Terminal track generation method and device, medium and product

Abstract

The embodiment of the invention provides a terminal track generation method and device, a medium and a product, and relates to the technical field of mobile communication. The method comprises the following steps: determining mapping data between signaling and a position of a terminal; acquiring a time feature and a position feature in the mapping data by using a space-time encoder; and inputting the time feature and the position feature into a preset attention mechanism model, and determining the actual position of the terminal. According to the scheme provided by the invention, the dependence on a single base station signal is reduced through deep association of the terminal signaling and the position mapping data, and the dependence on the positioning precision of the base station is broken through; a space-time encoder and an attention mechanism model are utilized to realize multi-modal mobile modeling; based on signaling and an attention mechanism, semantic and base station interaction relation modeling is enhanced, and the precision of track production or track recovery is improved.

Description

Technical Field

[0001] The present application relates to the field of mobile communication technology, and in particular to a terminal trajectory generation method, device, medium, and product. Background Art

[0002] With the development of mobile communication technology, terminal trajectory restoration technology based on base station signaling data plays an important role in urban traffic planning, travel behavior analysis and other fields. Existing technologies mainly improve trajectory accuracy through signaling data clustering, road network matching and rule correction, but still face the following bottlenecks: (1) Inherent limitations of base station positioning accuracy: Traditional methods such as density clustering algorithm clustering and stop point elimination in existing technologies can suppress the ping-pong effect, but rely on the geometric division of base station coverage. Due to the uneven distribution of base station density, rule-based correction is difficult to break through the physical positioning error, resulting in trajectory breakage or drift in complex urban areas or areas with sparse base stations. (2) Lack of processing of multimodal mobility patterns: A related technology extracts a single travel path through hierarchical clustering and path similarity calculation, but lacks modeling capabilities for mixed transportation modes, such as walking-subway-shared bicycle combination travel. The spatiotemporal characteristics of different transportation tools are significantly different, and traditional clustering algorithms cannot adaptively distinguish the dynamic switching of mobility modes, resulting in the accumulation of path fusion errors. (3) Insufficient utilization of semantic information and network topology: Although another related technology introduced road network matching, it did not deeply explore the topological relationship between base stations and the surrounding semantic environment. For example, the terminal migration patterns between base stations in dense commercial areas are significantly different from those in industrial areas, and the other related technology did not incorporate this information into the trajectory inference model.

[0003] Therefore, the existing technology has the problems of relying on base station positioning with insufficient accuracy, lacking multi-mode mobility modeling, and ignoring the interaction between semantics and base stations, resulting in limited trajectory restoration effect. Summary of the Invention

[0004] At least one embodiment of the present application provides a terminal trajectory generation method, apparatus, medium, and product for resolving the problems in the prior art of relying on base station positioning with insufficient accuracy, lacking multi-modal mobility modeling, and ignoring semantic interactions with base stations, resulting in limited trajectory restoration effectiveness.

[0005] In order to solve the above technical problems, this application is implemented as follows: In a first aspect, an embodiment of the present application provides a terminal trajectory generation method, comprising: Determining mapping data between signaling and a location of a terminal, including: obtaining communication signaling data and real trajectory data of the terminal; performing time sorting and outlier removal operations on the communication signaling data to generate a signaling sequence; determining, based on the signaling sequence and the actual location in the real trajectory data, the mapping data having a mapping relationship between the signaling and the location in a time alignment manner, and filtering out the outlier data; Using a spatiotemporal encoder, obtaining time features and position features in the mapping data; The time feature and the location feature are input into a preset attention mechanism model to determine the actual location of the terminal.

[0006] Optionally, using a spatiotemporal encoder to obtain temporal features in the mapping data includes: Normalizing the timestamp information in the mapping data using the time branch in the spatiotemporal encoder to determine an entry time parameter, an exit time parameter, and a duration parameter; For the entry time parameter, the exit time parameter, and the duration parameter, respectively, a corresponding time feature vector is generated through a trainable linear layer; Use the cosine function to capture the periodicity of the time component and determine the periodic characteristics; The corresponding time feature vector and the periodic feature are subjected to feature fusion to obtain the time feature in the mapping data.

[0007] Optionally, using a spatiotemporal encoder to obtain position features in the mapping data includes: Using a multi-layer perceptron in the spatiotemporal encoder, obtaining base station positions and relative positional relationships between different base stations in the mapping data; The base station position and the relative position relationship are comprehensively coded and represented to obtain position features in the mapping data.

[0008] Optionally, before inputting the time feature and the location feature into a preset attention mechanism model to determine the actual location of the terminal, the method further includes: Obtaining base station adjacency relationships, and determining a global spatiotemporal representation based on the base station adjacency relationships; For the movement trajectory of a single terminal, a local transfer subgraph is constructed, and a hierarchical attention mechanism is used based on the local transfer subgraph to extract personalized movement pattern features; Extract road network topology features and point of interest distribution features within the preset range of the base station; Generating a geographic semantic representation through a spatial attention network according to the road network topology characteristics and the distribution characteristics of the points of interest; Performing gated fusion on the global spatiotemporal representation, the personalized mobility pattern features, and the geographic semantic representation to output comprehensive base station information containing multi-dimensional interactive relationships; The initial attention mechanism model is trained using the comprehensive base station information to generate the attention mechanism model.

[0009] Optionally, obtaining a base station adjacency relationship and determining a global spatiotemporal representation based on the base station adjacency relationship includes: Obtain the spatial locations of all base stations in the target area, as well as the terminal transfer patterns between different base stations; Determine the adjacency relationship of the base stations by using the spatial positions of the base stations as nodes and the terminal transfer patterns as edges; constructing a global transition graph of base stations based on the base station adjacency relationship; A global spatiotemporal representation is determined based on the global transition graph.

[0010] Optionally, using the comprehensive base station information to train an initial attention mechanism model to generate the attention mechanism model includes: Obtaining sample time features and sample location features generated based on the sample; Generate a multi-time-step interactive representation using a multi-head self-attention mechanism according to the comprehensive base station information, the sample time features, and the sample position features; Constructing a multi-task loss function; the multi-task loss function includes a geographic coordinate reconstruction error term, a positive sample representation similarity term, and a negative sample comparison term, and optimizing model parameters through self-supervised learning; The attention mechanism model is constructed based on the interaction representation and the multi-task loss function.

[0011] In a second aspect, an embodiment of the present application provides a terminal trajectory generation device, including: A first determination module is configured to determine mapping data between the signaling and the location of the terminal; wherein the first determination module includes: a first acquisition unit configured to acquire communication signaling data and real trajectory data of the terminal; a first processing unit configured to perform time sorting and outlier removal operations on the communication signaling data to generate a signaling sequence; and a second processing unit configured to determine, based on the signaling sequence and the actual location in the real trajectory data, the mapping data having a mapping relationship between the signaling and the location in a time-aligned manner, and filter out outlier data. A first processing module is configured to obtain time features and position features in the mapping data using a spatiotemporal encoder; The second processing module is used to input the time feature and the location feature into a preset attention mechanism model to determine the actual location of the terminal.

[0012] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0013] In a fourth aspect, an embodiment of the present application provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0014] Compared with the prior art, the terminal trajectory generation method, device, medium and product provided in the embodiments of the present application combine the signaling and position of the terminal to generate mapping data between the two, extract the time and space features in the mapping data through the space-time encoder, and reduce the single dependence on the base station signal. Compared with the traditional base station air interface data dependence, this method uses the road network topology constraint to optimize the trajectory, reduce the error accumulation caused by uneven base station coverage, and improve the positioning accuracy in complex environments; use the space-time encoder to obtain the time features and position features in the mapping data, and select a time-segment dynamic model to fuse with the space-time features to solve the problem that a single model in the traditional method cannot adapt to complex mobile states; then use the attention mechanism to input the time features and the position features into a preset attention mechanism model to determine the actual position of the terminal and improve the semantic relevance of the trajectory restoration. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present application. The same reference symbols are used throughout the drawings to represent the same components. In the drawings: Figure 1 A schematic diagram of a process for generating a terminal trajectory according to an embodiment of the present application; Figure 2 A schematic diagram of the model training process provided in the embodiment of the present application; Figure 3 This is a schematic diagram of the structure of the terminal trajectory generation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The terms "first", "second", etc. in this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same type, and do not limit the number of objects, for example, the first object can be one or more. In addition, "or" in this application represents at least one of the connected objects. For example, "A or B" covers three options, namely, Option 1: including A but not including B; Option 2: including B but not including A; Option 3: including both A and B. The character " / " generally indicates that the objects associated before and after are in an "or" relationship.

[0017] The term "indication" in this application can be either a direct indication (or explicit indication) or an indirect indication (or implicit indication). A direct indication can be understood as the sender explicitly informing the receiver of specific information, the required operation, or the requested result, etc. in the instruction sent; an indirect indication can be understood as the receiver determining the corresponding information based on the instruction sent by the sender, or making a judgment and determining the required operation or request result based on the judgment result.

[0018] As described in the background technology, existing technologies for improving trajectory longitude often only process signaling data through rule-based methods, which cannot break through the positioning accuracy based on base station location, and the accuracy of base station positioning itself is low; existing technologies do not fully consider complex terminal mobility patterns and are not universally applicable. They can often only be used to process a certain type of travel mode, but cannot uniformly process data for multiple travel modes; existing technologies only use signaling data itself and additional information (such as road network information) to improve data quality, and often ignore semantic information around the base station that helps improve trajectory accuracy and related information such as the interaction relationship between base stations. To solve at least one of the above problems, the embodiments of the present application provide a terminal trajectory generation method that can reduce or avoid the occurrence of the above situation.

[0019] Please refer to Figure 1 , an embodiment of the present application provides a terminal trajectory generation method, comprising: Step 11: Determine mapping data between the signaling and location of the terminal.

[0020] In the embodiment of the present application, the core of this step is to establish an association between the interactive signaling between the terminal and the base station (such as access events, switching records) and the geographic location information of the base station to form a mapping relationship. Optionally, the communication event timestamp of the terminal, the base station identifier (such as the cell ID), the terminal unique identifier and other data are obtained to collect the signaling of the terminal. Combined with the base station ledger or geographic location data, the base station identifier is mapped to specific geographic coordinates (latitude and longitude) and road network information. The signaling data is associated with the road network information through the terminal identifier to generate an original data set containing time, base station location, and terminal movement path.

[0021] For example, suppose a terminal connects to base station A at time t1 (at location (x1, y1)) and then switches to base station B at time t2 (at location (x2, y2)). By associating the event time order in the signaling data with the base station locations, we can generate the initial trajectory point sequence: [(t1, x1, y1), (t2, x2, y2), ...].

[0022] Optionally, determining mapping data between signaling and location of the terminal includes: Obtain the terminal's communication signaling data and real trajectory data; Performing time sorting and outlier removal operations on the communication signaling data to generate a signaling sequence; Based on the signaling sequence and the actual position in the real trajectory data, the mapping data having a mapping relationship between the signaling and the position is determined in a time alignment manner, and abnormal data is filtered out.

[0023] In the embodiments of the present application, communication signaling data includes event records of the interaction between the terminal and the base station (such as access, switching, disconnection, etc.), including but not limited to timestamps, base station identifiers (such as cell ID, LAC), terminal identifiers (such as IMSI), and other information; real trajectory data can be obtained through high-precision positioning technology (such as GPS, Beidou) or user-reported trajectory points, including timestamps and geographic coordinates (latitude and longitude). The purpose of obtaining the communication signaling data and real trajectory data of the terminal is to establish an association between signaling events and physical locations, providing basic data for subsequent feature extraction and anomaly filtering.

[0024] For example, suppose a terminal interacts with base stations at the following times: t1 = 10:00:00, connecting to base station A (LAC = 12345, Cell = 67890, coordinates (116.4, 39.9)); t2 = 10:05:00, switching to base station B (LAC = 23456, Cell = 89012, coordinates (116.5, 39.8)). The actual trajectory data recorded via GPS is: t1 = 10:00:00, location (116.41, 39.92); t2 = 10:05:00, location (116.52, 39.78). The two timestamps are merged to form the initial dataset.

[0025] Furthermore, the communication signaling data is time-sorted and outlier removal is performed. Time sorting, for example, sorts signaling times by timestamp to ensure temporal continuity. Outlier removal includes, but is not limited to, removing records with duplicate timestamps or timestamps with small intervals; filtering out points where base station coordinates do not match the road network; and removing outliers using clustering or rate thresholds.

[0026] In this application, the signaling time is matched with the actual trajectory point according to the time window, and time alignment is used to generate candidate association pairs. The determination of the mapping relationship includes direct matching and difference matching. Direct matching is to directly establish the mapping if the signaling time coincides with the trajectory point time; difference matching is to estimate the signaling position by the difference of the trajectory point if there is no exact match, and then eliminate association pairs where the signaling and trajectory points are too far apart or have inconsistent directions.

[0027] Multi-dimensional verification can be used to filter out abnormal mapping data, including: checking whether the mapping location is within the road network to implement road network constraints; calculating the position and direction changes between adjacent mapping points to eliminate sudden changes (such as 90° sharp turns); and combining base station signal strength (RSRP) or MR (Measurement Report) data to filter low-quality signaling.

[0028] The steps of this application combine multi-dimensional information such as signaling, trajectory, and road network to achieve multi-source data fusion and improve mapping accuracy. Through time sorting, cluster analysis, and road network constraints, dynamic anomaly filtering is achieved, effectively suppressing signal drift. Interpolation matching and motion continuity verification reduce time window errors and achieve high-precision alignment, suitable for high-speed mobile scenarios. Through refined data cleaning and spatiotemporal alignment mechanisms, the reliability of signaling and position mapping is significantly improved, providing high-quality input for subsequent trajectory generation.

[0029] In a specific implementation, step 11 of the present application can be understood as a data preprocessing step. The communication signaling data and the actual travel trajectory data of the same batch of terminals are collected, which can be obtained through mobile phone operators and related applications (APPs). A mapping pair from the actual signaling trajectory to the actual travel location is formed, and the sample pairs therein are screened and abnormal data is eliminated. The communication signaling data of the terminal on the mobile device is collected, the abnormal data therein is removed, and the remaining data is arranged in chronological order to form the original signaling sequence data. The actual location data of the same terminal within the time period is collected and arranged according to time, and according to the time period corresponding to the terminal, a mapping from the communication signaling sequence to the actual trajectory sequence of the terminal is constructed, and the sample pairs are divided into a structured format according to the form of the time period, and the abnormal sample pairs are eliminated. In this way, a mapping from the communication signaling sequence to the actual trajectory sequence of the terminal is constructed.

[0030] Step 12: Using a spatiotemporal encoder, obtain the time features and position features in the mapping data.

[0031] In this application, the spatiotemporal encoder converts the time and space information in the mapping data into a high-dimensional feature vector to capture the spatiotemporal dependency: a sine / cosine function or a learnable position encoding (such as position embedding in Transformer) is used to map the timestamp to a periodic feature; optionally, the base station coordinates are converted into spatial semantic features using Beidou grid codes, geographic hashing, or Euclidean distance encoding.

[0032] Furthermore, time and position encoding can be interacted through convolutional layers or Transformer to generate joint spatiotemporal features.

[0033] Step 13: Input the time feature and the location feature into a preset attention mechanism model to determine the actual location of the terminal.

[0034] It should be noted that the attention mechanism model is used to dynamically weight key spatiotemporal features, suppressing noise and enhancing effective information, ultimately outputting a high-precision trajectory. The attention layers in the attention mechanism model include local attention and global attention. Local attention is used to calculate the association weights (such as cosine similarity) between the current trajectory point and its neighboring points. Global attention is used to incorporate prior knowledge such as road direction and traffic flow to adjust the weight distribution. Here, assuming the model detects that the terminal frequently switches base stations near a highway (for example, from base station C to base station D to base station E), the attention mechanism will enhance the weight of the road direction feature and suppress abnormal signals from remote base stations. The final output trajectory point (x, y) has an error within a preset range from the actual road position.

[0035] In this embodiment, a pre-trained attention mechanism model is used to input time and location features, and the actual location of the terminal can be obtained as the model output. By integrating communication signaling, spatiotemporal coding, and the attention mechanism model, this application improves the accuracy and robustness of trajectory generation, making it suitable for positioning needs in dynamic and complex environments.

[0036] Optionally, the step 12 of obtaining the time feature in the mapping data includes: Normalizing the timestamp information in the mapping data using the time branch in the spatiotemporal encoder to determine an entry time parameter, an exit time parameter, and a duration parameter; For the entry time parameter, the exit time parameter, and the duration parameter, respectively, a corresponding time feature vector is generated through a trainable linear layer; Use the cosine function to capture the periodicity of the time component and determine the periodic characteristics; The corresponding time feature vector and the periodic feature are subjected to feature fusion to obtain the time feature in the mapping data.

[0037] In this embodiment of the present application, the time branch in the space-time encoder is used to convert the original timestamps in the mapping data into interpretable time series parameters, eliminate dimension differences, and normalize the timestamp information in the mapping data to determine the entry time parameter, departure time parameter, and duration parameter. The entry time parameter is used to indicate the time when the terminal enters the base station coverage area; the departure time parameter is used to indicate the time when the terminal leaves the base station coverage area; and the duration parameter is used to indicate the length of time the terminal stays in the base station coverage area.

[0038] Through a trainable linear layer, such as a fully connected network, the three types of time parameters mentioned above—entry time, exit time, and duration—are mapped into high-dimensional vectors, capturing the latent semantic information of these time parameters. The linear layer weights are optimized through gradient descent to adapt to the temporal representation requirements of different scenarios.

[0039] The cosine function is introduced to periodically encode the time component, for example, converting timestamps into periodic features. Periodic features can reflect the repetitive patterns of terminal behavior. The time feature vector and the periodic features are concatenated or weighted to generate a composite time feature. For example, the attention mechanism can dynamically assign weights to the two features. The fusion process can suppress noise and improve the model's generalization ability for complex scenarios.

[0040] This step extracts multidimensional parameters (entry / exit / duration). The cosine function encodes periodic components, and a trainable linear layer plus fusion mechanism enhances nonlinear representation. Through parameter refinement, periodic modeling, and dynamic fusion, the representation capability of temporal features is systematically improved, providing reliable input for downstream tasks such as trajectory prediction and anomaly detection. This method improves the accuracy and robustness of spatiotemporal modeling by refining temporal feature extraction.

[0041] Optionally, the step 12 of obtaining the location features in the mapping data includes: Using a multi-layer perceptron in the spatiotemporal encoder, obtaining base station positions and relative positional relationships between different base stations in the mapping data; The base station position and the relative position relationship are comprehensively coded and represented to obtain position features in the mapping data.

[0042] In this embodiment, a multi-layer perceptron (MLP) in a spatiotemporal encoder is used to extract base station location features. The specific process is as follows: The MLP performs nonlinear mapping on the latitude and longitude coordinates of the base stations, converting the absolute physical location into a high-dimensional vector, enhancing the model's semantic understanding of geographic coordinates. The activation function (such as ReLU) in the MLP captures base station density distribution characteristics, such as the differential representation of densely populated base stations in urban areas and sparsely populated base stations in suburban areas. Parameters such as distance and azimuth between base stations are calculated to construct a spatial topological network between base stations and characterize the interactive relationships between base station coverage areas. The MLP performs normalization and weighting on the relative position parameters to eliminate dimensional differences and highlight the influence of key neighboring base stations. The absolute position vector and the relative position weight matrix are concatenated or attention-weighted to generate a joint spatial feature. For example, a gating mechanism is used to dynamically adjust the contribution ratio of the two feature types. Residual connections or batch normalization are introduced into the fusion process to avoid the vanishing gradient problem and improve the model's tolerance to base station signal drift. The fusion process of this application introduces residual connections or batch normalization to avoid the vanishing gradient problem and improve the model's tolerance to base station signal drift.

[0043] Specifically, the longitude and latitude positions in the signaling sequence and the spatial relationship between the front and rear base stations are represented in the following encoding method: , formula (1); , formula (2); in, Respectively represent the longitude and latitude of the base station and its relative position to the previous and next base stations; The feature representation of longitude and latitude information is used to obtain the contextual features before and after the base station and the base station's location. MLP is a multi-layer perceptron model. PosEncoder represents a comprehensive encoding representation of longitude and latitude information and relative position, which can be processed using either an MLP or a custom neural network.

[0044] To represent the timestamp information in the signaling sequence, use As a representation of timestamp comprehensive information, the specific representation method is: , formula (3); , formula (4); , formula (5); , formula (6); in, The length of time the terminal is within the base station's detection range, the time it enters, and the time it leaves, respectively. This encoding captures the temporal characteristics of the terminal's presence within the base station's detection range, capturing relevant features of the base station's coverage area and the periodic nature of the terminal's behavior. Furthermore, by introducing time difference information and processing specific timestamps, this method can handle time intervals of varying lengths, making it more general. Both are trainable parameters. Cos refers to the trigonometric function, which can be used to learn the periodic features in the representation and the duration features. Stitching with periodic features.

[0045] Optionally, before step 13 of the present application, the method further includes a step of generating the attention mechanism model, which includes: Obtaining base station adjacency relationships, and determining a global spatiotemporal representation based on the base station adjacency relationships; For the movement trajectory of a single terminal, a local transfer subgraph is constructed, and a hierarchical attention mechanism is used based on the local transfer subgraph to extract personalized movement pattern features; Extract road network topology features and point of interest distribution features within the preset range of the base station; Generating a geographic semantic representation through a spatial attention network according to the road network topology characteristics and the distribution characteristics of the points of interest; Performing gated fusion on the global spatiotemporal representation, the personalized mobility pattern features, and the geographic semantic representation to output comprehensive base station information containing multi-dimensional interactive relationships; The initial attention mechanism model is trained using the comprehensive base station information to generate the attention mechanism model.

[0046] In an embodiment of the present application, base station adjacency relationships are obtained. Base station adjacency relationships can be used to analyze the relationship between the spatial distribution of base station groups and the transfer patterns of terminals (including the transfer frequencies of terminals), such as confirming the relationship between the location of a base station and the transfer of a terminal. Determining a global spatiotemporal representation based on base station adjacency relationships solves the problem of traditional methods being overly dependent on local base stations, supports dynamic scheduling of cross-regional communication resources, and supports dynamic scheduling of cross-regional communication resources. The construction of a local transfer subgraph can model the base station switching sequence as a directed graph structure for the movement trajectory of a single terminal, capturing individual mobility preferences.

[0047] The system calculates the weights of inter-base station transition probabilities, focusing on high-frequency handoff paths and calculating spatial attention. It also dynamically adjusts path weights based on dwell time, distinguishing between commuting and random movement patterns and calculating temporal attention. This helps avoid individual trajectory ambiguity caused by overlapping base station signals and improves trajectory prediction accuracy.

[0048] Based on parameters such as road node and segment connectivity within the base station's preset range, a graph structure is constructed to represent regional traffic capacity, such as the connectivity differences between main roads and branch roads, enabling road network topology extraction. Points of interest (POI) density information for commercial districts, transportation hubs, and other locations is integrated to map regional functional attributes. For example, a high POI density in a commercial district corresponds to a high terminal concentration, enabling POI distribution analysis. A multi-head attention mechanism dynamically assigns weights to road topology and POI features, enabling the design of a spatial attention network.

[0049] The global spatiotemporal representation, personalized features, and geographic semantic representation are input, and the contribution ratio of each feature is dynamically adjusted through a trainable gating unit, such as a Sigmoid function, to train the initial attention mechanism model and generate the attention mechanism model.

[0050] For example, in areas with sparse base station coverage, the weights of geographic semantic representations are enhanced to compensate for signal loss. Model training strategy: A multi-task loss function is used to jointly optimize objectives such as base station load prediction and terminal trajectory restoration. An adaptive learning rate adjustment strategy is introduced to balance the gradient update speed of different feature dimensions.

[0051] The method of this application achieves refined representation of base station information through joint modeling of global and local features, such as base station cluster distribution and individual trajectories, multimodal geographic semantic fusion, and dynamic gating mechanism, providing high-precision model support for scenarios such as communication network optimization and user behavior analysis.

[0052] Furthermore, obtaining a base station adjacency relationship and determining a global spatiotemporal representation based on the base station adjacency relationship include: Obtain the spatial locations of all base stations in the target area, as well as the terminal transfer patterns between different base stations; Determine the adjacency relationship of the base stations by using the spatial positions of the base stations as nodes and the terminal transfer patterns as edges; constructing a global transition graph of base stations based on the base station adjacency relationship; A global spatiotemporal representation is determined based on the global transition graph.

[0053] In this application, the base station adjacency relationship is constructed to capture the base station transition pattern and spatial correlation. Based on the spatial location of all base stations in the study area, as well as the transfer pattern and frequency of the terminal as a whole between different base stations, the base station adjacency relationship is constructed, and a global transfer map of the base stations is constructed. , where V is the set of base station nodes, E is the weighted edge defined by the transfer pattern and frequency, which captures the global spatiotemporal characteristics of the base station, thereby further enriching the representation of the base station. , record its adjacent node set as ; is the original feature of the node. Then, for each node and its adjacent edge in the model, the representation of each node is as follows: , formula (7); , formula (8); , formula (9); , formula (10); , formula (11); , formula (12); Among them, formula (7) is used to express the initial representation of the node; A trainable weight matrix used to represent the global node feature transformation, mapping the original node features fi to a high-dimensional space; Represents the original characteristics of base station node i (such as coverage, traffic volume, geographic location code, etc.); Used to represent the set of neighboring nodes of base station node i (the adjacency relationship is defined based on the transfer frequency); Represents the attention coefficient vector, which is used to weight the adjacent edge features ; represents the edge feature (output of formula (8)), which represents the association strength from base station i to j. Represents the trainable weight matrix and bias term of the global edge feature; in formula (8) Represents the original edge features (such as transfer frequency, time interval, signal strength difference, etc.). Formula (9) is used to normalize the attention coefficient; Used to represent the normalized attention weight, reflecting the adjacent edges The contribution ratio to node i. Formula (10) represents the dynamic edge feature update, where aglobal‌ represents the global attention parameter vector, which is used to dynamically adjust the edge weight; and Represent the concatenated feature vectors of nodes i and j respectively (the symbol ∥ indicates the concatenation operation); It is used to represent the activation function with leakage and enhance the nonlinear expression ability. Formula (11) represents the updated attention coefficient. Indicates that the edge feature is updated based on The recalculated attention weights. Formula (12) is used to update the node representation. σ(⋅) represents the activation function (such as Sigmoid or ReLU), which enhances nonlinearity. Represents the updated global node representation, which is the high-dimensional feature after integrating neighborhood information.

[0054] Through the above parameter design, the model achieves global-local feature complementarity (group patterns and individual preferences) and multimodal information fusion (communication data plus geographic semantics), providing a high-resolution representation basis for base station optimization and trajectory prediction.

[0055] The updated node features here are are the original trainable parameters, and , are all attention coefficients, is the original edge feature, is the updated edge feature. In this way, the node feature further considers and integrates edge representation on the basis of the original Graph Attention Network (GAT) model, so that edge features are also considered in the global representation.

[0056] According to the change of base station location in the terminal's own travel mode, a local transfer graph is constructed, and the local transfer features used are obtained through the GAT model. Specifically, for each terminal u, the local transfer graph constructed ,in For terminal The base stations in the trajectory, It is the transfer relationship in its moving trajectory. Represents the local representation, then for each terminal u the local representation , the processing method used here is similar to the above content, the formula is: , formula (13); , formula (14); , formula (15); , formula (16); , formula (17); , formula (18).

[0057] Among them, the local transfer map ( ) In the related formulas, formula (13) represents the initial representation of the local node. A trainable weight matrix representing the transformation of local node features (sharing structure with the global one but independent parameters). represents the personalized features of base station i in the trajectory of terminal u (such as residence time, trajectory sequence, etc.). Formula (14) is used to generate local edge features. represents local edge features (e.g., the frequency and timestamp interval of terminal u's transitions from base station i to j). In formulas (15) to (18), the meaning of the parameters is consistent with the global formulas (9) to (12), but the scope is limited to the local transition graph of terminal u. alocal represents the local attention parameter vector, which captures individual movement preferences. Represents the local node representation of terminal u, reflecting its personalized mobility pattern.

[0058] here All A collection of .

[0059] Here use As a comprehensive characterization of the base station, use Represents additional features such as road network information, POI information, etc. .in, represents the global spatiotemporal representation (output of formula (12)), encoding the overall transfer law of the base station group; represents the local personalized representation (output of formula (18)), capturing the individual behavior pattern of the terminal; Represents additional environmental features (such as road network topology, POI distribution), which are encoded into vectors through neural networks: such as road network topology features: road connectivity, traffic capacity, etc.; POI distribution features: density of points of interest such as commercial areas and transportation hubs. Feature fusion methods can use splicing plus fully connected layers, such as the one in this application. Gated weighting can be used, such as introducing adaptive weights to distribute the contribution ratio of each feature.

[0060] here This shows that a comprehensive representation is obtained by fusing multiple features, which includes the interaction between base stations and the interaction between base stations and the context environment. It represents the global representation of the base station. By vectorizing environmental features such as road network structure and POI, their comprehensive representation can be obtained through neural networks.

[0061] For the road network structure, all roads are represented by vectors, and for the types of POIs, they are represented by one-hot encoding. For each base station, the number of roads within a certain range of the base station, the distance between the roads, the number and type of POIs around the base station and other related information are considered and counted, and a network based on the attention mechanism is designed to fully consider the importance and relevance of geographic semantic information, thereby obtaining .

[0062] Optionally, using the comprehensive base station information to train an initial attention mechanism model to generate the attention mechanism model includes: Obtaining sample time features and sample location features generated based on the sample; Generate a multi-time-step interactive representation using a multi-head self-attention mechanism according to the comprehensive base station information, the sample time features, and the sample position features; Constructing a multi-task loss function; the multi-task loss function includes a geographic coordinate reconstruction error term, a positive sample representation similarity term, and a negative sample comparison term, and optimizing model parameters through self-supervised learning; The attention mechanism model is constructed based on the interaction representation and the multi-task loss function.

[0063] In the embodiment of the present application, the temporal features (such as dwell time and transfer interval) and spatial features (such as base station latitude and longitude, POI distribution) of the samples are extracted from the terminal signaling data. The input feature matrix is ​​constructed by combining comprehensive information such as base station coverage and traffic load, and the original signaling data is converted into structured features to capture the spatiotemporal regularity of user mobility behavior. A multi-head self-attention mechanism is used to dynamically weight the spatiotemporal features of the base station. Each attention head independently learns the spatiotemporal correlation patterns of different dimensions (such as time dependence, spatial proximity, and semantic similarity), and finally the multi-head outputs are spliced ​​to obtain a comprehensive interactive representation. The input features are linearly mapped into query, key, and value vectors; through attention weight calculation, the focus is adaptively focused on key time steps and base station nodes.

[0064] Constructing a multi-task loss function involves designing a geo-reconstruction error term, a positive sample representation similarity term, and a negative sample comparison term. The total loss function is then constructed based on these terms. The geo-reconstruction error term minimizes the root mean square error (RMSE) between the predicted and actual locations, ensuring that the latitude and longitude output by the model are consistent with the true trajectory. The positive sample representation similarity term maximizes the representation similarity (e.g., cosine similarity) of the true trajectory of the same user, improving the model's ability to capture user behavior patterns. The negative sample comparison term introduces adversarial negative examples (e.g., randomly perturbed trajectories or trajectories of other users) to distinguish between positive and negative examples using a contrastive loss, enhancing model robustness.

[0065] Based on the interaction representation and multi-task loss function, the constructed attention mechanism model can jointly optimize the multi-task loss using the backpropagation algorithm and update the model weights through gradient descent. The self-supervision mechanism reduces the reliance on manually labeled data and automatically learns representations by leveraging the spatiotemporal correlations between base stations. The output of the constructed attention mechanism model integrates the attention weight matrix of the base station's global context (such as network topology) and local behavior patterns (such as user movement preferences), supporting downstream tasks such as trajectory generation and base station coverage correction.

[0066] This application uses a multi-head mechanism and differential attention to suppress noise interference from irrelevant base station nodes and enhance the weight of key spatiotemporal features. It also combines base station attributes (coverage), environmental semantics (road network, POI), and user behavior (transfer frequency) to enhance the physical interpretability of representations. This application achieves joint optimization of geographic coordinate regression and representation comparison tasks, avoiding information loss from phased training.

[0067] Specifically, the steps to generate the attention mechanism model include: Use TransformerEncoder based on self-attention mechanism to perform comprehensive spatiotemporal representation of multimodal representation ,in, represents the sample time characteristics, represents the sample position feature, It is a comprehensive base station representation determined after processing the sample time features and sample location features generated based on the samples. It contains information of multiple time steps. By using the attention mechanism to assign different weights to different time steps, the information of each time step is fully considered in the generation of each position of the actual trajectory.

[0068] By using actual location information as labels, combining a self-supervised learning module with a mean square error loss function, and designing a custom loss function, the model is trained to achieve the task of reconstructing the signaling trajectory sequence to the precise location. In the process of designing the custom loss function, the weight of each of the above components is controlled by coefficients. The specific formula is as follows: , formula (19); Among them, formula (19) is the total loss function; is a hyperparameter used to adjust the weight of the contrast loss term and balance the importance of the position regression task and the representation learning task. The training objective of formula (19) is: Minimize Used to improve the accuracy of longitude and latitude prediction. ‌Maximize − , which is used to enhance the model’s ability to distinguish between real trajectories and abnormal trajectories. is the calculation based on longitude and latitude, expressed as root mean square error loss; and It is calculated by self-supervised learning and is expressed as positive sample contrast loss and negative sample contrast loss respectively. is the loss function of the terminal's existing movement trajectory, and It is the loss function of the representation obtained by randomly sampling random nodes of the same length, and is calculated as follows: , formula (20).

[0069] , formula (21).

[0070] , formula (22).

[0071] Where N represents the total number of samples, that is, the number of trajectory location points; Indicates the actual latitude and longitude of the i-th location; represents the latitude and longitude of the i-th location predicted by the model. Formula (20) is used to measure the average error of the two-dimensional Euclidean distance between the predicted location and the actual location (the unit is consistent with the input data, usually degrees or projected plane coordinate units). M represents the number of positive samples or negative samples; is a positive sample, representing the representation of the i-th true trajectory, which is obtained by averaging the representation vectors of all base station nodes in the trajectory; is a negative sample, representing the representation of the i-th abnormal trajectory, which is usually constructed by the following methods: randomly sampling the trajectories of other users; perturbing the trajectory of the same user (such as randomly replacing some nodes); generating adversarial negative examples (such as through generative models); It is used to represent the similarity score that maximizes the true trajectory. is the representation of each actual trajectory, which is obtained by averaging the corresponding node representations in the trajectory among the above node representations. Definition of . ( ) represent the actual longitude and predicted longitude respectively, They represent the actual latitude and the predicted latitude respectively, and λ is used to adjust the weight.

[0072] Using the above loss function, the model is trained by adjusting parameters. The best model is saved, and the terminal's signaling data is then fed into the trained model for processing to generate the terminal's actual trajectory location. After training in these steps, the model can analyze the discrepancy between the base station's location and the terminal's actual location, thereby generating the actual terminal's travel trajectory for use in more downstream tasks.

[0073] It should be noted that, referring to Figure 2 As shown, the training process of the above model includes: starting, dividing the data set; selecting model hyperparameters, model training evaluation, whether the model effect is optimal, if not, continuing to execute the step of selecting model hyperparameters, if optimal, saving the model parameters, saving the node representation, and ending.

[0074] In a specific embodiment, the present application describes a method for generating terminal trajectories based on communication signaling data as follows: Data preprocessing: Collect signaling data and actual travel data of users. Construct a mapping from actual signaling trajectories to actual travel locations, and filter out abnormal data. Feature extraction: Collect spatiotemporal position information in signaling data through a spatiotemporal encoder to obtain a spatiotemporal representation. Construct a base station interaction representation to capture the conversion pattern and spatial correlation of the base station, as well as the interaction relationship with the surrounding environment. Combine auxiliary information and graph neural networks to obtain a comprehensive representation. Use a model based on an attention mechanism to learn the global and local spatiotemporal relationships of the signaling sequence, capture the difference between it and the actual location, and generate the actual location of the user.

[0075] In summary, the solution of this application significantly improves the accuracy and generalization ability of terminal trajectory generation through deep modeling of communication signaling data and fusion of multi-source information, and shows obvious advantages in processing efficiency, multimodal adaptability and practical application scenarios. The specific technical effects are as follows: (1) Positioning accuracy is significantly improved: The global-local transfer graph modeling of this application combines base station adjacency (switching frequency, spatial correlation) with individual terminal trajectory patterns to capture group patterns and individual preferences; and the designed multimodal attention mechanism dynamically integrates base station attributes, road network connectivity, and POI semantic features to improve environmental perception. Compared with traditional rule-based and clustering methods, positioning error has been reduced, mainly due to the nonlinear modeling ability of deep learning models for complex spatiotemporal patterns. With the support of multi-source information fusion (POI, road network), positioning accuracy has been improved compared to methods that rely solely on trajectory information. By enhancing the interactive perception ability of base stations and the environment, signal drift interference is reduced.

[0076] (2) It supports unified modeling of multiple travel modes such as walking, driving, and public transportation, eliminating the need to design dedicated rules for a single travel mode, thereby improving processing efficiency. Through the spatiotemporal encoder and self-supervised learning module, it adaptively extracts the implicit features of different travel modes (such as speed changes and stop patterns).

[0077] (3) Innovations have been made in multi-source data fusion and modeling: environmental features such as road density and POI type distribution are encoded into high-dimensional vectors, and weighted fusion is performed through the attention mechanism to enhance base station context perception; base station-road and base station-POI association matrices are constructed to capture the correlation between user behavior and geographic semantics (such as high probability of staying in commercial areas).

[0078] (4) By using a multi-task loss function (geographic coordinate reconstruction plus positive and negative sample comparison), the model's dependence on labeled data is reduced, the generalization ability of the model in sparse signaling scenarios is improved, and self-supervised learning optimization is achieved; the introduction of perturbation trajectories and cross-user negative examples enhances the model's robustness to noisy data (such as base station switching anomalies).

[0079] (5) Downstream application scalability and diversified scenario support: Traffic management: Generate high-precision vehicle trajectories to optimize traffic light control and congestion prediction; Urban safety: Identify traffic accident areas through abnormal trajectory detection (such as long stops and frequent turns); Business analysis: Combine POI data to analyze user consumption behavior and regional business vitality.

[0080] This proposal can serve applications in smart cities, transportation planning, public safety, and other fields, filling the technological gap in generating high-precision trajectory data. It also improves signaling data utilization, creating value-added data services for operators (such as user behavior analysis and regional traffic forecasting). Through multi-source fusion modeling, self-supervised optimization, and a dynamic graph attention mechanism, this proposal achieves reliable mapping from low-precision signaling data to high-resolution trajectories, providing core technical support for smart cities and mobile behavior analysis.

[0081] The above describes various methods of the embodiments of the present application. The following further provides apparatuses for implementing the above methods.

[0082] Please refer to Figure 3 , an embodiment of the present application further provides a terminal trajectory generating device, comprising: A first determination module 31 is configured to determine mapping data between the terminal's signaling and location. The first determination module 31 includes: a first acquisition unit configured to acquire communication signaling data and actual trajectory data of the terminal; a first processing unit configured to perform time sorting and outlier removal operations on the communication signaling data to generate a signaling sequence; and a second processing unit configured to determine, based on the signaling sequence and the actual location in the actual trajectory data, the mapping data having a mapping relationship between the signaling and the location in a time-aligned manner, and to filter out outlier data. A first processing module 32 is configured to obtain time features and position features in the mapping data using a spatiotemporal encoder; The second processing module 33 is used to input the time feature and the location feature into a preset attention mechanism model to determine the actual location of the terminal.

[0083] Optionally, the first processing module 32 includes: a third processing unit, configured to normalize the timestamp information in the mapping data using the time branch in the spatiotemporal encoder to determine an entry time parameter, an exit time parameter, and a duration parameter; a fourth processing unit, configured to generate corresponding time feature vectors for the entry time parameter, the exit time parameter, and the duration parameter respectively through a trainable linear layer; A first determining unit is configured to capture the periodicity of the time component using a cosine function and determine a periodic feature; The second acquisition unit is configured to perform feature fusion on the corresponding time feature vector and the periodic feature to acquire the time feature in the mapping data.

[0084] Optionally, the first processing module 32 further includes: a fifth processing unit, configured to obtain the base station positions and the relative positional relationships between different base stations in the mapping data using a multilayer perceptron in the spatiotemporal encoder; The third acquisition unit is used to perform comprehensive coding representation on the base station position and the relative position relationship to obtain the position feature in the mapping data.

[0085] Optionally, the terminal trajectory generating device of the present application further includes: A construction module is configured to obtain a base station adjacency relationship and determine a global spatiotemporal representation based on the base station adjacency relationship; A third processing module is configured to construct a local transfer subgraph for the movement trajectory of a single terminal, and extract personalized movement pattern features using a hierarchical attention mechanism based on the local transfer subgraph; The fourth processing module is used to extract the road network topology characteristics and the distribution characteristics of points of interest within the preset range of the base station; a fifth processing module, configured to generate a geographic semantic representation through a spatial attention network according to the road network topology characteristics and the POI distribution characteristics; a sixth processing module, configured to perform gated fusion on the global spatiotemporal representation, the personalized mobility pattern features, and the geographic semantic representation, and output comprehensive base station information including multi-dimensional interactive relationships; The seventh processing module is used to use the comprehensive base station information to train the initial attention mechanism model to generate the attention mechanism model.

[0086] Optionally, the above building blocks include: a fourth acquiring unit, configured to acquire the spatial positions of all base stations in the target area and a terminal transfer mode between different base stations; A second determining unit is configured to determine the base station adjacency relationship by using the spatial position of the base station as a node and the terminal transfer pattern as an edge; A first construction unit is configured to construct a global transfer graph of base stations based on the base station adjacency relationship; The third determining unit is configured to determine a global spatiotemporal representation according to the global transition graph.

[0087] Optionally, the seventh processing module includes: A fifth acquiring unit, configured to acquire a sample time feature and a sample position feature generated based on the sample; a sixth processing unit, configured to generate a multi-time-step interaction representation using a multi-head self-attention mechanism based on the comprehensive base station information, the sample time feature, and the sample position feature; The second construction unit is used to construct a multi-task loss function; the multi-task loss function includes a geographic coordinate reconstruction error term, a positive sample representation similarity term, and a negative sample comparison term, and optimizes model parameters through self-supervised learning; The third building module is used to build the attention mechanism model based on the interaction representation and the multi-task loss function.

[0088] It should be noted that the device in this embodiment is a device corresponding to the above-mentioned method, and the implementation methods in the above-mentioned embodiments are all applicable to the embodiments of this device and can achieve the same technical effects. The above-mentioned device provided in the embodiment of this application can implement all the method steps implemented in the above-mentioned method embodiment and can achieve the same technical effects. The parts and beneficial effects of this embodiment that are the same as those in the method embodiment will not be specifically described here.

[0089] The present application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the terminal trajectory generation method embodiment described above and achieves the same technical effects. To avoid repetition, the details are omitted here. The computer-readable storage medium may be, for example, a read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0090] The present application also provides a computer program product including computer instructions. When the computer instructions are executed by a processor, the various processes of the terminal trajectory generation method embodiment described above are implemented, and the same technical effects can be achieved. To avoid repetition, they are not described here.

[0091] It should be noted that the collection, collection, updating, analysis, processing, use, transmission, and storage of user personal information involved in the technical solutions disclosed herein comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken with respect to personal information to prevent unauthorized access to user personal information data and maintain the security of user personal information and network security.

[0092] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0093] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a more preferred embodiment. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of this application.

[0094] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A terminal trajectory generation method, characterized in that: include: Determining mapping data between signaling and a location of a terminal, including: obtaining communication signaling data and real trajectory data of the terminal; performing time sorting and outlier removal operations on the communication signaling data to generate a signaling sequence; determining, based on the signaling sequence and the actual location in the real trajectory data, the mapping data having a mapping relationship between the signaling and the location in a time alignment manner, and filtering out the outlier data; Using a spatiotemporal encoder, obtaining time features and position features in the mapping data; The time feature and the location feature are input into a preset attention mechanism model to determine the actual location of the terminal.

2. The method according to claim 1, characterized in that Using a spatiotemporal encoder to obtain temporal features in the mapping data includes: Normalizing the timestamp information in the mapping data using the time branch in the spatiotemporal encoder to determine an entry time parameter, an exit time parameter, and a duration parameter; For the entry time parameter, the exit time parameter, and the duration parameter, respectively, a corresponding time feature vector is generated through a trainable linear layer; Use the cosine function to capture the periodicity of the time component and determine the periodic characteristics; The corresponding time feature vector and the periodic feature are subjected to feature fusion to obtain the time feature in the mapping data.

3. The method according to claim 1, characterized in that Using a spatiotemporal encoder to obtain position features in the mapping data includes: Using a multi-layer perceptron in the spatiotemporal encoder, obtaining base station positions and relative positional relationships between different base stations in the mapping data; The base station position and the relative position relationship are comprehensively coded and represented to obtain position features in the mapping data.

4. The method according to claim 1, wherein Before inputting the time feature and the location feature into a preset attention mechanism model to determine the actual location of the terminal, the method further includes: Obtaining base station adjacency relationships, and determining a global spatiotemporal representation based on the base station adjacency relationships; For the movement trajectory of a single terminal, a local transfer subgraph is constructed, and a hierarchical attention mechanism is used based on the local transfer subgraph to extract personalized movement pattern features; Extract road network topology features and point of interest distribution features within the preset range of the base station; Generating a geographic semantic representation through a spatial attention network according to the road network topology characteristics and the distribution characteristics of the points of interest; Performing gated fusion on the global spatiotemporal representation, the personalized mobility pattern features, and the geographic semantic representation to output comprehensive base station information containing multi-dimensional interactive relationships; The initial attention mechanism model is trained using the comprehensive base station information to generate the attention mechanism model.

5. The method according to claim 5, characterized in that Obtaining a base station adjacency relationship and determining a global spatiotemporal representation based on the base station adjacency relationship includes: Obtain the spatial locations of all base stations in the target area, as well as the terminal transfer patterns between different base stations; Determine the adjacency relationship of the base stations by using the spatial positions of the base stations as nodes and the terminal transfer patterns as edges; constructing a global transition graph of base stations based on the base station adjacency relationship; A global spatiotemporal representation is determined based on the global transition graph.

6. The method according to claim 5, characterized in that Using the comprehensive base station information, training the initial attention mechanism model to generate the attention mechanism model includes: Obtaining sample time features and sample location features generated based on the sample; Generate a multi-time-step interactive representation using a multi-head self-attention mechanism according to the comprehensive base station information, the sample time features, and the sample position features; Constructing a multi-task loss function; the multi-task loss function includes a geographic coordinate reconstruction error term, a positive sample representation similarity term, and a negative sample comparison term, and optimizing model parameters through self-supervised learning; The attention mechanism model is constructed based on the interaction representation and the multi-task loss function.

7. A terminal trajectory generating device, characterized in that: include: A first determination module is configured to determine mapping data between the signaling and the location of the terminal; wherein the first determination module includes: a first acquisition unit configured to acquire communication signaling data and real trajectory data of the terminal; a first processing unit configured to perform time sorting and outlier removal operations on the communication signaling data to generate a signaling sequence; and a second processing unit configured to determine, based on the signaling sequence and the actual location in the real trajectory data, the mapping data having a mapping relationship between the signaling and the location in a time-aligned manner, and filter out outlier data. A first processing module is configured to obtain time features and position features in the mapping data using a spatiotemporal encoder; The second processing module is used to input the time feature and the location feature into a preset attention mechanism model to determine the actual location of the terminal.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

9. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 6.

Images