Traffic state prediction method and device, terminal and medium

Abstract

The invention provides a traffic state prediction method and device, a terminal and a medium, and relates to the technical field of intelligent traffic, and the method comprises the steps: recognizing a key traffic road section as a key node based on the historical traffic state data of each traffic road section; constructing a double-layer traffic network structure according to an identification result; a first traffic network structure in the structure comprises all traffic road sections and a spatial connection relation thereof, and a second traffic network structure only comprises key traffic road sections serving as key nodes and a spatial connection relation thereof; and extracting network-level spatial features and key node spatial features based on a double-layer traffic network structure, guiding fusion of the network-level spatial features and the key node spatial features based on prior knowledge information related to each traffic road section, and predicting the traffic state based on the fused spatial features. According to the method, the key node characteristics in the traffic network can be combined, the traffic road segment priori knowledge is utilized to guide feature fusion, and the traffic state prediction accuracy and stability are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a traffic condition prediction method, device, terminal, and medium. Background Technology

[0002] Currently, existing traffic state prediction methods mainly include statistical analysis-based methods and machine learning and deep learning-based methods. Statistical analysis methods typically rely on linear assumptions or fixed model structures, making it difficult to effectively characterize the nonlinear changes in traffic states within complex traffic networks, thus limiting prediction accuracy. Furthermore, most existing machine learning and deep learning-based methods, when introducing neural network models to jointly model the temporal and spatial characteristics of traffic states, assume that all traffic segments have equal importance within the network, failing to differentiate and model key nodes with significant operational fluctuations and substantial impacts on surrounding segments and overall traffic flow. This results in models that cannot accurately reflect the dominant role of key nodes in the overall traffic state evolution. Moreover, existing schemes that improve prediction accuracy by fusing multiple spatial features or multi-source traffic feature information typically employ a uniform fusion method or fixed weights during feature fusion, ignoring the differences in structural attributes, operational stability, and connectivity among different traffic segments. Due to the lack of effective utilization of prior information about traffic segments, the feature fusion results fail to fully reflect the heterogeneity of the traffic network, thus limiting the predictive performance of the model in complex traffic scenarios.

[0003] Therefore, existing traffic condition prediction methods still have shortcomings in terms of key node characterization and the rationality of feature fusion, and need to be improved and developed. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a traffic state prediction method, device, terminal and medium that can combine the characteristics of key nodes in the traffic network and make full use of prior knowledge of traffic segments for feature fusion, so as to improve the accuracy and stability of traffic state prediction.

[0005] The technical solution adopted by this invention to solve the technical problem is as follows: In a first aspect, the present invention discloses a traffic condition prediction method, wherein the method includes: Historical traffic status data of each traffic segment in the target traffic network is obtained, and based on the historical traffic status data, key traffic segments that are key nodes and non-key traffic segments that are non-key nodes are identified in the target traffic network. A two-layer traffic network structure is constructed based on the key traffic segments that are key nodes and the non-key traffic segments that are non-key nodes. The first traffic network structure includes all traffic segments in the target traffic network and their corresponding spatial connections, while the second traffic network structure only includes the key traffic segments that are key nodes and their corresponding spatial connections. Based on the two-layer traffic network structure, spatial features are extracted from the historical traffic state data to obtain network-level spatial features and key node spatial features. Feature fusion parameters are generated based on the prior knowledge information related to each traffic segment. The network-level spatial features and the key node spatial features are weighted and fused using the feature fusion parameters to obtain fused spatial features, and traffic conditions are predicted based on the fused spatial features.

[0006] Optionally, identifying key traffic segments as key nodes and non-key traffic segments as non-key nodes in the target traffic network based on the historical traffic state data includes: Based on the historical traffic state data, the traffic state fluctuation index corresponding to each traffic segment in the target traffic network is calculated, and the traffic state fluctuation index corresponding to each traffic segment is compared with a preset fluctuation threshold to obtain the corresponding comparison result; If the comparison results show that the traffic state fluctuation index corresponding to the traffic segment is greater than the preset fluctuation threshold, then the traffic segment is determined to be a key traffic segment as a key node. If the comparison results show that the traffic state fluctuation index corresponding to the traffic segment is not greater than the preset fluctuation threshold, then the traffic segment is determined to be a non-critical traffic segment as a non-critical node.

[0007] Optionally, the step of calculating the traffic state fluctuation index corresponding to each traffic segment in the target traffic network based on the historical traffic state data includes: The historical traffic state data of each traffic segment in the target traffic network are organized according to the time dimension to construct a historical traffic state time series that reflects the evolution of the operating state of each traffic segment. Based on the historical traffic state time series, a traffic state fluctuation index is calculated for each traffic segment; the traffic state fluctuation index is used to characterize the degree of change in the operating state of the traffic segment in the historical traffic state time series.

[0008] Optionally, the network-level spatial features are: ; ; in, This represents the network-level spatial features extracted from the first transportation network structure. The order of spatial feature propagation is represented. These are the trainable parameters corresponding to the first traffic network structure. L This is the normalized Laplace matrix corresponding to the first transportation network structure. for The traffic state vector of each traffic segment in the target traffic network at any given time. It is the identity matrix. For degree matrix, This is the adjacency matrix corresponding to the first traffic network structure; Furthermore, the spatial characteristics of the key nodes are as follows: ; ; in, To extract the spatial features of key nodes from the second transportation network structure. These are the trainable parameters corresponding to the second transportation network structure. This is the Laplace matrix corresponding to the second transportation network structure. This is the adjacency matrix corresponding to the second transportation network structure.

[0009] Optionally, the generation of feature fusion parameters based on the acquired prior knowledge information related to each of the traffic segments includes: Based on the acquired prior knowledge information related to each of the traffic segments, fusion weight parameters and bias parameters are generated. The fusion weight parameters are: ; The bias parameter is: ; in, To integrate weight parameters, For bias parameters, This refers to prior knowledge information corresponding to traffic segments. The first trainable parameter corresponding to the fusion weight parameters. This is the second trainable parameter corresponding to the fusion weight parameters. This is the third trainable parameter corresponding to the bias parameter. This is the fourth trainable parameter corresponding to the bias parameter.

[0010] Optionally, the step of weightedly fusing the network-level spatial features and the key node spatial features using the feature fusion parameters to obtain the fused spatial features includes: The network-level spatial features are spliced ​​and combined with the key node spatial features to obtain the spliced ​​and combined spatial features; According to the preset weighted fusion formula, the spliced ​​spatial features are weighted and fused using the fusion weight parameter and the bias parameter to obtain the fused spatial features; The preset weighted fusion formula is: ; in, For the integrated spatial features, The spatial characteristics after splicing and combining. This represents the network-level spatial features extracted from the first transportation network structure. To extract the spatial features of key nodes from the second transportation network structure. To integrate weight parameters, This is the bias parameter.

[0011] Optionally, the prediction of traffic conditions based on the fused spatial features includes: The fused spatial features are input into a pre-trained temporal prediction model to predict traffic conditions, and the traffic condition prediction results of the target traffic network in the future within a preset time period are obtained from the temporal prediction model.

[0012] Secondly, the present invention also discloses a traffic condition prediction device, wherein the device comprises: The data acquisition module is used to acquire historical traffic status data for each traffic segment in the target traffic network; The key node identification module is used to identify key traffic segments that are key nodes and non-key traffic segments that are non-key nodes in the target traffic network based on the historical traffic status data. A traffic network construction module is used to construct a two-layer traffic network structure, including a first traffic network structure and a second traffic network structure, based on key traffic segments that are key nodes and non-key traffic segments that are non-key nodes. The first traffic network structure includes all traffic segments in the target traffic network and their corresponding spatial connections, while the second traffic network structure only includes key traffic segments that are key nodes and their corresponding spatial connections. The spatial feature extraction module is used to extract spatial features from the historical traffic state data based on the two-layer traffic network structure, obtain network-level spatial features and key node spatial features, and generate feature fusion parameters based on the acquired prior knowledge information related to each traffic segment. The feature fusion module is used to perform weighted fusion of the network-level spatial features and the key node spatial features using the feature fusion parameters to obtain fused spatial features. The traffic condition prediction module is used to predict traffic conditions based on the fused spatial features.

[0013] Thirdly, the present invention discloses a terminal, comprising: a memory, a processor, and a traffic state prediction program stored in the memory and executable on the processor, wherein the traffic state prediction program, when executed by the processor, implements the steps of the traffic state prediction method as described above.

[0014] Fourthly, the present invention discloses a computer-readable storage medium storing a computer program that can be executed to implement the steps of the traffic state prediction method as described above.

[0015] This invention provides a traffic state prediction method, device, terminal, and medium. The traffic state prediction method includes: acquiring historical traffic state data of each traffic segment in a target traffic network, and identifying key traffic segments as key nodes and non-key traffic segments as non-key nodes in the target traffic network based on the historical traffic state data; constructing a two-layer traffic network structure including a first traffic network structure and a second traffic network structure based on the key traffic segments as key nodes and the non-key traffic segments as non-key nodes; extracting spatial features from the historical traffic state data based on the two-layer traffic network structure to obtain network-level spatial features and key node spatial features, and generating feature fusion parameters based on the acquired prior knowledge information related to each traffic segment; weightedly fusing the network-level spatial features and the key node spatial features using the feature fusion parameters to obtain fused spatial features, and predicting traffic state based on the fused spatial features. Therefore, this invention identifies key nodes in the target traffic network and constructs a first traffic network structure encompassing all traffic segments and a second traffic network structure containing only key nodes, forming a two-layer traffic network structure. This effectively characterizes the dominant influence of key nodes on the overall operation of the target traffic network, achieving an effective depiction of the spatiotemporal evolution characteristics of traffic states and improving the adaptability of prediction results to complex traffic scenarios. Then, network-level spatial features and key node spatial features are extracted from the first and second traffic network structures respectively, achieving a multi-level characterization of the spatial correlation characteristics of traffic states and enhancing the expressive power of traffic network structure features. Furthermore, prior knowledge information of traffic segments is combined to generate feature fusion parameters for adaptive fusion of network-level spatial features and key node spatial features, avoiding information distortion caused by using uniform fusion weights, thereby improving the accuracy and stability of traffic state prediction. In other words, the technical solution of this application can effectively characterize the dominant role of key nodes and the heterogeneous characteristics between traffic segments by combining the characteristics of key nodes in the traffic network and utilizing prior knowledge of traffic segments for feature fusion, thus improving the accuracy and stability of traffic state prediction. Attached Figure Description

[0016] Figure 1 This is a flowchart of a preferred embodiment of the traffic state prediction method in this invention; Figure 2 This is a schematic diagram of a traffic state prediction method that specifically considers key nodes and guides feature fusion based on prior knowledge, as disclosed in this invention. Figure 3This is a functional principle block diagram of a preferred embodiment of the traffic condition prediction device in this invention; Figure 4 This is a functional principle block diagram of a preferred embodiment of the terminal in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0018] With the continuous acceleration of urbanization, the scale and complexity of transportation systems are constantly increasing, and traffic operation status exhibits obvious spatiotemporal correlation and dynamic uncertainty. As one of the key technologies in intelligent transportation systems, traffic state prediction is widely used in scenarios such as traffic guidance, traffic control, and travel decision support, and its prediction accuracy directly affects the level of traffic management and services.

[0019] Currently, existing traffic state prediction methods mainly include statistical analysis-based methods and machine learning and deep learning-based methods. Statistical analysis methods typically rely on linear assumptions or fixed model structures, making it difficult to effectively characterize the nonlinear changes in traffic states within complex traffic networks, thus limiting prediction accuracy. With the improvement of computing power, deep learning-based traffic state prediction methods have gradually become a research hotspot. By introducing neural network models to jointly model the temporal and spatial characteristics of traffic states, prediction performance has been improved to some extent.

[0020] In existing deep learning methods, traffic networks are typically abstracted as a network structure composed of multiple traffic segments or detection nodes, and spatially relevant features of traffic states are extracted based on this network structure. However, most existing machine learning and deep learning-based methods assume that each traffic segment has equal importance in the network when introducing neural network models to jointly model the temporal and spatial features of traffic states. They fail to differentiate and model key nodes in the traffic network that have large fluctuations in their operational status and have a significant impact on surrounding road segments and overall traffic operation. This makes it difficult for the models to accurately reflect the dominant role of key nodes in the evolution of the overall traffic state.

[0021] Furthermore, existing schemes that improve prediction accuracy by fusing multiple spatial features or multi-source traffic feature information typically employ a uniform fusion method or fixed weights during feature fusion, neglecting the differences in structural attributes, operational stability, and connectivity among different traffic segments. Due to the lack of effective utilization of prior information about traffic segments, the feature fusion results fail to fully reflect the heterogeneous characteristics of the traffic network, thus limiting the model's prediction performance in complex traffic scenarios.

[0022] Therefore, existing traffic state prediction methods suffer from insufficient ability to characterize key nodes, neglect of traffic segment differences during feature fusion, and limited implementation flexibility. To address these issues, this application provides a traffic state prediction scheme that combines the characteristics of key nodes in the traffic network and fully utilizes prior knowledge of traffic segments for feature fusion, thereby improving the accuracy and stability of traffic state prediction.

[0023] Please see Figure 1 , Figure 1 This is a flowchart of the traffic state prediction method in this invention. For example... Figure 1 As shown, the traffic condition prediction method according to an embodiment of the present invention includes: Step S11: Obtain historical traffic status data for each traffic segment in the target traffic network, and identify key traffic segments that are key nodes and non-key traffic segments that are non-key nodes in the target traffic network based on the historical traffic status data.

[0024] In this embodiment, historical traffic status data refers to operational status data over multiple historical time periods, such as indicators that characterize traffic operation status, including operating speed, traffic flow, and occupancy rate.

[0025] In this embodiment, after obtaining historical traffic state data for each traffic segment in the target traffic network, key traffic segments that can serve as key nodes in the target traffic network are identified based on the historical traffic state data. Specifically, traffic state fluctuation indicators corresponding to each traffic segment in the target traffic network are calculated based on the historical traffic state data, and the traffic state fluctuation indicators corresponding to each traffic segment are compared with preset fluctuation thresholds to obtain corresponding comparison results. If the comparison results show that the traffic state fluctuation indicator corresponding to the traffic segment is greater than the preset fluctuation threshold, then the traffic segment is determined to be a key traffic segment serving as a key node. If the comparison results show that the traffic state fluctuation indicator corresponding to the traffic segment is not greater than the preset fluctuation threshold, then the traffic segment is determined to be a non-key traffic segment serving as a non-key node. Understandably, based on the historical traffic state data corresponding to each traffic segment, the traffic state fluctuation index corresponding to each traffic segment in the target traffic network is calculated. Based on the comparison results of the traffic state fluctuation index and the preset fluctuation threshold, the key traffic segments in the target traffic network that can be used as key nodes are identified. That is, traffic segments whose traffic state fluctuation index is greater than the preset fluctuation threshold can be used as key traffic segments of key nodes, and traffic segments whose traffic state fluctuation index is not greater than the preset fluctuation threshold can be used as non-key traffic segments of non-key nodes.

[0026] For example, the degree of change in the operational status of traffic segments within a historical time window can be mapped to a statistical dispersion index, and the operational stability of traffic segments can be measured accordingly. That is, when the traffic state fluctuation index corresponding to a certain traffic segment is greater than a preset fluctuation threshold, the traffic segment is identified as a key node, and the other traffic segments are identified as non-key nodes. In this way, a set of key nodes that have a significant impact on the overall traffic state evolution can be selected from the target traffic network.

[0027] In this embodiment, traffic state fluctuation indices are calculated for each traffic segment in the target traffic network based on historical traffic state data. Specifically, this includes: organizing historical traffic state data for each traffic segment in the target traffic network according to the time dimension to construct a historical traffic state time series reflecting the evolution of the operational state of each traffic segment; calculating traffic state fluctuation indices for each traffic segment in the target traffic network based on the historical traffic state time series; and using these traffic state fluctuation indices to characterize the degree of change in the operational state of a traffic segment within the historical traffic state time series. It can be understood that the historical traffic state data for each traffic segment in the target traffic network is processed hierarchically and analyzed step-by-step to extract key fluctuation information characterizing the temporal changes in traffic state, thereby identifying key nodes in the traffic network. Specifically, historical traffic state data is organized and constructed along the time dimension to reflect the evolution of the operational state of each traffic segment. Then, based on this historical traffic state time series, traffic state fluctuation indices can be calculated from aspects such as the magnitude of traffic state changes, fluctuation frequency, dispersion, and stability characteristics. Finally, the importance of traffic segments is comprehensively evaluated based on the traffic state fluctuation indices to complete the determination of key nodes.

[0028] The historical traffic status time series is as follows: ;in, The length of the historical time window. For the target transportation network within the historical time window length Historical traffic status time series within the region.

[0029] It should be noted that traffic condition fluctuation indicators can be characterized using standard deviation. This is achieved by statistically analyzing the time series of traffic conditions of a road segment within a historical time window, and calculating the corresponding standard deviation to reflect the fluctuation level of traffic conditions for that road segment. The specific calculation method for the standard deviation can be set according to actual application needs. For example, it can be calculated based on different time window lengths, different sample statistical methods, or different weighting strategies. This embodiment does not limit this.

[0030] Step S12: Construct a two-layer traffic network structure including a first traffic network structure and a second traffic network structure based on the key traffic segments that are the key nodes and the non-key traffic segments that are the non-key nodes; the first traffic network structure includes all traffic segments in the target traffic network and their corresponding spatial connections, and the second traffic network structure only includes the key traffic segments that are the key nodes and their corresponding spatial connections.

[0031] It should be noted that the transportation network structure includes the spatial connections between transportation segments. Transportation segments are abstracted as network nodes, and the spatial connections between segments are abstracted as network edges, thus forming the corresponding transportation network structure. Spatial connections can be physical connections or spatial adjacency relationships.

[0032] In this embodiment, a quantitative assessment of the criticality of each traffic segment in the target traffic network is completed. After obtaining the determination results of critical traffic segments as key nodes and non-critical traffic segments as non-critical nodes in the traffic network, a two-layer traffic network structure can be constructed based on the determination results, providing basic support for spatial feature modeling. That is, based on the identification results of key nodes, a two-layer traffic network structure is constructed. This two-layer traffic network structure consists of a first traffic network structure and a second traffic network structure. In the two-layer traffic network structure, the first traffic network structure includes all traffic segments in the target traffic network (critical traffic segments as key nodes and non-critical traffic segments as non-critical nodes). The first traffic network structure, consisting of key traffic segments and their spatial connections, is used to characterize the overall spatial topology of the target traffic network. The second traffic network structure only includes key traffic segments that are key nodes and their corresponding spatial connections. In other words, in the second traffic network structure, the influence of non-key nodes is weakened or shielded. By weakening or shielding the influence of non-key nodes in the spatial feature propagation process, the key nodes are strengthened in the spatial modeling process. For example, the influence of non-key nodes can be weakened or shielded in the following ways: deleting non-key nodes and their associated edges, reducing the connection weights of non-key nodes, or mapping the connection relationships of non-key nodes to a preset low influence value.

[0033] Step S13: Based on the two-layer traffic network structure, spatial features are extracted from the historical traffic state data to obtain network-level spatial features and key node spatial features, and feature fusion parameters are generated based on the prior knowledge information related to each traffic segment.

[0034] In this embodiment, to fully characterize the spatial correlation of traffic states, spatial features are extracted from historical traffic state data based on the first and second traffic network structures in the two-layer traffic network structure. This yields network-level spatial features reflecting the overall spatial correlation of the traffic network, as well as key node spatial features highlighting the spatial influence of key nodes. Specifically, spatial features are extracted from historical traffic state data based on the first traffic network structure to obtain network-level spatial features, which characterize the overall spatial correlation of the traffic network. Spatial features are also extracted from historical traffic state data based on the second traffic network structure to obtain key node spatial features, which highlight the influence of key nodes in spatial feature modeling. This results in two types of spatial feature representations with different emphases.

[0035] It is understandable that by modeling spatial features at different levels of the transportation network structure, the integrity of the overall spatial information of the transportation network can be maintained while further strengthening the role of key nodes in the spatial feature propagation process, providing richer and more distinctive spatial feature representations for subsequent feature fusion and temporal prediction. The spatial feature extraction methods can be one or more of graph-based feature propagation, neighborhood aggregation, or spatial correlation modeling.

[0036] For example, spatial features can be extracted from historical traffic state data based on the first traffic network structure to obtain network-level spatial features, which can be used to characterize the overall spatial correlation between traffic segments in the traffic network. For instance, spectral domain local graph convolution can be used to extract spatial features. That is, based on the Laplacian matrix, spatial feature propagation operators on the graph structure are used to model historical traffic state data to extract network-level spatial features. This allows for spatial propagation and aggregation of traffic state information within a limited neighborhood, effectively characterizing the spatial correlation between different traffic segments in the traffic network.

[0037] Among them, the network-level spatial features are: ; ; in, This represents the network-level spatial features extracted from the first transportation network structure. The order of spatial feature propagation is represented. These are the trainable parameters corresponding to the first traffic network structure. L This is the normalized Laplace matrix corresponding to the first transportation network structure. for The traffic state vector of each traffic segment in the target traffic network at any given time. It is the identity matrix. For degree matrix, This is the adjacency matrix corresponding to the first traffic network structure; Furthermore, to highlight the spatial correlation characteristics between key nodes in the target traffic network, spatial features are extracted from historical traffic state data based on a second traffic network structure to obtain the spatial features of key nodes. This involves using the same or similar spatial feature propagation method as the network-level spatial feature extraction to model the traffic state data on the second traffic network structure. This second traffic network structure is built upon the key node identification results, retaining only key nodes and their corresponding spatial connections, while weakening or masking the influence of non-key nodes during spatial feature propagation. Based on this second traffic network structure, a corresponding adjacency matrix is ​​constructed. and Laplace matrix .

[0038] The key node spatial characteristics are as follows: ; ; in, To extract the spatial features of key nodes from the second transportation network structure. These are the trainable parameters corresponding to the second transportation network structure. This is the Laplace matrix corresponding to the second transportation network structure. This is the adjacency matrix corresponding to the second transportation network structure.

[0039] It should be noted that, since the second traffic network structure focuses more on the mutual influence between key nodes in terms of spatial connectivity, the above spatial feature extraction process can effectively enhance the role of key nodes in the spatial evolution of traffic state, making the obtained key node spatial features more sensitive to key change areas in the traffic network.

[0040] In this embodiment, after obtaining the network-level spatial features and key node spatial features respectively, in order to make full use of the differences in structural attributes and operational characteristics of traffic segments, prior knowledge information related to each traffic segment can be obtained. Based on the prior knowledge information, feature fusion parameters are adaptively generated to adjust the relative contributions of network-level spatial features and key node spatial features, so as to guide the fusion process of network-level spatial features and key node spatial features. The prior knowledge information includes, but is not limited to, historical statistical features of traffic status, topological connectivity of traffic segments, road grade, speed limit information, historical congestion frequency or node importance index, which are used to characterize the differences in structural attributes and operational characteristics of different traffic segments.

[0041] Specifically, fusion weight parameters and bias parameters are generated based on the acquired prior knowledge information related to each traffic segment. It can be understood that, in order to ensure that different traffic segments obtain fusion weights that match their own characteristics during feature fusion, the fusion weight parameters and bias parameters required for feature fusion can be adaptively generated based on prior knowledge information.

[0042] The fusion weight parameters are as follows: ; Furthermore, the bias parameter is: ; in, To integrate weight parameters, For bias parameters, This refers to prior knowledge information corresponding to traffic segments. The first trainable parameter corresponding to the fusion weight parameters. This is the second trainable parameter corresponding to the fusion weight parameters. This is the third trainable parameter corresponding to the bias parameter. This is the fourth trainable parameter corresponding to the bias parameters. These trainable parameters are used to learn the mapping relationship between prior knowledge information and feature fusion parameters.

[0043] Step S14: Use the feature fusion parameters to perform weighted fusion of the network-level spatial features and the key node spatial features to obtain fused spatial features, and predict traffic conditions based on the fused spatial features.

[0044] In this embodiment, after adaptively generating feature fusion parameters based on prior knowledge information to adjust the relative contributions of network-level spatial features and key node spatial features, the feature fusion parameters can be used to perform weighted fusion of network-level spatial features and key node spatial features. It can be understood that by introducing prior knowledge information about traffic segments, adaptive fusion of network-level spatial features and key node spatial features is performed to obtain fused spatial features that simultaneously reflect the overall spatial correlation characteristics of the traffic network and the influence of key nodes. By dynamically adjusting the relative contributions of different spatial features based on prior knowledge information, the information imbalance problem caused by using fixed fusion weights can be avoided, thereby improving the relevance and rationality of spatial feature expression. That is, through a spatial feature fusion method guided by prior knowledge, different traffic segments can obtain fusion weights that match their structural attributes and operational characteristics during feature fusion, thereby avoiding the problem of weakening key node information due to using uniform fusion weights and improving the ability of fused spatial features to characterize the spatial heterogeneity of traffic states.

[0045] Specifically, network-level spatial features are concatenated with key node spatial features to obtain concatenated spatial features. These concatenated features are then weighted and fused according to a preset weighted fusion formula, using fusion weight parameters and bias parameters. In essence, the obtained network-level spatial features are concatenated with key node spatial features to construct the spatial feature input for fusion. For example, concatenating network-level spatial features with key node spatial features... .

[0046] The above-mentioned preset weighted fusion formula is: ; in, The fused spatial features are used to characterize the traffic state spatial representation, which comprehensively considers the overall spatial characteristics of the network and the influence of key nodes under the guidance of prior knowledge. This represents the network-level spatial features extracted from the first transportation network structure. To extract the spatial features of key nodes from the second transportation network structure. This represents the spatial characteristics after splicing and combining, namely the network-level spatial characteristics and key node spatial characteristics after splicing and combining. To integrate weight parameters, This is the bias parameter.

[0047] In this embodiment, after weighted fusion of network-level spatial features and key node spatial features using feature fusion parameters to obtain fused spatial features, traffic state prediction can be performed based on the fused spatial features. Specifically, the fused spatial features are input into a pre-trained temporal prediction model to predict traffic state, and the traffic state prediction result of the target traffic network in the future within a preset time period is obtained from the output of the temporal prediction model. For example, the traffic state prediction result of the target traffic network in one or more future time periods is output by the temporal prediction model.

[0048] The time-series prediction model can be expressed as: ; in, It is the spatial feature after integration. For the target transportation network in the future continuous Traffic condition prediction results within a time step This is a spatiotemporal modeling function used for traffic condition prediction. This refers to prior knowledge information related to each traffic segment in the target transportation network. For the transportation network structure, For traffic sections to be collected. This refers to the spatial connections between traffic segments. Traffic state information related to key nodes is extracted from historical traffic state time series to highlight the impact of key nodes in the evolution of traffic states. This is a time series of historical traffic conditions.

[0049] As can be seen, in this embodiment of the invention, by identifying key nodes in the target traffic network, a first traffic network structure containing all traffic segments and a second traffic network structure containing only key nodes are constructed, forming a two-layer traffic network structure. This effectively characterizes the dominant influence of key nodes on the overall operation of the target traffic network, achieves an effective characterization of the spatiotemporal evolution characteristics of traffic states, and improves the adaptability of prediction results to complex traffic scenarios. Then, network-level spatial features and key node spatial features are extracted from the first and second traffic network structures, respectively, to achieve a multi-level characterization of the spatial correlation characteristics of traffic states, enhance the expressive ability of traffic network structure features, and generate feature fusion parameters for adaptive fusion of network-level spatial features and key node spatial features by combining prior knowledge information of traffic segments. This avoids the information distortion problem caused by using a unified fusion weight, thereby improving the accuracy and stability of traffic state prediction. In other words, the technical solution of this application can effectively characterize the dominant role of key nodes in the traffic network and the heterogeneity characteristics between traffic segments by combining the characteristics of key nodes in the traffic network and using prior knowledge of traffic segments for feature fusion, thereby improving the accuracy and stability of traffic state prediction.

[0050] For example, see Figure 2As shown, by acquiring traffic state data, identifying key nodes, constructing a two-layer traffic network, extracting spatial features, fusing features guided by prior knowledge, and making temporal predictions, the prediction of future traffic conditions can be achieved. That is, traffic state data and traffic network structure information are acquired, and then the traffic state fluctuation index of each traffic segment is calculated based on the traffic state data. Based on the traffic state index, key traffic segments as key nodes are identified, that is, the nodes corresponding to the key traffic segments in the traffic network structure information are determined as key nodes. In a two-layer traffic network structure, the first traffic network structure includes all traffic segments and their spatial connections in the target traffic network, while the second traffic network structure only includes key traffic segments that serve as key nodes and their corresponding spatial connections. Based on the first traffic network structure, spatial features are extracted from traffic state data to obtain network-level spatial features, which are used to characterize the overall spatial correlation of the traffic network. Based on the second traffic network structure, spatial features are extracted from traffic state data to obtain key node spatial features. Then, feature fusion is guided by prior knowledge. Specifically, feature fusion parameters are adaptively generated based on prior knowledge. The network-level spatial features and key node spatial features are weighted and fused using the feature fusion parameters. The fused spatial features are then input into a pre-trained time-series prediction model to predict traffic state, resulting in the traffic state prediction results of the target traffic network for one or more future time periods.

[0051] In one embodiment, such as Figure 3 As shown, based on the above traffic state prediction method, the present invention also provides a traffic state prediction device, comprising: Data acquisition module 11 is used to acquire historical traffic status data of each traffic segment in the target traffic network; The key node identification module 12 is used to identify key traffic segments that are key nodes and non-key traffic segments that are non-key nodes in the target traffic network based on the historical traffic status data. The traffic network construction module 13 is used to construct a two-layer traffic network structure including a first traffic network structure and a second traffic network structure based on the key traffic segments that are the key nodes and the non-key traffic segments that are the non-key nodes; the first traffic network structure includes all traffic segments in the target traffic network and their corresponding spatial connections, and the second traffic network structure only includes the key traffic segments that are the key nodes and their corresponding spatial connections. The spatial feature extraction module 14 is used to extract spatial features from the historical traffic state data based on the two-layer traffic network structure, obtain network-level spatial features and key node spatial features, and generate feature fusion parameters based on the acquired prior knowledge information related to each traffic segment. Feature fusion module 15 is used to perform weighted fusion of the network-level spatial features and the key node spatial features using the feature fusion parameters to obtain fused spatial features; Traffic state prediction module 16 is used to predict traffic state based on the fused spatial features.

[0052] Furthermore, it is worth noting that the working process of the traffic state prediction device provided in this embodiment is the same as that of the traffic state prediction method described above, so it will not be repeated here. For details, please refer to the working process of the traffic state prediction method described above.

[0053] Figure 4 A schematic diagram of the structure of a terminal provided in an embodiment of this application. The terminal may include: The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0054] When the processor 502 executes the program, it implements the traffic state prediction method provided in the above embodiments.

[0055] Furthermore, the terminal also includes: Communication interface 503 is used for communication between memory 501 and processor 502.

[0056] The memory 501 is used to store computer programs that can run on the processor 502.

[0057] Memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0058] If the memory 501, processor 502, and communication interface 503 are implemented independently, they can be interconnected via a bus to communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, only one line is used in the diagram, but this does not imply that there is only one bus or one type of bus.

[0059] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0060] Processor 502 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0061] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the traffic state prediction method described above.

[0062] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein.

[0063] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0064] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can read and execute instructions from and from an instruction execution system, apparatus or device).

[0065] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0066] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A traffic condition prediction method, characterized in that, The method includes: Historical traffic status data of each traffic segment in the target traffic network is obtained, and based on the historical traffic status data, key traffic segments that are key nodes and non-key traffic segments that are non-key nodes are identified in the target traffic network. A two-layer traffic network structure is constructed based on the key traffic segments that are key nodes and the non-key traffic segments that are non-key nodes. The first traffic network structure includes all traffic segments in the target traffic network and their corresponding spatial connections, while the second traffic network structure only includes the key traffic segments that are key nodes and their corresponding spatial connections. Based on the two-layer traffic network structure, spatial features are extracted from the historical traffic state data to obtain network-level spatial features and key node spatial features. Feature fusion parameters are generated based on the prior knowledge information related to each traffic segment. The network-level spatial features and the key node spatial features are weighted and fused using the feature fusion parameters to obtain fused spatial features, and traffic conditions are predicted based on the fused spatial features.

2. The traffic condition prediction method according to claim 1, characterized in that, The process of identifying critical traffic segments as key nodes and non-critical traffic segments as non-key nodes in the target traffic network based on the historical traffic state data includes: Based on the historical traffic state data, the traffic state fluctuation index corresponding to each traffic segment in the target traffic network is calculated, and the traffic state fluctuation index corresponding to each traffic segment is compared with a preset fluctuation threshold to obtain the corresponding comparison result; If the comparison results show that the traffic state fluctuation index corresponding to the traffic segment is greater than the preset fluctuation threshold, then the traffic segment is determined to be a key traffic segment as a key node. If the comparison results show that the traffic state fluctuation index corresponding to the traffic segment is not greater than the preset fluctuation threshold, then the traffic segment is determined to be a non-critical traffic segment as a non-critical node.

3. The traffic condition prediction method according to claim 2, characterized in that, The step of calculating the traffic state fluctuation index corresponding to each traffic segment in the target traffic network based on the historical traffic state data includes: The historical traffic state data of each traffic segment in the target traffic network are organized according to the time dimension to construct a historical traffic state time series that reflects the evolution of the operating state of each traffic segment. Based on the historical traffic state time series, a traffic state fluctuation index is calculated for each traffic segment; the traffic state fluctuation index is used to characterize the degree of change in the operating state of the traffic segment in the historical traffic state time series.

4. The traffic condition prediction method according to claim 1, characterized in that, The network-level spatial features are: ； ； in, This represents the network-level spatial features extracted from the first transportation network structure. The order of spatial feature propagation is represented. These are the trainable parameters corresponding to the first traffic network structure. L This is the normalized Laplace matrix corresponding to the first transportation network structure. for The traffic state vector of each traffic segment in the target traffic network at any given time. It is the identity matrix. For degree matrix, This is the adjacency matrix corresponding to the first traffic network structure; Furthermore, the spatial characteristics of the key nodes are as follows: ； ； in, To extract the spatial features of key nodes from the second transportation network structure. These are the trainable parameters corresponding to the second transportation network structure. This is the Laplace matrix corresponding to the second transportation network structure. This is the adjacency matrix corresponding to the second transportation network structure.

5. The traffic condition prediction method according to claim 4, characterized in that, The generation of feature fusion parameters based on the acquired prior knowledge information related to each of the traffic segments includes: Based on the acquired prior knowledge information related to each of the traffic segments, fusion weight parameters and bias parameters are generated. The fusion weight parameters are: ; The bias parameter is: ; in, To integrate weight parameters, For bias parameters, This refers to prior knowledge information corresponding to traffic segments. The first trainable parameter corresponding to the fusion weight parameters. This is the second trainable parameter corresponding to the fusion weight parameters. This is the third trainable parameter corresponding to the bias parameter. This is the fourth trainable parameter corresponding to the bias parameter.

6. The traffic condition prediction method according to claim 5, characterized in that, The step of using the feature fusion parameters to weight and fuse the network-level spatial features and the key node spatial features to obtain the fused spatial features includes: The network-level spatial features are spliced ​​and combined with the key node spatial features to obtain the spliced ​​and combined spatial features; According to the preset weighted fusion formula, the spliced ​​spatial features are weighted and fused using the fusion weight parameter and the bias parameter to obtain the fused spatial features; The preset weighted fusion formula is: ; in, For the integrated spatial features, The spatial characteristics after splicing and combining. This represents the network-level spatial features extracted from the first transportation network structure. To extract the spatial features of key nodes from the second transportation network structure. To integrate weight parameters, This is the bias parameter.

7. The traffic condition prediction method according to any one of claims 1 to 6, characterized in that, The prediction of traffic conditions based on the fused spatial features includes: The fused spatial features are input into a pre-trained temporal prediction model to predict traffic conditions, and the traffic condition prediction results of the target traffic network in the future within a preset time period are obtained from the temporal prediction model.

8. A traffic condition prediction device, characterized in that, The device includes: The data acquisition module is used to acquire historical traffic status data for each traffic segment in the target traffic network; The key node identification module is used to identify key traffic segments that are key nodes and non-key traffic segments that are non-key nodes in the target traffic network based on the historical traffic status data. A traffic network construction module is used to construct a two-layer traffic network structure, including a first traffic network structure and a second traffic network structure, based on key traffic segments that are key nodes and non-key traffic segments that are non-key nodes. The first traffic network structure includes all traffic segments in the target traffic network and their corresponding spatial connections, while the second traffic network structure only includes key traffic segments that are key nodes and their corresponding spatial connections. The spatial feature extraction module is used to extract spatial features from the historical traffic state data based on the two-layer traffic network structure, obtain network-level spatial features and key node spatial features, and generate feature fusion parameters based on the acquired prior knowledge information related to each traffic segment. The feature fusion module is used to perform weighted fusion of the network-level spatial features and the key node spatial features using the feature fusion parameters to obtain fused spatial features. The traffic condition prediction module is used to predict traffic conditions based on the fused spatial features.

9. A terminal, characterized in that, include: The system includes a memory, a processor, and a traffic state prediction program stored in the memory and executable on the processor, wherein the traffic state prediction program, when executed by the processor, implements the steps of the traffic state prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that can be executed to implement the steps of the traffic state prediction method as described in any one of claims 1 to 7.

Images