the number (k) of base stations to be predicted.

[0131] According to one embodiment, the output data value may be expressed as a percentage (%) of the time the terminal stays at the base station. For example, the electronic device may represent the time the terminal stays (or will stay) at the corresponding base station for a specified time (e.g., 4 hours) per base station as a percentage. The softmax activation function used for the FC neural network (720) may output a value normalized to a value between 0 and 1. The sum of the output values ​​may always be 1.

[0132] FIG. 8 is a diagram illustrating the accuracy of a terminal location prediction model according to one embodiment. To verify the proposed terminal location prediction model, location and time information of the terminal's movement in active mode were recorded, and a dataset was collected. Power consumption was evaluated using a Stacked GRU Model that learns a previous movement path of 12 hours and predicts the next 8 hours. When two input sequences are integrated into a segment to reduce the number of input sequences by half, the power consumption of the learning model can be reduced by approximately 21.1%. The power consumption reduction effect is greater the longer the input sequence (when learning a longer previous path or when UE handover occurs more frequently), and the power saving effect can be increased by increasing the size of the integrated segment. However, since the number of segments to be integrated affects the prediction accuracy, the appropriate size of the segment can be determined after analyzing the data of the applied system.

[0133] FIG. 9 is a diagram illustrating the performance of a resulting model of data preprocessing according to one embodiment. Referring to FIG. 9, the X-axis of the graph represents the registration area (RA), and the Y-axis of the graph represents the number of signalings associated with the occurrence of a Registration Update. Referring to FIG. 9, when data preprocessing (input sequence aggregation, ISA) that groups and integrates the input sequences of the present disclosure is performed, when the size of the RA is larger than the conventional 12%, it exhibits a level of prediction accuracy similar to that of using one-hot encoding. Furthermore, when the size of the RA is smaller than the conventional 12%, the data preprocessing method of the present disclosure can exhibit higher prediction accuracy than that of using one-hot encoding. Additionally, across the entire range, the data preprocessing method of the present disclosure can exhibit better performance than simply applying vector embeddings.

[0134] A data preprocessing method according to one embodiment can be applied to a network function load prediction model. The data preprocessing method of the present disclosure can also be applied to a network function (NF) load prediction model that supports network management and optimization by predicting the performance and capacity of network functions. The NF load prediction model may aim to proactively respond to situations such as traffic increase or failure. In the NF load prediction model, the trend of the NF load information can be used as a sequence input to the prediction model. For example, the NF load information that can be used as a sequence input may include CPU usage, memory usage, Session / UE count, delay time, and / or throughput.

[0135] In addition, the electronic device may perform data reduction on the transformed data, as in operation 135. In the case of an NF load prediction model, the electronic device may use the collected data to perform vector embedding based on the relationship between NF load information based on the characteristics of the NF (type (e.g., PCF, UDM, CHF, ...), location, computational resource, capacity).

[0136] A data preprocessing method according to one embodiment can be applied to a network function anomaly detection model. The data preprocessing method of the present disclosure can also be applied to an NF anomaly detection model that learns the normal distribution of data and detects anomalies based thereon. In an NF anomaly detection model, inter-NF interaction quality information can be used as a sequence input to a prediction model. For example, inter-NF interaction quality information that can be used as a sequence input may include key indicators that measure network performance and reliability, such as delay, attempt count, fail count, and timeout count.

[0137] In addition, the electronic device may perform data reduction on the converted data, as in operation 135. In the case of an NF anomaly detection model, the electronic device may use the collected data to perform vector embedding based on the relationship between inter-NF interoperability quality information based on the characteristics of the NF (type (e.g., PCF, UDM, CHF, ...), location, computational resource, capacity).

[0138] FIG. 10 illustrates an example of a functional structure of a terminal according to an embodiment of the present disclosure. The configuration exemplified in FIG. 10 can be understood as a configuration of a terminal. Terms such as '...part', '...unit', etc. used below refer to a unit that processes at least one function or operation, and this may be implemented in hardware or software, or a combination of hardware and software. The terminal of FIG. 10 may correspond to the terminal of FIG. 4.

[0139] Referring to FIG. 10, the terminal may include a communication unit (1005), a storage unit (1010), and a control unit (1015).

[0140] The communication unit (1005) can perform functions for transmitting and receiving signals through a wireless channel. For example, the communication unit (1005) can perform a conversion function between a baseband signal and a bit sequence according to the physical layer specifications of the system. For example, when transmitting data, the communication unit (1005) can generate complex symbols by encoding and modulating the transmitted bit sequence. Also, when receiving data, the communication unit (1005) can restore the received bit sequence by demodulating and decoding the baseband signal. Additionally, the communication unit (1005) can up-convert the baseband signal into an RF band signal and transmit it through an antenna, and down-convert the RF band signal received through the antenna into a baseband signal. For example, the communication unit (1005) may include a transmission filter, a reception filter, an amplifier, a mixer, an oscillator, a DAC, an ADC, etc.

[0141] Additionally, the communication unit (1005) may include a plurality of transmission and reception paths. Furthermore, the communication unit (1005) may include at least one antenna array composed of a plurality of antenna elements. In terms of hardware, the communication unit (1005) may be composed of a digital circuit and an analog circuit (e.g., a radio frequency integrated circuit (RFIC)). Here, the digital circuit and the analog circuit may be implemented as a single package. Additionally, the communication unit (1005) may include a plurality of RF chains. Furthermore, the communication unit (1005) may perform beamforming.

[0142] The communication unit (1005) can transmit and receive signals as described above. Accordingly, all or part of the communication unit (1005) may be referred to as a 'transmitter', a 'receiver', or a 'transmitter / receiver'. Furthermore, in the following description, transmission and reception performed via a wireless channel are used to mean that processing as described above is performed by the communication unit (1005).

[0143] The storage unit (1010) can store data such as basic programs, application programs, and setting information for the operation of the terminal. The storage unit (1010) may be composed of volatile memory, non-volatile memory, or a combination of volatile memory and non-volatile memory. Additionally, the storage unit (1010) can provide the stored data upon a request from the control unit (1015).

[0144] The control unit (1015) can control the overall operations of the terminal. For example, the control unit (1015) can transmit and receive signals through the communication unit (1005). Additionally, the control unit (1015) writes and reads data to and from the storage unit (1010). Furthermore, the control unit (1015) can perform the functions of the protocol stack required by the communication standard. To this end, the control unit (1015) may include at least one processor or microprocessor, or be part of a processor. Also, part of the communication unit (1005) and the control unit (1015) may be referred to as a communication processor (CP). According to various embodiments, the control unit (1015) can control the terminal to perform synchronization using a wireless communication network. For example, the control unit (1015) can control the terminal to perform operations according to various embodiments described below.

[0145] According to various embodiments of the present disclosure, a terminal may be composed of ME (mobile equipment) and USIM (Universal Mobile Telecommunications Service (UMTS) Subscriber Identity Module). The ME may include MT (mobile terminal) and TE (terminal equipment). The MT may be a part where a wireless access protocol operates, and the TE may be a part where a control function operates. For example, in the case of a wireless communication terminal (e.g., a mobile phone), the MT and TE may be integrated, and in the case of a laptop, the MT and TE may be separated. The present disclosure may describe the ME and USIM as distinct entities depending on the operation of each component, but is not limited thereto; it is understood that the various embodiments of the present disclosure may be described by including the ME and USIM as a terminal (e.g., UE) or by referring to the ME as a terminal.

[0146] FIG. 11 illustrates an example of the functional structure of a core network object according to one embodiment of the present disclosure. It illustrates the configuration of a core network object in a wireless communication system according to various embodiments of the present disclosure. The configuration exemplified in FIG. 11 can be understood as the configuration of a device having at least one function among network entities including the AMF of FIG. 5. Terms such as '...part', '...unit', etc. used below refer to a unit that processes at least one function or operation, and this may be implemented in hardware or software, or a combination of hardware and software. The network entity of FIG. 11 may correspond to the network entity illustrated in FIG. 4 through FIG. 7.

[0147] Referring to FIG. 11, the core network object is configured to include a communication unit (1105), a storage unit (1110), and a control unit (1115).

[0148] The communication unit (1105) can provide an interface for performing communication with other devices within the network. That is, the communication unit (1105) can convert a bit sequence transmitted from a core network object to another device into a physical signal, and convert a physical signal received from another device into a bit sequence. That is, the communication unit (1105) can transmit and receive signals. Accordingly, the communication unit (1105) may be referred to as a modem, a transmitter, a receiver, or a transceiver. At this time, the communication unit (1105) can enable the core network object to communicate with other devices or systems via a backhaul connection (e.g., wired backhaul or wireless backhaul) or via a network.

[0149] The storage unit (1110) can store data such as basic programs, application programs, and configuration information for the operation of a core network object. The storage unit (1110) may be composed of volatile memory, non-volatile memory, or a combination of volatile memory and non-volatile memory. Additionally, the storage unit (1110) can provide the stored data upon request from the control unit (1115).

[0150] The control unit (1115) can control the overall operations of the core network object. For example, the control unit (1115) can transmit and receive signals through the communication unit (1105). Additionally, the control unit (1115) writes and reads data to and from the storage unit (1110). To this end, the control unit (1115) may include at least one processor. According to various embodiments of the present disclosure, the control unit (1115) can control the core network object to perform synchronization using a wireless communication network. For example, the control unit (1115) can control the core network object to perform operations according to various embodiments described below.

[0151] Terms used in the foregoing description to identify connection nodes, terms referring to network entities, terms referring to messages, terms referring to interfaces between network entities, terms referring to various identification information, etc., are examples provided for the convenience of explanation. Accordingly, the present disclosure is not limited to the foregoing terms, and other terms referring to objects having equivalent technical meanings may be used.

[0152] The operations of the embodiments described above can be realized by providing a memory device storing program code in any component within the device. That is, the control unit within the device can execute the operations described above by reading and executing the program code stored in the memory device by a processor or a CPU (Central Processing Unit).

[0153] The entities or various components of terminal devices and modules described in this disclosure may be operated using hardware circuits, such as, for example, complementary metal oxide semiconductor-based logic circuits, firmware, software, and / or a combination of hardware and firmware and / or software embedded in a machine-readable medium. For example, various electrical structures and methods may be implemented using electrical circuits such as transistors, logic gates, and application-specific semiconductors.

[0154] Methods according to the claims or embodiments described in the specification of the present disclosure may be implemented in the form of hardware, software, or a combination of hardware and software.

[0155] When implemented in software, a computer-readable storage medium may be provided for storing one or more programs (software modules). One or more programs stored in the computer-readable storage medium are configured for execution by one or more processors within an electronic device. One or more programs include instructions that cause the electronic device to execute methods according to the claims or embodiments described in the specification of this disclosure.

[0156] These programs (software modules, software) may be stored in random access memory, non-volatile memory including flash memory, read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic disc storage device, compact disc-ROM (CD-ROM), digital versatile discs (DVDs), or other forms of optical storage devices, magnetic cassettes. Alternatively, they may be stored in a memory composed of some or all of these. Additionally, each constituent memory may include multiple units.

[0157] Additionally, the program may be stored on an attachable storage device that can be accessed via a communication network such as the Internet, Intranet, LAN (local area network), WAN (wide area network), or SAN (storage area network), or a combination thereof. Such a storage device may be connected to a device performing an embodiment of the present disclosure through an external port. Additionally, a separate storage device on a communication network may be connected to a device performing an embodiment of the present disclosure.

[0158] In the specific embodiments of the present disclosure described above, the components included in the disclosure are expressed in a singular or plural form according to the specific embodiments presented. However, the singular or plural expression is selected to suit the situation presented for convenience of explanation, and the present disclosure is not limited to singular or plural components; even if a component is expressed in the plural form, it may be composed of a singular form, and even if a component is expressed in the singular form, it may be composed of a plural form.

[0159] Meanwhile, although specific embodiments have been described in the detailed description of the present disclosure, it is understood that various modifications are possible within the scope of the present disclosure. Therefore, the scope of the present disclosure should not be limited to the described embodiments, but should be defined by the claims set forth below as well as equivalents thereof.

[0160] A method performed by a mobility management device according to one embodiment of the present disclosure may include: an operation of obtaining a first input sequence based on a past movement path of a terminal; an operation of obtaining a second input sequence based on a converted number or vector by converting a plurality of consecutive data points included in the first input sequence into a single number or vector; and an operation of performing training of a terminal location prediction model on the second input sequence.

[0161] According to one embodiment, the operation of converting to a single number includes the operation of multiplying each of the values ​​of the plurality of consecutive data points by a power of the number of candidate base stations, and the power of the number of candidate base stations multiplied by the value of any first data point among the plurality of consecutive data points and the power of the number of candidate base stations multiplied by the value of any second data point may be different from each other.

[0162] According to one embodiment, the operation of converting into a single vector includes: an operation of summing the vectors of the plurality of consecutive data points; and an operation of converting the summed single vector based on the time the terminal stayed at the candidate base station, and the time the terminal stayed at the candidate base station can be determined based on the first input sequence.

[0163] According to one embodiment, the operation of acquiring the second input sequence may further include an operation of performing dimensionality reduction based on the location of candidate base stations or the degree of proximity of candidate base stations with respect to the converted number or vector.

[0164] According to one embodiment, the operation of performing the learning may include an operation of grouping a plurality of consecutive GRUs included in the input GRU layer of the stacked GRU (gated recurrent unit) for the learning.

[0165] According to one embodiment, the stacked GRU further includes a hidden GRU layer, and the operation of performing the learning may further include an operation of grouping a plurality of consecutive GRUs included in the hidden GRU layer.

[0166] According to one embodiment, the method further includes an operation to generate the learning result output data, and the output data may indicate the time during which the terminal is predicted to stay at the candidate base station.

[0167] A mobility management device according to one embodiment of the present disclosure comprises: a transceiver; and a controller coupled to the transceiver, wherein the controller may be configured to: acquire a first input sequence based on the past movement path of a terminal, and acquire a second input sequence based on a converted number or vector by converting a plurality of consecutive data points included in the first input sequence into a single number or vector, and to perform learning of a terminal location prediction model on the second input sequence.

[0168] According to one embodiment, the control unit is configured to multiply each of the values ​​of the plurality of consecutive data points by a power of the number of candidate base stations, and the power of the number of candidate base stations multiplied by the value of any first data point among the plurality of consecutive data points and the power of the number of candidate base stations multiplied by the value of any second data point may be different from each other.

[0169] According to one embodiment, the control unit is configured to sum vectors of a plurality of consecutive data points and convert the summed vector based on the time the terminal stayed at the candidate base station, and the time the terminal stayed at the candidate base station can be determined based on the first input sequence.

[0170] According to one embodiment, the control unit may be configured to perform dimensionality reduction on the converted number or vector based on the location of candidate base stations or the degree of proximity of candidate base stations to each other.

[0171] According to one embodiment, the control unit may be configured to group a plurality of consecutive GRUs included in the input GRU layer of the stacked GRU (gated recurrent unit) for learning.

[0172] According to one embodiment, the stacked GRU further includes a hidden GRU layer, and the control unit may be configured to group a plurality of consecutive GRUs included in the hidden GRU layer.

[0173] According to one embodiment, the control unit is further configured to generate the learning result output data, and the output data may indicate the time during which the terminal is predicted to stay at the candidate base station.

Claims

1. In a method performed by a mobility management device, An operation to obtain a first input sequence based on the past movement path of a terminal; An operation of obtaining a second input sequence based on a converted number or vector by converting a plurality of consecutive data points included in the first input sequence into a single number or vector; and A method comprising the operation of performing training of a terminal location prediction model for the above second input sequence.

2. In Paragraph 1, The operation of converting to a single number includes the operation of multiplying each of the values ​​of the plurality of consecutive data points by a power of the number of candidate base stations, and A method in which the power of the number of candidate base stations multiplied by the value of any first data point among the plurality of consecutive data points and the power of the number of candidate base stations multiplied by the value of any second data point are different.

3. In Paragraph 1, The operation of converting to a single vector above comprises: an operation of summing the vectors of the plurality of consecutive data points; and The operation includes converting the above-mentioned summed vector based on the time the terminal stayed at the candidate base station, and A method in which the time the terminal stays at the candidate base station is determined based on the first input sequence.

4. In Paragraph 1, The operation of acquiring the second input sequence further includes the operation of performing dimensionality reduction based on the location of candidate base stations or the degree of proximity of candidate base stations to each other with respect to the converted number or vector.

5. In Paragraph 1, A method comprising an operation to perform the above learning, wherein the operation includes grouping a plurality of consecutive GRUs included in the input GRU layer of the stacked GRU (gated recurrent unit) for the above learning.

6. In Paragraph 5, The above stacked GRU further includes a hidden GRU layer, and A method in which the operation of performing the above learning further includes an operation of grouping a plurality of consecutive GRUs included in the hidden GRU layer.

7. In Paragraph 1, It further includes an operation to generate the above-mentioned learning result output data, and A method in which the above output data indicates the time during which the terminal is predicted to stay at the candidate base station.

8. In a mobility management device, a transceiver; and It includes a controller coupled to the above-mentioned transmitting and receiving unit, and the controller comprises: Obtain a first input sequence based on the past movement path of the terminal, and By converting a plurality of consecutive data points included in the first input sequence into a single number or vector, a second input sequence based on the converted number or vector is obtained, and A mobility management device configured to perform learning of a terminal location prediction model for the above second input sequence.

9. In Paragraph 8, The above control unit is configured to multiply each of the values ​​of the plurality of consecutive data points by a power of the number of candidate base stations, and A mobility management device in which the power of the number of candidate base stations multiplied by the value of any first data point among the plurality of consecutive data points and the power of the number of candidate base stations multiplied by the value of any second data point are different.

10. In Paragraph 8, The control unit above sums the vectors of the plurality of consecutive data points, and The above-mentioned summed vector is configured to be converted based on the time the terminal stayed at the candidate base station, and A mobility management device in which the time the terminal stays at the candidate base station is determined based on the first input sequence.

11. In Paragraph 8, A mobility management device configured such that the control unit performs dimensionality reduction based on the location of candidate base stations or the degree of proximity of candidate base stations to each other for the converted number or vector.

12. In Paragraph 8, The above control unit is a mobility management device configured to group a plurality of consecutive GRUs included in the input GRU layer of the stacked GRU (gated recurrent unit) for learning.

13. In Paragraph 12, The above stacked GRU further includes a hidden GRU layer, and The above control unit is a mobility management device configured to group a plurality of consecutive GRUs included in the above hidden GRU layer.

14. In Paragraph 8, The above control unit is further configured to generate the learning result output data, and A mobility management device in which the above output data indicates the time the terminal is predicted to stay at the candidate base station.

Images