Traffic information prediction method and device, equipment, storage medium and product

Abstract

The invention discloses a traffic information prediction method, device and equipment, a storage medium and a product, and relates to the technical fields of artificial intelligence, big data analysis, Internet of Things data application, intelligent transportation, location service, wireless communication data processing and the like. The method comprises the steps that spatial features of a to-be-predicted road section are obtained, a road network diagram is processed by a model for processing diagram structure data and then output, the road network diagram comprises nodes and edges used for representing the road section, the weight of the edges is determined according to historical data corresponding to the road network diagram, the spatial-temporal features of the to-be-predicted road section are obtained, and the spatial-temporal features of the to-be-predicted road section are obtained; according to the method, an event sequence determined according to historical data is processed by a sequence processing model and then output, event elements in the event sequence comprise time coding information and event information of a preset type of event, and road section features of a road section to be predicted are constructed according to spatial features and spatial-temporal features. The traffic information of the to-be-predicted road section in the future time period is predicted based on the road section features, and low-cost accurate prediction of the traffic information can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical fields of artificial intelligence, big data analysis, Internet of Things data application, intelligent transportation, location service and wireless communication data processing, and particularly relates to a traffic information prediction method and device, equipment, storage medium and product. BACKGROUND

[0002] With the acceleration of urbanization and the increasing demand for intelligent traffic management, it is particularly important to accurately predict traffic information such as traffic flow and congestion.

[0003] At present, traffic prediction mainly relies on the following technical solutions. The scheme based on fixed sensors deploys sensors on the road and predicts according to the directly detected traffic flow, speed and occupancy rate and other information of the cross section. However, this scheme has high deployment cost and limited coverage. The scheme based on floating car data maps the real-time position data uploaded by the floating vehicle to the road network through a map matching algorithm, calculates the driving speed and travel time of the vehicle on the road section, etc. However, this scheme has sparse data source, insufficient accuracy, and limited data acquisition cost and availability. The scheme based on traditional mobile communication data uses signaling data generated by the interaction between mobile phone users and mobile communication network, and analyzes the movement mode or connection time of users between different base stations for prediction. However, this scheme has low positioning accuracy, and the base station coverage range may be large and irregular, which is difficult to meet the accurate prediction requirements of road section level and long time sequence. The spatiotemporal prediction scheme based on deep learning mainly focuses on processing dense and high-quality data sources, and cannot process sparse data.

[0004] Therefore, the traffic information prediction in the prior art is still not perfect and needs to be improved. SUMMARY

[0005] The present application provides a traffic information prediction method, device, equipment, storage medium and product, which can solve the problem of low precision and poor accuracy in the prior art.

[0006] According to an aspect of the present application, a traffic information prediction method is provided, comprising:

[0007] obtaining a spatial feature corresponding to a preset historical period of a to-be-predicted road section, wherein the spatial feature is output by processing a road network graph by a first model, the first model is a model for processing graph structure data, the road network graph includes nodes for representing road sections and edges between the nodes, the weight of the edge is determined according to historical data of a preset historical period corresponding to the road network graph, and the historical data is traffic historical data for determining position information of a target vehicle;

[0008] acquire a space feature corresponding to a preset historical period of the to-be-predicted road section, wherein the space feature is output by a first model after processing a road network graph, the first model is a model for processing graph structure data, the road network graph includes nodes for representing road sections and edges between the nodes, a weight of the edge is determined according to historical data of a preset historical period corresponding to the road network graph, and the historical data is traffic historical data for determining position information of a target vehicle;

[0009] construct a road section feature of the to-be-predicted road section according to the space feature and the space-time feature;

[0010] predict traffic information of a future period of the to-be-predicted road section based on the road section feature.

[0011] According to another aspect of the present application, a traffic information prediction device is provided, which comprises:

[0012] a space feature acquisition module configured to acquire a space feature corresponding to a preset historical period of a to-be-predicted road section, wherein the space feature is output by a first model after processing a road network graph, the first model is a model for processing graph structure data, the road network graph includes nodes for representing road sections and edges between the nodes, a weight of the edge is determined according to historical data of a preset historical period corresponding to the road network graph, and the historical data is traffic historical data for determining position information of a target vehicle;

[0013] a space-time feature acquisition module configured to acquire a space-time feature corresponding to the preset historical period of the to-be-predicted road section, wherein the space-time feature is output by a second model after processing an event sequence, the second model is a model for processing sequences, the event sequence is determined according to the historical data, the event sequence includes a plurality of event elements, and the event elements include time coding information and event information of a preset type of event;

[0014] a feature construction module configured to construct a road section feature of the to-be-predicted road section according to the space feature and the space-time feature;

[0015] a traffic information prediction module configured to predict traffic information of a future period of the to-be-predicted road section based on the road section feature.

[0016] According to another aspect of the present application, an electronic device is provided, which comprises:

[0017] at least one processor;

[0018] and a memory in communication connection with the at least one processor;

[0019] The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the traffic information prediction method according to any embodiment of the present invention.

[0020] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the traffic information prediction method according to any embodiment of the present invention.

[0021] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the traffic information prediction method according to any embodiment of the present invention.

[0022] The technical solution of this invention involves obtaining spatial features corresponding to a preset historical time period for a road segment to be predicted. The spatial features are output by a first model processing a road network map. The first model is a model for processing graph structure data. The road network map includes nodes representing road segments and edges between the nodes. The weights of the edges are determined based on historical data corresponding to the preset historical time period of the road network map. The historical data is traffic historical data used to determine the location information of target vehicles. The spatiotemporal features corresponding to the preset historical time period for the road segment to be predicted are also obtained. These spatiotemporal features are output by a second model processing an event sequence. The second model is a model for processing sequences. The event sequence is determined based on historical data and includes multiple event elements, including time-encoded information and event information of a preset type of event. Based on the spatial and spatiotemporal features, road segment features are constructed for the road segment to be predicted. Finally, traffic information for the future time period of the road segment to be predicted is predicted based on these road segment features. By adopting the above technical solution, the weights of edges in the road network map and the event sequence are determined based on historical traffic data. The spatial characteristics of the road segment to be predicted are output by the understanding of the road network structure using the model used to process graph structure data, and the spatiotemporal characteristics of the road segment to be predicted are output by the processing capability of the model used to process sequences for sparse time-series events. The spatiotemporal dynamic information contained in the historical traffic data is fully explored, and the spatial and spatiotemporal characteristics are fused before prediction, thereby achieving accurate prediction of traffic information.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a traffic information prediction method provided by an embodiment of the present invention;

[0026] Figure 2 This is a flowchart of another traffic information prediction method provided by an embodiment of the present invention;

[0027] Figure 3 This is a schematic diagram of the structure of a traffic information prediction device according to an embodiment of the present invention;

[0028] Figure 4 This is a schematic diagram of the structure of an electronic device that implements the traffic information prediction method of this invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] To facilitate understanding of the technical solutions of the embodiments of the present invention, related technologies are described below. In the prior art, traffic prediction mainly relies on the following types of technical solutions.

[0032] Fixed sensor-based solutions deploy sensors such as inductive loop detectors, cameras, and radar on roads. Predictions are made based on directly detected traffic flow, speed, and occupancy information at cross-sections. For example, vehicles can be identified and counted by analyzing camera video streams, or inductive loop detectors can sense vehicle passage. However, this approach is costly to deploy, requiring civil engineering and wiring, and maintenance is expensive. Sensors are susceptible to environmental damage, and coverage is limited, typically only covering urban arterial roads, highways, or key intersections, with insufficient coverage for secondary urban roads, suburbs, and newly developed areas.

[0033] Floating car data-based solutions utilize real-time location data uploaded by Global Positioning System (GPS) devices or mobile applications (APPs) installed on taxis, buses, ride-hailing vehicles, and logistics vehicles. Map matching algorithms are used to map GPS point sequences onto the road network, calculating vehicle speeds and travel times on road segments. For example, aggregating speed information from a large number of floating cars can estimate the average speed and congestion levels of road segments. However, this approach relies on data sources limited to specific vehicle groups, making it difficult to fully represent overall traffic flow, especially in areas or times with fewer floating cars, exhibiting sparsity. For instance, at night or in areas with low vehicle traffic, the data points are insufficient to accurately reflect real-time road conditions, limiting cost and availability.

[0034] Traditional mobile communication data-based solutions utilize signaling data generated from interactions between mobile phone users and mobile communication networks. By analyzing user movement patterns or connection durations between different base stations, they indirectly infer regional population density, origin-destination (OD) distribution, or general traffic conditions. For example, mobile signaling data can be used to statistically analyze changes in the number of active users in a specific area to reflect traffic congestion levels. However, this approach typically only locates at the base station (BS) or location area code (LAC) level, resulting in coarse spatial resolution and difficulty in accurately mapping to specific urban road segments. Base station coverage can be vast and irregular, especially in suburban areas, making it difficult to obtain continuous and accurate vehicle trajectory and speed information. Furthermore, traditional models or simple statistical methods struggle to effectively handle the sparsity, noise, and complex spatiotemporal dependencies of mobile communication data. They are usually limited to regional or short-term state estimations, failing to meet the requirements for accurate predictions at the road segment level and over long time series.

[0035] Currently, spatiotemporal prediction schemes based on deep learning are mainly designed and optimized to handle dense and high-quality data sources. For example, they use graph neural networks (GNNs) to learn road network topology and convolutional networks to handle time dependencies. However, these schemes are only suitable for processing relatively dense and regular traffic data in spatiotemporal terms provided by fixed sensors and cannot handle sparse data.

[0036] Figure 1 This is a flowchart of a traffic information prediction method according to an embodiment of the present invention. This embodiment is applicable to situations involving traffic information prediction. The method can be executed by a traffic information prediction device, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method includes:

[0037] Step 101: Obtain the spatial features corresponding to the preset historical time period of the road segment to be predicted. The spatial features are output by the first model after processing the road network map. The first model is a model for processing graph structure data. The road network map includes nodes for representing road segments and edges between the nodes. The weight of the edges is determined according to the historical data of the preset historical time period corresponding to the road network map. The historical data is traffic historical data used to determine the location information of the target vehicle.

[0038] In this embodiment of the invention, the road segment to be predicted can be any road segment that is passable by vehicles. The preset historical time period can be any period of time in the past, and can be set according to actual needs, without any specific limitation, such as the most recent day, the most recent week, or the most recent month.

[0039] For example, the spatial features corresponding to the road segment to be predicted can be the spatial topological features of the road segment in the road network. These features can be used to characterize at least one dimension of the road segment's structural location, structural importance, geometric features, connectivity, neighborhood environment, and spatial propagation patterns of traffic flow. The spatial features are output by the first model after processing the road network map. The first model is a model used to process graph-structured data, and the specific model type is not limited. It can be a neural network model, such as a graph neural network-based model, like GNN, Graph Convolutional Network (GCN), Graph Attention Network (GAT), Graph Sample and Aggregate (GraphSAGE), Message Passing Neural Network (MPNN), and Chebyshev Convolutional Network (ChebNet). Among them, GAT can learn the importance of the influence of different neighboring road segments on the current road segment through the attention mechanism, making it more suitable for complex urban road networks.

[0040] For example, the urban road network can be abstracted as a graph, which can be denoted as: , among which, nodes Representing a road segment, each node can contain static attributes such as segment length, road class, speed limit, and number of lanes, etc. This represents the connection relationships between road segments (usually connected by intersections). Edges can be weighted, and the weight of each edge is determined based on historical data for a preset historical period corresponding to the road network map.

[0041] In this embodiment of the invention, historical data refers to traffic history data used to determine the location information of a target vehicle, such as GPS data or sensor data. It can also be low-cost sparse data, such as sparse mobile data (hereinafter referred to as sparse mobile data or sparse data) used to determine the location information of a target vehicle. The target vehicle can be any vehicle that passes through the road segment to be predicted within a preset historical time period. The sparse mobile communication data can be sparse mobile communication data, such as General Packet Radio Service (GPRS) call detail records, Wireless Fidelity (WiFi) probe data, Bluetooth signal data, or low-frequency location data generated by other IoT devices. The above data has the characteristics of low acquisition cost, wide coverage, sparseness, and low frequency. The vehicle can be equipped with relevant communication modules with the above communication methods, such as a Subscriber Identity Module (SIM) card, WiFi module, Bluetooth module, and other communication modules. The aforementioned communication module can communicate with preset wireless communication objects (such as base stations, wireless hotspots, Bluetooth devices, and other IoT devices). During communication, sparse motion data is generated. This sparse motion data can be used to identify the preset wireless communication object, and the location of the preset wireless communication object (referred to as the object's location) can generally be determined through methods such as table lookup. This allows for the determination of the target vehicle's location information. Due to the sparsity of historical data, the vehicle location information determined using historical data also exhibits sparsity. By using historical data to determine edge weights, the first model can learn richer features of the road segment to be predicted within a preset historical time period, improving prediction accuracy.

[0042] Optionally, the weight of the edge is determined based on the physical adjacency between road segments, the estimated travel time of the road segments, and the traffic flow correlation between the road segments, wherein the estimated travel time and / or the traffic flow correlation are determined based on the historical data. This allows for a more comprehensive characterization of the relationships between road segments. The physical adjacency can be, for example, a connection type, such as end-to-end connection, T-junction, or Y-junction; the estimated travel time can be the estimated travel duration of vehicles traveling on the road segments represented by the connected nodes; and the traffic flow correlation can be the travel duration between the road segments represented by the connected nodes, such as the time spent moving from one road segment to another.

[0043] Step 102: Obtain the spatiotemporal features corresponding to the preset historical time period of the road segment to be predicted. The spatiotemporal features are output by the second model after processing the event sequence. The second model is a model used to process the sequence. The event sequence is determined based on the historical data. The event sequence includes multiple event elements. The event elements include time encoding information and event information of preset type events.

[0044] For example, the spatiotemporal features corresponding to the road segment to be predicted can be dynamic features characterizing events occurring on the road segment and the time-related aspects of these events. Multiple types of possible events can be predefined, denoted as preset event types. The specific types are not limited and can be set based on the source of historical data. Taking GPRS call detail record (CDR) data as an example, preset event types could include events entering the road segment, events leaving the road segment, CDR generation events, dwelling events, and multi-location recurring events, etc. The spatiotemporal features are output by a second model after processing the event sequence. The second model is the model used to process the sequence; the specific model type is not limited, such as a neural network model. For example, it could be a model based on a self-attention mechanism, such as a Transformer model, or a newer variant of an RNN model, such as a Gated Recurrent Unit (GRU). Other models could also be used, such as Neural Ordinary Differential Equations (Neural ODEs) or Time-Aware Long Short-Term Memory (Time-Aware LSTM) networks. Optionally, the first and second models can also serve as the internal network structure of an end-to-end deep learning model, enabling the direct output of traffic information prediction results from historical data through the prediction of this end-to-end deep learning model, avoiding the explicit separation of the first and second models. For example, a Spatio-Temporal Graph Neural Network (STGNN) can be designed to integrate temporal and spatial processing within the same network architecture.

[0045] For example, the event sequence includes multiple event elements, each of which includes time-encoded information and event information for a preset type of event. The time-encoded information can be obtained by encoding the relevant information of the timestamp corresponding to the preset type of event; the timestamp is used to characterize the occurrence time of the corresponding preset type of event. The event information can include encoded information related to the event type, event-related call detail records (CDRs) such as traffic and duration estimates, and the confidence level of the association between the base station location and the road segment to be predicted.

[0046] For example, the event sequence is input into a second model for processing. The second model can learn the dependencies between different preset types of events, as well as the activity patterns of the target vehicle and the evolution of traffic-related patterns, to obtain the spatiotemporal features of the road segment to be predicted corresponding to a preset historical time period. For example, the second model is a model based on a self-attention mechanism, such as the standard Transformer encoder or its variants. The self-attention mechanism can directly calculate the dependency weights between any two events in the event sequence (regardless of the time interval), which makes it naturally suitable for capturing long-distance dependencies and burst patterns that may exist in sparse data. For example, a call detail record (CDR) event at a distant base station tens of minutes ago may predict the traffic that will arrive at the current road segment later, which can overcome the gradient or memory problems that RNN-type models may encounter when processing long sparse sequences.

[0047] Step 103: Construct the road segment features of the road segment to be predicted based on the spatial features and the spatiotemporal features.

[0048] In this step, the spatial features extracted in the previous steps and the dynamic spatiotemporal features can be fused to obtain a more comprehensive and integrated representation of the features of the road segment to be predicted, which is denoted as the road segment feature. The specific fusion method is not limited, such as direct splicing, element-by-element operation, or weighted fusion, etc., and one or more fusion methods can be used.

[0049] Step 104: Based on the road segment characteristics, predict the traffic information of the road segment to be predicted for future time periods.

[0050] For example, after obtaining the fused road segment features, these features can be input into the prediction layer to output a prediction of traffic information for the road segment in the future. This traffic information can be information characterizing traffic conditions. The prediction layer can be one or more layers, and can be a fully connected network (FNC) or a multilayer perceptron (MLP), etc. The type of predicted traffic information is not limited, such as traffic flow, speed, or congestion level. For example, traffic flow can be a prediction of the number of vehicles passing through the road segment per unit time in the future, and can be a relative value or an estimated absolute value; average speed can be a prediction of the average speed of the road segment in the future; congestion level can be a prediction of the congestion state of the road segment in the future, such as smooth traffic, slow traffic, and congestion, and can be converted into a multi-class classification problem. The specific duration of the future period is not limited, such as 15 minutes, 30 minutes, or 1 hour.

[0051] The traffic information prediction method of this invention determines the edge weights and event sequences in a road network graph based on historical traffic data. It utilizes a model for processing graph-structured data to understand the road network structure and outputs the spatial features of the road segment to be predicted. It also utilizes a model for processing sequences to handle sparse temporal events and outputs the spatiotemporal features of the road segment to be predicted. This fully mines the spatiotemporal dynamic information contained in historical traffic data, and then fuses the spatial and spatiotemporal features before prediction, thereby achieving accurate traffic information prediction. Furthermore, when using low-cost, sparse mobile communication data, the above scheme can also achieve accurate traffic information prediction at a low cost.

[0052] Figure 2 This is a flowchart of another traffic information prediction method provided by an embodiment of the present invention. This embodiment is an optimization based on the above-mentioned optional embodiments. The historical data is sparse data based on mobile communication used to determine the location information of the target vehicle. Figure 2 As shown, the method includes:

[0053] Step 201: Determine the driving trajectory of the candidate vehicle based on the candidate data. The candidate data is sparse data based on mobile communication used to determine the location information of the candidate vehicle. The driving trajectory includes multiple sparse trajectory points, and each trajectory point is associated with time information and driving segment.

[0054] It should be noted that the candidate data and the aforementioned historical data can be of the same type. For example, the historical data is traffic history data used to determine the location information of the target vehicle, and the candidate data is traffic history data used to determine the location information of the candidate vehicles. Preferably, the historical data is sparse data based on mobile communication used to determine the location information of the target vehicle, and the candidate data is sparse data based on mobile communication used to determine the location information of the candidate vehicles.

[0055] For example, to make traffic information predictions more accurate, sparse movement data can be collected comprehensively over a large area and designated as candidate data. Then, sparse movement data for predicting the road segment to be predicted can be selected from this data, which is the historical data mentioned earlier. Candidate vehicles are any vehicles that generate sparse movement data within the collection area.

[0056] For example, after obtaining the candidate data, the candidate data is aggregated using candidate vehicles as the aggregation dimension to obtain candidate data corresponding to each candidate vehicle. This allows for the determination of the driving trajectory of each candidate vehicle, which includes multiple sparse trajectory points. Each trajectory point is associated with time information and a driving segment. The time information can be time-related information about the candidate vehicle's journey within the driving segment, which can be determined from the time-related information in the candidate data.

[0057] Optionally, candidate data refers to data generated when candidate vehicles communicate with preset wireless communication objects. This step may specifically include: determining the object identifier of the preset wireless communication object contained in the candidate data; determining the object location and communication signal coverage area of ​​the preset wireless communication object based on the object identifier; for each candidate vehicle, sorting the candidate data according to the time information of the current candidate vehicle to obtain the object connection sequence corresponding to the current candidate vehicle; and, based on the communication signal coverage area, matching the object connection sequence with a preset road network map using a preset matching method to obtain the driving trajectory corresponding to the current candidate vehicle. The object connection sequence includes multiple object elements, and each object element includes time information and object location. Therefore, the road segment-level driving trajectory of each candidate vehicle can be accurately obtained.

[0058] The preset wireless communication target can be a base station, a Wi-Fi hotspot, a Bluetooth device, or other IoT devices. The object identifier of the preset wireless communication target can include an identifier indicating the identity of the preset wireless communication target. Taking traffic history data as GPRS call detail record data as an example, the preset wireless communication target is a base station.

[0059] For example, GPRS call detail record (CDR) data may include many key fields, such as user identifier, timestamp, location information, network information, device information, session information, traffic volume information, and network type. The user identifier can be, for example, PRI_IDENTITY or Charging Party International Mobile Subscriber Identity (ChargingPartyIMSI) (which needs to be anonymized); the timestamp can be, for example, START_DATE, END_DATE, CUST_LOCAL_START_DATE, CUST_LOCAL_END_DATE, or TimeStampOfSGSN (network device recorded time), where the first two are Coordinated Universal Time (UTC) and the middle two are local time; location information can include CallingCellID, LAC, and CellIdentity (CellId); network information can include AccessNetworkAddress, Serving GPRS Support Node (SGSN) address, and Gateway GPRS Support Node (GGSN) address; and device information can be International Mobile Equipment Identity (IMSI). Identity (IMEI) (needs anonymization); session information may include session ID (SESSION_ID), etc.; traffic volume information may include total traffic, uplink traffic, and downlink traffic, etc.; network type may include Radio Access Technology Type (RATType). The object identifier for the preset wireless communication object can be CallingCellID, LAC, or CellId.

[0060] For example, after obtaining the original candidate data, it can be preprocessed, such as by cleaning and standardizing, to lay the foundation for subsequent trajectory reconstruction and feature extraction. This mainly includes key steps such as data source identification, data cleaning, time unification, and base station location mapping. Data cleaning can include removing invalid records (such as records with missing fields, abnormal times, or incorrect location information), processing duplicate call detail records (CDRs), and user anonymization, such as replacing sensitive fields like PRI_IDENTITY, ChargingPartyIMSI, and IMEI with hashes or other techniques to generate unique anonymous user IDs. When processing duplicate CDRs, a deduplication algorithm based on a combination of CDR_ID, SESSION_ID, timestamp, and user identifier can be used. For similar CDR records within a time window, duplicate records are identified and removed by comparing key fields (such as user ID, base station location, and time difference threshold), retaining the record with the highest data quality or the most accurate timestamp. After data cleaning, time unification and time zone conversion can be performed. For example, all relevant timestamps can be converted to Unix timestamps or other standard formats under the local time zone. This can be done using CUST_LOCAL_START_DATE, CUST_LOCAL_END_DATE, or by combining it with TimeZoneOfSGSN.

[0061] For example, the object identifier of the preset wireless communication object in the candidate data can be used as the primary key and matched with a hash table or database query of a pre-prepared base station parameter library. The base station parameter library may contain data such as base station ID, latitude and longitude coordinates, and coverage area. Through query matching, the object location and communication signal coverage area of ​​the preset wireless communication object can be determined, such as the base station latitude and longitude coordinates and coverage area.

[0062] Subsequently, the processed GPRS call detail records (CDRs) are converted into sparse driving trajectories of vehicles on the actual road network. This overcomes the problems of low base station positioning accuracy and sparse CDR frequency, providing accurate road segment-level data input for subsequent traffic information prediction. Specifically, the preprocessed GPRS CDRs can be grouped by anonymous user ID. Each candidate vehicle (corresponding to an anonymous user ID) can be processed sequentially as the current candidate vehicle. The data is sorted according to the time information (such as CUST_LOCAL_START_DATE or START_DATE) in the candidate data corresponding to the current candidate vehicle, resulting in the object connection sequence corresponding to the current candidate vehicle, such as the original base station connection sequence. The sequence is in the form of [(timestamp1, cell_coord1), (timestamp2, cell_coord2), ...], where cell_coord is the latitude and longitude of the base station. The preset road network map can be, for example, high-precision urban digital map road network data (including information such as road segments, intersections, road grades, and speed limits). The preset matching method can be based on a map matching algorithm, such as one based on a Hidden Markov Model (HMM), Kalman filtering, or deep learning. Using this preset matching method, the sparse base station coordinate sequences of candidate vehicles are mapped to the actual road network. Considering the relatively low positioning accuracy of base stations (which may cover multiple road segments), probabilistic or fuzzy matching strategies can be employed.

[0063] The following explanation uses a probabilistic map matching method based on Hidden Markov Models (HMM) as an example. For instance, for a base station coordinate point, road segments within the communication signal coverage area of ​​that base station are designated as candidate road segments, and the association probabilities between the base station and multiple candidate road segments are calculated. Specifically, the emission probability and transition probability can be calculated separately. Then, based on the emission and transition probabilities, the Viterbi algorithm or other optimization methods are used to find the most probable sequence of road segments, which serves as the sparse driving trajectory of the vehicle, i.e., the driving trajectory corresponding to the current candidate vehicle. The trajectory is in the form of [(timestamp1, road_segment_id1), (timestamp2, road_segment_id2), ...], where road_segment_id is the road segment identifier. Among them, the transmission probability reflects the spatial attenuation law of the base station signal strength and follows the logarithmic distance path loss model. The transfer probability reflects the motion constraints and preferences of vehicles in the road network and takes into account the road network connectivity, travel time rationality and route selection habits. By combining the transmission probability and the transfer probability, the spatial coverage characteristics of the communication network and the topological constraints of the traffic network are effectively integrated.

[0064] The transmission probability represents the likelihood that a base station signal will cover the road segment; the closer the distance, the higher the probability. It can be calculated as follows, assuming the base station location is... Candidate road sections The center point is The emission probability is then defined as:

[0065]

[0066] in, The shortest Euclidean distance from the base station location to the road segment, in meters; This is the standard deviation of the base station positioning error, which is usually set to 1 / 3 of the base station's coverage radius, approximately 200-500 meters.

[0067] Among them, for road segment transfers between adjacent time points, from road segment to section of road The transition probability is:

[0068]

[0069] in, For road section The set of directly adjacent road segments; From arrive Path cost; A low probability value (approximately 0.01) allows transfers between non-adjacent but reachable road segments; The temperature parameter controls the sharpness of the probability distribution.

[0070] in,

[0071] in, For the road network from arrive The shortest path distance; To estimate the average speed (based on road class, urban roads are typically 20-60 km / h); The time interval between two call detail records; For velocity variance; This refers to the number of jumps in the road segment; For weight parameters (e.g., ).

[0072] Optionally, based on the coverage area of ​​the communication signal, a preset matching method is used to match the object connection sequence with a preset road network map, including: determining the sparsity index of the object connection sequence; in response to the sparsity index being greater than a preset threshold, interpolating the object connection sequence to obtain an interpolated object connection sequence; and matching the interpolated object connection sequence with the preset road network map using the preset matching method. Thus, by setting a sparsity index to evaluate the sparsity of the object connection sequence, if it is too sparse, interpolation can be performed to further improve prediction accuracy, while preserving the sparsity characteristics of the original data during the interpolation process.

[0073] The sparsity index can be determined based on at least one of the following: the difference in time information between adjacent object elements in the object connection sequence, the distance between object locations, and the number of road segments traversed. Preset thresholds can be set according to actual needs; for example, a preset threshold of 10 minutes for the time information difference, 2 kilometers for the distance between object locations, and 5 for the number of road segments. During interpolation, methods such as linear interpolation or more complex path inference methods based on road network constraints can be used to supplement possible road segments between two known road segment points. The specific interpolation method is not limited.

[0074] Optionally, determining the sparsity index of the object connection sequence includes: for each pair of object elements in the object connection sequence, calculating a first difference in the time information of the current object element pair and a second difference in the object position; calculating a first product of the first difference and the second difference; calculating the sum of the first products of all object element pairs; calculating a third difference in the time information of the last object element and the first object element and a fourth difference in the object position; calculating a second product of the third difference and the fourth difference; and determining the sparsity index of the object connection sequence based on the quotient of the sum and the second product. This quantifies the sparsity index and improves the accuracy of the standard used to determine whether interpolation processing is needed.

[0075] For example, the sparsity index can be expressed as:

[0076]

[0077] in, Sparsity index; This is the time interval between adjacent records, also known as the first difference. This represents the spatial distance, also known as the second difference. This is the total duration, also known as the third difference. This is the total distance, also known as the fourth difference. Optionally, the preset threshold is 0.7. If the value is greater than 0.7, it can be determined that the object connection sequence is too sparse and interpolation processing is required.

[0078] Step 202: Determine the driving trajectory containing trajectory points that meet preset conditions as the target driving trajectory. The preset conditions are that the time information associated with the trajectory points is within the preset historical time period, and the associated driving segment is the segment to be predicted.

[0079] For example, after determining the driving trajectory corresponding to each candidate vehicle, for the road segment to be predicted, the driving trajectory that passes through the road segment to be predicted within a preset historical time period is found as the target driving trajectory. That is, the time information associated with at least one trajectory point in the target driving trajectory is within the preset historical time period, and the associated driving segment is the road segment to be predicted.

[0080] Step 203: Identify the candidate vehicles corresponding to the target driving trajectory as the target vehicles.

[0081] For example, after determining the target driving trajectory, the candidate vehicles corresponding to the target driving trajectory are determined as the target vehicles. That is, the candidate vehicles corresponding to the target driving trajectory are vehicles that pass through the road segment to be predicted within a preset historical time period.

[0082] Step 204: Determine the candidate data corresponding to the target vehicle as historical data.

[0083] For example, after identifying the target vehicle, the candidate data corresponding to the target vehicle is identified as historical data.

[0084] Step 205: Obtain the spatial characteristics of the preset historical time period of the road segment to be predicted.

[0085] The spatial features are output by the first model after processing the road network map. The first model is a neural network model used to process graph structure data. The road network map includes nodes for representing road segments and edges between the nodes. The weight of the edges is determined according to the historical data of the preset historical period corresponding to the road network map.

[0086] For example, the first model can be a Generative Neural Network (GNN) model. The GNN model learns the static spatial topological features of the road network. By aggregating information from neighboring road segments, it generates a feature vector (spatial feature) for each road segment, encoding its structural location, connectivity, and neighborhood environment, which serves as the input for subsequent static spatial information fusion. Optionally, the initial input to the GNN can be the static attributes of the road segments, or it can combine historical (e.g., previous moment) traffic conditions as the initial features of the nodes. The GNN iteratively learns the embedding of each road segment node through neighbor node information aggregation and update operations. This embedding contains its topological structure information and local neighborhood features within the road network. Each road segment... Obtain a spatial feature vector In other words, the spatial features in this step, the physical meaning of this vector includes: topological representation, which encodes the location information and connectivity features of the road segment in the entire road network, reflecting its role as a traffic hub or edge road segment; neighborhood influence modeling, which captures the impact of the static attributes of adjacent road segments (such as road grade, number of lanes, and speed limit) on the traffic characteristics of the current road segment through multi-layer graph convolution aggregation; spatial propagation features, which reflect the spatial propagation law of traffic flow in the road network, including the potential path and intensity of upstream congestion propagating downstream; structural importance measurement, which implicitly characterizes the strategic importance of the road segment in the road network, such as whether it is a key connection point, bottleneck road segment, or main passage; and geometric and functional attribute fusion, which integrates the distribution patterns of the geometric features (length and direction) and functional features (road type and traffic capacity) of the road segment in the spatial neighborhood.

[0087] Step 206: Obtain the spatiotemporal characteristics of the preset historical time period of the road segment to be predicted.

[0088] The spatiotemporal features are output by the second model after processing the event sequence. The second model is a neural network model used to process the sequence. The event sequence is determined based on historical data and includes multiple event elements, including time-encoded information and event information of preset event types.

[0089] In this step, the second model can be a Transformer model, which is used to extract dynamic spatiotemporal features from sparse movement events. First, a sparse event sequence with irregular time is constructed for each road segment. Then, the time information is encoded using an encoding method designed for irregular timestamps. With the help of the powerful self-attention mechanism of Transformer, this step can effectively capture long-distance spatiotemporal dependencies and burst patterns in sparse data and generate dynamic spatiotemporal feature vectors.

[0090] For example, for a road segment to be predicted, records associated with it within a past time period T (a preset historical time period) can be filtered from the sparse data corresponding to all driving trajectories (e.g., matching the road segment or connecting to a base station covering the road segment). These records form a temporally irregular and event-sparse sequence, denoted as the event data sequence. Each event can be represented as (timestamp, user_id_anon, event_feature). The event_feature can include event type (such as entry or exit estimation and call detail record (CDR) generation), CDR association information (such as traffic and duration estimation), and the confidence level of the association between the base station location and the road segment. The key is to preserve the timestamp irregularity of the original data.

[0091] For example, to enable the Transformer to handle event sequences with sparsity and temporal irregularities, this embodiment of the invention has a special design for time encoding. Optionally, the time encoding information is determined in the following way: Fourier temporal feature encoding is performed on the timestamps corresponding to the preset type of event to obtain first encoding information, wherein the timestamps are used to characterize the occurrence time of the corresponding preset type of event; a second encoding information is determined based on a preset mapping relationship, wherein the preset mapping relationship is obtained by learning the mapping relationship from timestamps to high-dimensional vectors through a neural network containing trainable parameter matrices of multiple time scales; the time interval between the preset type of events with adjacent timestamps is calculated and encoded to obtain third encoding information; the time encoding information is determined based on the first encoding information, the second encoding information, and the third encoding information. Thus, the timestamps of events can be encoded to adapt to the second model, improving the processing performance of the second model.

[0092] For example, considering the irregular timestamp characteristics of sparse mobile data, a continuous-time encoding method is employed. Fourier temporal feature encoding is used to encode the timestamps. Mapped to a combination of sine and cosine functions: ,in , The first encoded information is the encoding dimension. A trainable parameter matrix with multiple time scales is designed, and a neural network learns the mapping relationship from timestamps to high-dimensional vectors to achieve learnable temporal embedding; the resulting encoded information is the second encoded information. Temporal difference encoding is then performed to calculate the time interval between adjacent events. This information is then encoded as an additional feature to capture the temporal distance information between sparse events, and the resulting encoded information is the third encoded information.

[0093] For example, the event_feature is vectorized to obtain event information. Event elements are determined based on the time encoding information and the event information, and then an event sequence is constructed. The event elements may also include the encoded, anonymized user ID (user_id_anon).

[0094] For example, by using Transformer to process the above-mentioned sparse event embedding sequence with time encoding, i.e., event sequence, the spatiotemporal features corresponding to a preset historical time period of the road segment to be predicted can be obtained, denoted as road segment. in the past Dynamic spatiotemporal feature vectors within a time period This vector encodes the activity patterns and evolutionary laws learned from sparse events.

[0095] Step 207: Construct the road segment characteristics of the road segment to be predicted based on spatial and spatiotemporal characteristics.

[0096] In this step, the static spatial features extracted by GNN and the dynamic spatiotemporal features extracted by Transformer can be deeply fused using advanced methods such as attention mechanisms to generate a comprehensive feature representation. For example, the static spatial features obtained from GNN... (in (for spatial feature dimensions) and dynamic spatiotemporal features obtained from the Transformer (in By fusing the spatiotemporal features, road segments are obtained. Comprehensive feature representation of (road segment to be predicted) That is, the characteristics of the road segment.

[0097] Optionally, the spatial features and the spatiotemporal features are fused based on at least one of the following fusion methods (1) to (4) to obtain the road segment features of the road segment to be predicted:

[0098] (1) The spatial features and the spatiotemporal features are directly spliced ​​together.

[0099] For example, spatial features and spatiotemporal features can be directly concatenated, such as:

[0100]

[0101] (2) Perform element-wise operations on the spatial features and the spatiotemporal features, wherein the element-wise operations include element-wise addition or element-wise multiplication.

[0102] For example, before performing element-by-element operations, it is necessary to ensure That is, the spatial features and the spatiotemporal features have the same dimensions. If they are not the same, they can be made the same by linear transformation before performing element-by-element operations.

[0103] (3) The spatial features and the spatiotemporal features are processed based on the gating mechanism to obtain the feature weights corresponding to the spatial features and the spatiotemporal features respectively, and the spatial features and the spatiotemporal features are weighted and fused based on the feature weights.

[0104] For example, based on the characteristics of the input features, the importance of spatial and spatiotemporal features is dynamically learned, and different feature weights are assigned.

[0105] For example, spatial features and spatiotemporal characteristics (If the dimensions are different, they can be mapped to the same dimension through a linear layer first.) The input is fed into a gating mechanism to calculate the gating weights, which can then be used to concatenate the features. (in (This represents features that may have undergone dimensionality transformation) or both can be input into one or more small fully connected layers (with activation functions such as Sigmoid or Softmax) to generate their respective feature weights. and .

[0106] For example, the feature weights can be generated using the following expression:

[0107]

[0108] in, It is a weight matrix. It's a bias. It's the Sigmoid activation function, and its output is... It can be used as one of the features (e.g.) If the feature weight of one feature is equal to the feature weight of another feature, then the feature weight of the other feature can be... Alternatively, calculate the weights of the two features separately and then normalize them.

[0109] After determining the feature weights corresponding to spatial and spatiotemporal features, the calculated feature weights are used to perform a weighted sum of the spatial and spatiotemporal features (or features after dimensional transformation):

[0110]

[0111] This fusion approach enables the model to automatically adjust its reliance on different modal information based on the data. For example, spatial structure may be more important during periods of stable traffic, while temporal characteristics may be more critical during periods of dynamic change, such as the occurrence or dissipation of traffic congestion.

[0112] (4) Using a cross-attention mechanism, a query is determined based on one of the spatial features and the spatiotemporal features, and a key and value are determined based on the other feature. Attention is calculated based on the determined query, key and value to obtain the spatial and spatiotemporal features enhanced by interaction. The spatial and spatiotemporal features enhanced by interaction are then fused.

[0113] This fusion approach allows for deeper information interaction and mutual enhancement between spatial and spatiotemporal features, rather than simply a weighted combination. It employs a cross-attention mechanism, where features from one modality (spatial or spatiotemporal) act as queries, and features from the other modality (spatiotemporal or spatial) act as keys and values.

[0114] For example, temporal-to-spatial attention enhances spatial features:

[0115] Query: Spatial Features ,in, This is the weight matrix;

[0116] Key: Spatiotemporal feature sequence (Transformer in processing road segments) When dealing with a sparse event sequence, the output representation of each time step within it, i.e., the time-coded information in each event element, is denoted as... After linear transformation ,in, This is the weight matrix;

[0117] Value: The spatiotemporal feature sequence after linear transformation ,in, This is the weight matrix;

[0118] Attention calculation: .in, It is a spatial feature enhanced with spatiotemporal information.

[0119] For example, spatial features are enhanced with spatiotemporal features (Spatial-to-Temporal Attention):

[0120] Query: Spatiotemporal characteristics ,in, This is the weight matrix;

[0121] Key: Spatial features (can be the spatial features themselves, or the road segments to be considered) The spatial feature set of the GNN output of its first- or second-order neighboring road segments. After linear transformation ,in, This is the weight matrix;

[0122] Value: A set of spatial features after a linear transformation ,in, This is the weight matrix;

[0123] Attention calculation: .in, It is a spatiotemporal feature enhanced with spatial information.

[0124] For example, after performing attention calculations, the enhanced interaction features are obtained. and These features can be spliced ​​with the original features (spatial features and spatiotemporal features) or further weighted and fused, without any specific limitations.

[0125] This fusion approach can capture how a specific pattern in one modality depends on a specific aspect of another modality. For example, how the spatial attributes of a road segment (such as a bottleneck segment) amplify or diminish the impact of a specific time pattern (such as the morning rush hour).

[0126] For example, when at least two of the fusion methods in (1) to (4) above are used, the fusion features output by each fusion method can be fused a second time. For example, after splicing, dimensionality reduction and final feature integration can be performed through one or more fully connected layers to obtain the final feature. That is, the characteristics of the road segment, for example:

[0127]

[0128] Step 208: Predict traffic information for future time periods of the road segment to be predicted based on road segment characteristics.

[0129] For example, the fused feature vector Input is fed into one or more prediction layers to predict traffic information for the road segment to be predicted at a future time period (e.g., the next 15 minutes, 30 minutes, or 1 hour).

[0130] Optionally, during the model training phase, due to the lack of sensors in the target prediction area (the area containing the road segment to be predicted), it is difficult to obtain large-scale ground truth labels. In this embodiment of the invention, the following methods can be used to address this: small-label area migration: train the model in an area covered by a small number of sensors or floating car data, and then migrate the model to an unlabeled area for prediction; semi-supervised or self-supervised learning: design pre-training tasks using the characteristics of GPRS data itself, such as predicting the next hop position of the trajectory and mask modeling, to learn a general spatiotemporal representation, and then fine-tune it on a small amount of labeled data; using simulation data, combining traffic simulation software to generate simulated traffic flow and corresponding GPRS call detail records for pre-training; manual annotation or small-scale deployment verification: conduct short-term manual observation or temporarily deploy low-cost sensors on key road segments for model verification and calibration. The loss function used during training can be selected according to the prediction objective, such as using Mean Squared Error (MSE) or Mean Absolute Error (MAE) for flow or speed prediction; and using Cross-Entropy Loss for congestion level prediction. The optimizer used during training can be selected according to actual needs, such as using a suitable optimization algorithm to optimize model parameters.

[0131] The traffic information prediction method provided in this invention determines the driving trajectories of candidate vehicles based on sparse data from mobile communication, identifies the driving trajectory containing trajectory points that meet preset conditions as the target driving trajectory, and then determines the historical data of the road segment to be predicted for a preset historical period. Based on the sparse historical data, the weights of the edges in the road network graph and the event sequence are determined. The spatial features of the road segment to be predicted are output using the understanding of the road network structure by a model for processing graph structure data, and the spatiotemporal features of the road segment to be predicted are output using the processing capability of the sparse temporal events by a model for processing sequences. This fully mines the spatiotemporal dynamic information contained in the sparse data, and then performs prediction after fusing the spatial and spatiotemporal features, thereby achieving accurate prediction of traffic information at low cost. This solution can process and fully utilize widely covered but sparse, low-frequency, and low-precision mobile communication data such as GPRS call detail records, effectively overcoming data quality limitations. By deeply fusing GNN's understanding of road network structure and Transformer model's ability to process sparse long-term events, it can fully mine the spatiotemporal dynamic information contained in sparse data, effectively integrating spatial topology and spatiotemporal dynamic features. This enables accurate prediction of traffic information for specific target road segments in areas lacking sensor coverage (such as suburbs and new towns) over a longer period (e.g., 1 hour), thus providing a new approach to traffic situation awareness that is low-cost, widely covered, and highly effective.The technical solution provided by this invention fully recognizes that the static spatial structure of the road network and the dynamic spatiotemporal events occurring on the road network are two closely related and inseparable aspects. Therefore, it leverages the synergistic potential of combining the two models, rather than simply choosing between two single models. It employs a dedicated "Separate-then-Fuse" architecture, clearly distinguishing between static topology and dynamic events—two physically distinct pieces of information. It does not treat GNN and Transformer as options, nor does it vaguely couple them together. Instead, it explicitly distinguishes between two types of information: the "static road network topology" processed by GNN, and the "dynamic sparse movement events" processed by Transformer. This clear division of labor and explicit physical meaning fully utilizes each... The model's respective advantages make its feature processing in different dimensions more targeted and effective than general methods in related technologies. Furthermore, the solutions designed for sparse mobile data, such as data preprocessing, sparse trajectory reconstruction and map matching, and continuous time coding specifically designed for irregular timestamps, are all designed to overcome the challenges of sparse, low-frequency, and inaccurate positioning of sparse mobile data. When fusing spatial and spatiotemporal features, the model intelligently integrates GNN's deep understanding of "static space" with Transformer's keen capture of "long-term dynamics," achieving a "1+1>2" effect. This fusion is not a simple feature splicing, but a targeted combination of two heterogeneous information sources to generate a more comprehensive basis for judging future traffic conditions.

[0132] Furthermore, the technical solutions provided by the embodiments of this invention have significant application prospects and value in many application scenarios. For example, they can reduce traffic monitoring costs. For traffic management departments, they can significantly reduce the cost of deploying and maintaining expensive sensors in areas such as secondary arterial roads, branch roads, suburbs, and newly built urban areas, and can achieve large-scale traffic situation awareness using existing mobile communication network infrastructure. They can improve traffic management efficiency. The refined road segment-level traffic prediction information provided can support more intelligent traffic signal timing optimization, traffic indication strategy formulation, congestion warning and guidance, and emergency route planning, thereby improving urban traffic operation efficiency. They can empower smart city applications. As an important supplement to the traffic perception layer of smart cities, they can provide data support for urban planning, public transportation scheduling optimization, and commercial site selection analysis (based on pedestrian / vehicle flow heatmaps). They can improve navigation and travel services. Navigation map service providers can integrate this low-cost prediction information to provide users with more accurate real-time traffic conditions and route planning, especially in areas where traditional data sources are insufficient.

[0133] Figure 3 This is a schematic diagram of the structure of a traffic information prediction device according to an embodiment of the present invention. Figure 3As shown, the device includes:

[0134] The spatial feature acquisition module 301 is used to acquire the spatial features corresponding to a preset historical time period of the road segment to be predicted. The spatial features are output by a first model after processing the road network map. The first model is a model for processing graph structure data. The road network map includes nodes for representing road segments and edges between the nodes. The weight of the edges is determined according to the historical data of the preset historical time period corresponding to the road network map. The historical data is traffic historical data used to determine the location information of the target vehicle.

[0135] The spatiotemporal feature acquisition module 302 is used to acquire the spatiotemporal features corresponding to the preset historical time period of the road segment to be predicted. The spatiotemporal features are output by the second model after processing the event sequence. The second model is a model for processing the sequence. The event sequence is determined according to the historical data. The event sequence includes multiple event elements. The event elements include time encoding information and event information of preset type events.

[0136] The feature construction module 303 is used to construct the road segment features of the road segment to be predicted based on the spatial features and the spatiotemporal features;

[0137] The traffic information prediction module 304 is used to predict the traffic information of the road segment to be predicted for future time periods based on the road segment characteristics.

[0138] The traffic information prediction device provided in this embodiment of the invention determines the weights of edges in a road network graph and the event sequence based on historical traffic data. It uses a model for processing graph-structured data to understand the road network structure and outputs the spatial features of the road segment to be predicted. It also uses a model for processing sequences to process sparse temporal events and outputs the spatiotemporal features of the road segment to be predicted. This fully mines the spatiotemporal dynamic information contained in historical traffic data and integrates spatial and spatiotemporal features to achieve accurate prediction of traffic information.

[0139] Optionally, the weight of the edge is determined based on the physical adjacency between road segments, the estimated travel time of the road segments, and the traffic flow correlation between road segments, wherein the estimated travel time and / or the traffic flow correlation are determined based on the historical data.

[0140] Optionally, the historical data is sparse mobile communication-based data used to determine the location information of the target vehicle. The device further includes: a trajectory determination module, used to determine the driving trajectory of the candidate vehicle based on the candidate data, wherein the candidate data is traffic historical data used to determine the location information of the candidate vehicle, and the driving trajectory includes multiple sparse trajectory points, each trajectory point being associated with time information and a driving segment; a target trajectory determination module, used to determine the driving trajectory containing trajectory points that meet preset conditions as the target driving trajectory, wherein the preset conditions are that the time information associated with the trajectory point is within the preset historical time period, and the associated driving segment is the segment to be predicted; a target vehicle determination module, used to determine the candidate vehicle corresponding to the target driving trajectory as the target vehicle; and a historical data determination module, used to determine the candidate data corresponding to the target vehicle as historical data.

[0141] Optionally, the candidate data is data generated by a candidate vehicle communicating with a preset wireless communication object; wherein, the trajectory determination module includes: an object identifier determination unit, used to determine the object identifier of the preset wireless communication object contained in the candidate data; a location determination unit, used to determine the object location and communication signal coverage range of the preset wireless communication object based on the object identifier; and a trajectory determination unit, used to sort the candidate data for each candidate vehicle according to the time information in the candidate data corresponding to the current candidate vehicle to obtain the object connection sequence corresponding to the current candidate vehicle, and based on the communication signal coverage range, use a preset matching method to match the object connection sequence with a preset road network map to obtain the driving trajectory corresponding to the current candidate vehicle, wherein the object connection sequence includes multiple object elements, and the object elements include time information and object location.

[0142] Optionally, based on the coverage area of ​​the communication signal, a preset matching method is used to match the object connection sequence with a preset road network map, including: determining the sparsity index of the object connection sequence; in response to the sparsity index being greater than a preset threshold, interpolating the object connection sequence to obtain an interpolated object connection sequence; and matching the interpolated object connection sequence with the preset road network map using the preset matching method.

[0143] Optionally, determining the sparsity index of the object connection sequence includes: for each pair of object elements formed by two adjacent object elements in the object connection sequence, calculating a first difference in the time information of the current object element pair and a second difference in the object position, and calculating a first product of the first difference and the second difference; calculating the sum of the first products of all object element pairs; calculating a third difference in the time information of the last object element and the first object element and a fourth difference in the object position, and calculating a second product of the third difference and the fourth difference; and determining the sparsity index of the object connection sequence based on the quotient of the sum and the second product.

[0144] Optionally, the time encoding information is determined in the following manner: Fourier time feature encoding is performed on the timestamps corresponding to the preset type events to obtain first encoding information, wherein the timestamps are used to characterize the occurrence time of the corresponding preset type events; a second encoding information is determined based on a preset mapping relationship, wherein the preset mapping relationship is obtained by learning the mapping relationship from timestamps to high-dimensional vectors through a neural network containing trainable parameter matrices of multiple time scales; the time interval between the preset type events with adjacent timestamps is calculated, and the time interval is encoded to obtain third encoding information; and the time encoding information is determined based on the first encoding information, the second encoding information, and the third encoding information.

[0145] Optionally, the feature construction module is used to fuse the spatial features and the spatiotemporal features based on at least one of the following fusion methods to obtain the road segment features of the road segment to be predicted: directly concatenating the spatial features and the spatiotemporal features; performing element-wise operations on the spatial features and the spatiotemporal features, wherein the element-wise operations include element-wise addition or element-wise multiplication; processing the spatial features and the spatiotemporal features based on a gating mechanism to obtain the feature weights corresponding to the spatial features and the spatiotemporal features respectively, and performing weighted fusion of the spatial features and the spatiotemporal features based on the feature weights; employing a cross-attention mechanism to determine the query based on one feature of the spatial features and the spatiotemporal features, and determine the key and value based on the other feature, performing attention calculation based on the determined query, key, and value to obtain the interactively enhanced spatial features and spatiotemporal features, and fusing the interactively enhanced spatial features and spatiotemporal features.

[0146] Optionally, the first model is a graph neural network-based model; and / or, the second model is a self-attention mechanism-based model.

[0147] Optionally, the historical data is call detail records (CDRs) for General Packet Radio Service.

[0148] The traffic information prediction device provided in this embodiment of the invention can execute the traffic information prediction method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0149] Figure 4 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0150] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0151] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0152] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as traffic information prediction methods.

[0153] In some embodiments, the traffic information prediction method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the traffic information prediction method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the traffic information prediction method by any other suitable means (e.g., by means of firmware).

[0154] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0155] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0156] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0157] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0158] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0159] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0160] This invention provides a computer program product, including a computer program that, when executed by a processor, implements the traffic information prediction method provided in the above embodiments.

[0161] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0162] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A traffic information prediction method, characterized in that, include: The spatial features corresponding to a preset historical time period of the road segment to be predicted are obtained. The spatial features are output by a first model after processing the road network map. The first model is a model for processing graph structure data. The road network map includes nodes for representing road segments and edges between the nodes. The weight of the edges is determined according to the historical data of the preset historical time period corresponding to the road network map. The historical data is traffic historical data used to determine the location information of the target vehicle. The spatiotemporal features corresponding to the preset historical time period of the road segment to be predicted are obtained. The spatiotemporal features are output by the second model after processing the event sequence. The second model is a model for processing the sequence. The event sequence is determined according to the historical data. The event sequence includes multiple event elements. The event elements include time encoding information and event information of preset type events. The road segment features to be predicted are constructed based on the spatial features and the spatiotemporal features. Based on the road segment characteristics, traffic information for the future time period of the road segment to be predicted is predicted.

2. The method according to claim 1, characterized in that, The weight of the edge is determined based on the physical adjacency between road segments, the estimated travel time of the road segments, and the traffic flow correlation between road segments, wherein the estimated travel time and / or the traffic flow correlation are determined based on the historical data.

3. The method according to claim 1, characterized in that, The historical data is sparse mobile communication-based data used to determine the location information of the target vehicle. The method further includes: The driving trajectory of the candidate vehicle is determined based on the candidate data, wherein the candidate data is sparse data based on mobile communication used to determine the location information of the candidate vehicle, and the driving trajectory includes multiple sparse trajectory points, each trajectory point being associated with time information and driving segment. The driving trajectory containing trajectory points that meet preset conditions is determined as the target driving trajectory, wherein the preset conditions are that the time information associated with the trajectory points is within the preset historical time period, and the associated driving segment is the segment to be predicted. The candidate vehicles corresponding to the target driving trajectory are identified as the target vehicles; The candidate data corresponding to the target vehicle is determined as historical data.

4. The method according to claim 3, characterized in that, The candidate data is the data generated when the candidate vehicle communicates with a preset wireless communication object; The step of determining the driving trajectory of a candidate vehicle based on candidate data includes: Determine the object identifier of the preset wireless communication object contained in the candidate data; The object location and communication signal coverage range of the preset wireless communication object are determined based on the object identifier; For each candidate vehicle, the time information in the candidate data corresponding to the current candidate vehicle is sorted to obtain the object connection sequence corresponding to the current candidate vehicle. Based on the coverage of the communication signal, the object connection sequence is matched with the preset road network map using a preset matching method to obtain the driving trajectory corresponding to the current candidate vehicle. The object connection sequence includes multiple object elements, and the object elements include time information and object location.

5. The method according to claim 4, characterized in that, Based on the coverage area of ​​the communication signal, a preset matching method is used to match the object connection sequence with a preset road network map, including: Determine the sparsity index of the object join sequence; In response to the sparsity index being greater than a preset threshold, the object connection sequence is interpolated to obtain the interpolated object connection sequence. A preset matching method is used to match the interpolated object connection sequence with a preset road network map.

6. The method according to claim 5, characterized in that, The sparsity index for determining the object connection sequence includes: For each pair of object elements formed by two adjacent object elements in the object connection sequence, calculate the first difference in the time information of the current object element pair and the second difference in the object position, and calculate the first product of the first difference and the second difference; Calculate the sum of the first products of all element pairs of objects; Calculate the third difference in time information between the last object element and the first object element, and the fourth difference in object position; then calculate the second product of the third difference and the fourth difference. The sparsity index of the object connection sequence is determined based on the quotient of the sum and the second product.

7. The method according to claim 1, characterized in that, The time-encoded information is determined in the following way: The timestamps corresponding to the preset type events are Fourier time feature encoded to obtain first encoded information, wherein the timestamps are used to characterize the occurrence time of the corresponding preset type events; The second encoded information of the timestamp corresponding to the preset type event is determined based on the preset mapping relationship, wherein the preset mapping relationship is obtained by learning the mapping relationship from timestamp to high-dimensional vector through a neural network containing trainable parameter matrices of multiple time scales; Calculate the time interval between the preset type events with adjacent timestamps, and encode the time interval to obtain third encoded information; Time encoding information is determined based on the first encoding information, the second encoding information, and the third encoding information.

8. The method according to claim 1, characterized in that, The road segment features to be predicted are constructed based on the spatial features and the spatiotemporal features, including: The spatial features and the spatiotemporal features are fused based on at least one of the following fusion methods to obtain the road segment features of the road segment to be predicted: The spatial features and the spatiotemporal features are directly spliced ​​together; Element-wise operations are performed on the spatial features and the spatiotemporal features, wherein the element-wise operations include element-wise addition or element-wise multiplication; The spatial features and the spatiotemporal features are processed based on a gating mechanism to obtain the feature weights corresponding to the spatial features and the spatiotemporal features respectively. The spatial features and the spatiotemporal features are then weighted and fused based on the feature weights. A cross-attention mechanism is adopted to determine the query based on one of the spatial features and the spatiotemporal features, and to determine the key and value based on the other feature. Attention is calculated based on the determined query, key and value to obtain the interactive enhanced spatial features and spatiotemporal features, and then the interactive enhanced spatial features and spatiotemporal features are fused.

9. The method according to any one of claims 1-8, characterized in that, The first model is a graph neural network-based model; and / or, the second model is a self-attention mechanism-based model.

10. The method according to any one of claims 1-8, characterized in that, The historical data refers to call detail records (CDRs) for general packet radio services.

11. A traffic information prediction device, characterized in that, include: The spatial feature acquisition module is used to acquire the spatial features corresponding to a preset historical time period of the road segment to be predicted. The spatial features are output by the first model after processing the road network map. The first model is a model for processing graph structure data. The road network map includes nodes for representing road segments and edges between the nodes. The weight of the edges is determined according to the historical data of the preset historical time period corresponding to the road network map. The historical data is traffic historical data used to determine the location information of the target vehicle. The spatiotemporal feature acquisition module is used to acquire the spatiotemporal features corresponding to the preset historical time period of the road segment to be predicted. The spatiotemporal features are output by the second model after processing the event sequence. The second model is a model for processing the sequence. The event sequence is determined based on the historical data. The event sequence includes multiple event elements. The event elements include time encoding information and event information of preset type events. The feature construction module is used to construct the road segment features of the road segment to be predicted based on the spatial features and the spatiotemporal features; The traffic information prediction module is used to predict the traffic information of the road segment to be predicted for future time periods based on the road segment characteristics.

12. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the traffic information prediction method according to any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the traffic information prediction method according to any one of claims 1-10.

14. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the traffic information prediction method according to any one of claims 1-10.

Images