Traffic volume prediction method under adverse weather conditions

Abstract

The invention relates to the technical field of intelligent traffic systems, in particular to a traffic volume prediction method under adverse weather conditions, which comprises the following steps: acquiring and preprocessing multi-source heterogeneous data of roads in a target area; the multi-source heterogeneous data comprises meteorological data, historical traffic flow data, real-time traffic sensor data, dynamic road event data and road network topology data; performing space-time alignment on the preprocessed multi-source heterogeneous data, and determining a fusion feature matrix; based on a pre-trained CNN-LSTM-Transformer hybrid prediction model, inputting the fusion feature matrix, and outputting a traffic volume prediction result in a specified time period in the future; and generating a dynamic traffic control instruction set according to the traffic volume prediction result. The problems that in the prior art, the data dimension is single, spatial-temporal feature extraction is insufficient, and prediction control is disjointed are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent transportation systems, and in particular to a method for predicting traffic volume under adverse weather conditions. Background Art

[0002] With the acceleration of urbanization, traffic congestion caused by severe weather has become a difficult problem in urban governance. Traditional traffic volume forecasting methods mainly rely on historical traffic flow data (such as vehicle speed and volume), but have the following shortcomings:

[0003] 1. Data is limited in dimension. Existing models rarely integrate multiple sources of data, including meteorological data, real-time events, and road network topology. Especially in inclement weather, meteorological parameters such as precipitation and visibility are strongly correlated with traffic flow, while dynamic events such as road construction and accidents further exacerbate road network uncertainty. Relying solely on historical traffic flow data will result in predictions that deviate significantly from reality.

[0004] 2. Inadequate spatiotemporal feature extraction. For example, at the spatial level, complex spatial correlations in road network topology (such as the propagation of congestion between adjacent road sections) are not effectively modeled. Traditional methods (such as spatial weight matrices) have difficulty capturing multi-scale spatial dependencies (such as the interaction between directly adjacent road sections and the distant road network). At the temporal level, severe weather can cause traffic flow to exhibit non-stationary characteristics (such as sudden congestion), but mainstream time series models (such as ARIMA and single LSTM) are insufficient in their ability to jointly model long-term features (daily / weekly patterns) and short-term mutations.

[0005] 3. Prediction and control are disconnected. Existing technologies often remain at the prediction stage, lacking a closed-loop mechanism to convert prediction results into dynamic control instructions. For example, when a critical road section is predicted to be overloaded, emergency strategies such as traffic light adjustment and route diversion cannot be triggered in real time, resulting in delayed control responses.

[0006] Attempts to improve existing technologies:

[0007] Some studies have attempted to fuse meteorological data, but they have only simply concatenated features and have not addressed spatiotemporal alignment issues (e.g., differences in sampling frequencies between radar reflectivity maps and traffic flow data);

[0008] Some solutions use graph convolutional networks (GCNs) to process road network topology, but do not introduce a multi-scale neighbor aggregation mechanism, resulting in the loss of long-range spatial dependencies.

[0009] Most prediction models use a single architecture (such as CNN or Transformer), which has limitations in capturing local spatial features (CNN), long-term and short-term temporal dependencies (LSTM), and global interactions (Transformer) of traffic flow.

[0010] Therefore, the present invention provides a method for predicting traffic volume under adverse weather conditions to solve the above problems. Summary of the Invention

[0011] In view of the above situation, in order to overcome the defects of the existing technology, the present invention provides a traffic volume prediction method under adverse weather conditions to solve the problems of single data dimension, insufficient extraction of spatiotemporal features and disconnection between prediction and control in the existing technology.

[0012] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a traffic volume prediction method under adverse weather conditions, the prediction method comprising: obtaining multi-source heterogeneous data of roads in the target area and preprocessing the multi-source heterogeneous data, the multi-source heterogeneous data including meteorological data, historical traffic flow data, real-time traffic sensor data, dynamic road event data and road network topology data; performing spatiotemporal alignment on the preprocessed multi-source heterogeneous data to determine a fusion feature matrix; based on a pre-trained CNN-LSTM-Transformer hybrid prediction model, inputting the fusion feature matrix and outputting a traffic volume prediction result for a specified time period in the future; and generating a dynamic traffic control instruction set based on the traffic volume prediction result.

[0013] A further improvement of the present application is that the meteorological data includes precipitation, visibility, wind speed, road surface temperature, weather type code and radar reflectivity map; the historical traffic flow data includes timestamp, road section ID, average speed, traffic volume and lane occupancy; the real-time traffic sensor stream data includes geomagnetic coil detector data, microwave radar data and electronic police camera data; the dynamic road event data includes road construction plans, traffic accident reports and emergency event data; the road network topology data is a directed graph structure, including latitude and longitude coordinates, number of lanes, slope, and edge attributes include road section length, traffic capacity and real-time impedance weight.

[0014] A further improvement of the present application lies in the preprocessing of multi-source heterogeneous data, including: filling missing data in meteorological data with the spatiotemporal Kriging interpolation method; extracting periodic characteristics and event-related characteristics of historical traffic flow data, the periodic characteristics including daily and weekly Fourier basis expansions, and the event-related characteristics including holiday markers and morning and evening peak markers; performing streaming calculations on real-time traffic sensor stream data using a sliding time window, and using a spatiotemporal joint outlier detection algorithm to clean outliers; determining the road construction and traffic accident geographic fence information contained in dynamic road event data, and calculating the impact radius through a spatiotemporal propagation model; and generating target road segment embedding vectors for road network topology data through a multi-scale graph convolutional network.

[0015] A further improvement of the present application is that the target road segment embedding vector is generated by a multi-scale graph convolutional network, including: constructing an adjacency matrix based on the road network topology data, the adjacency matrix characterizing the spatial connectivity between road segments; performing multi-scale sampling on the adjacency matrix to obtain a set of neighboring road segments of the target road segment within multiple preset neighborhood radii; using a graph convolution operator to separately aggregate the road segment attribute features of neighboring road segments of each scale; splicing or weighted fusion of aggregated features of different scales to generate an embedding vector for the target road segment.

[0016] A further improvement of the present application is that the neighborhood radius preset in the multi-scale sampling includes at least a radius of a directly adjacent road section, a mid-range neighborhood radius, and a long-range neighborhood radius.

[0017] A further improvement of the present application is that the spatiotemporal alignment of the pre-processed multi-source heterogeneous data and the determination of the fusion feature matrix include: unifying data with different sampling frequencies to the same time granularity and aligning the time dimension of the multi-source heterogeneous data using spline interpolation; aggregating discrete data into hexagonal units based on the H3 geographic grid system and aligning the spatial dimensions of the multi-source heterogeneous data; and using a gated multimodal fusion mechanism to perform feature fusion on the multi-source heterogeneous data, the expression is:

[0018] F fusion =g⊙F traffic +(1-g)⊙F weather ,g=σ(W g [F traffic ‖F weather ]+η·b emergency ) (1),

[0019] In expression (1), F traffic represents the traffic feature vector, F weather represents the meteorological feature vector, ‖ represents the vector concatenation operation, W g represents the learnable weight matrix, η represents the meteorological urgency scalar, and η∈[0,1], b emergency represents the emergency bias vector, and σ represents the Sigmoid activation function.

[0020] A further improvement of the present application is that the pre-trained CNN-LSTM-Transformer hybrid prediction model is constructed according to the following structure, including:

[0021] The spatial feature extraction layer uses a multi-scale graph convolutional network, and the calculation expression is:

[0022]

[0023] In expression (2), represents the output features of the lth layer, represents the adjacency matrix, represents the degree matrix, represents the normalization operator, represents the learnable weight matrix;

[0024] The time series modeling layer captures the bidirectional dependency of traffic flow through bidirectional LSTM units, and the hidden state update expression is:

[0025]

[0026] In expression (3), x t represents the input feature at time t, represents the forward hidden state, represents the backward hidden state, represents the forward history state, Represents a backward future state;

[0027] And a multimodal interaction layer whose spatiotemporal features are fused through the Transformer-XL architecture, using relative position encoding to process long sequences:

[0028]

[0029] In expression (4), Represents the relative position encoding between positions i and j, which is used to adjust the position relationship when calculating the attention weight. i represents the target time step, j represents the source time step, and w |i-j| represents the learnable position weight vector, and d represents the feature dimension.

[0030] A further improvement of the present application is that the dynamic traffic control instruction set is generated according to the traffic volume prediction results, including: when the predicted traffic volume exceeds the carrying capacity of the target section, a three-level response mechanism is activated, and the triggering conditions and execution actions of the three-level response mechanism are: when the ratio of the predicted traffic volume to the carrying capacity of the target section is greater than the first warning value and less than or equal to the second warning value, the signal light cycle of the target section is adjusted to the yellow flashing mode; when the ratio of the predicted traffic volume to the carrying capacity of the target section is greater than the second warning value and less than or equal to the third warning value, a suggestion to detour the target section is issued through the variable information board, and the weight of the target section is reallocated; when the ratio of the predicted traffic volume to the carrying capacity of the target section is greater than the third warning value, the target section is forced to be diverted, and the road network impedance matrix of the target section is updated.

[0031] A further improvement of the present application is that the reallocation of the weight of the target road section includes increasing the travel time cost or impedance value of the target road section in the real-time road network impedance matrix; the forced diversion of the target road section includes issuing diversion instructions through the navigation platform, ensuring public transportation priority through the signal control system, issuing detour guidance through variable information boards and linking the electronic police system to implement lane control.

[0032] The beneficial effects of the present invention are: by integrating multi-source heterogeneous data such as meteorological, real-time events, and road network topology, and using a pre-trained CNN-LSTM-Transformer hybrid prediction model, accurate prediction of traffic volume under adverse weather conditions is achieved. This method not only solves the problems of single data dimension and insufficient spatiotemporal feature extraction in existing technologies, but also realizes a closed-loop mechanism of prediction and control by generating a dynamic traffic control instruction set, effectively improving the efficiency of urban traffic management and the ability to cope with severe weather. Specifically, the beneficial effects of the present invention include the following aspects:

[0033] First, by introducing multi-source heterogeneous data, the prediction model's input dimensions are enriched, improving the accuracy and reliability of the prediction results. In severe weather conditions, meteorological parameters and dynamic events have a particularly significant impact on traffic flow. The prediction method of this invention fully considers these factors, resulting in more realistic prediction results.

[0034] Secondly, the present invention adopts a pre-trained CNN-LSTM-Transformer hybrid prediction model, which combines the spatial feature extraction capability of convolutional neural network (CNN), the temporal modeling capability of long short-term memory network (LSTM) and the global interaction capability of Transformer. It can fully capture the spatiotemporal characteristics of traffic flow and improve the performance of the prediction model.

[0035] Furthermore, the present invention achieves spatiotemporal alignment and feature fusion of heterogeneous multi-source data, ensuring the consistency and validity of the input data. By aligning the temporal dimension of data with different sampling frequencies, and aligning the spatial dimension based on a geographic grid system, the present invention can fully leverage the complementary advantages of multi-source data and improve the accuracy of prediction results.

[0036] Finally, this invention achieves a close connection between prediction results and traffic control by generating a dynamic traffic control instruction set. When impending traffic congestion is predicted on a key road section, emergency strategies such as signal light adjustment and route diversion can be triggered in real time, effectively alleviating traffic pressure and improving urban traffic efficiency.

[0037] In summary, the present invention provides a method for predicting traffic volume under adverse weather conditions. This method has significant advantages such as rich data dimensions, sufficient extraction of spatiotemporal features, and a closed loop of prediction and control. It is of great significance for improving the level of urban traffic management and the ability to cope with severe weather. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The figure is a schematic flow chart of a method for predicting traffic volume under adverse weather conditions according to the present invention. DETAILED DESCRIPTION

[0039] The following will describe various embodiments of the present invention in detail with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0040] Based on the above problems, the inventors provide the following solutions:

[0041] First, a multi-source heterogeneous data integration framework was constructed. This framework integrates data from various sources, including meteorological data, historical traffic flow data, real-time traffic sensor data, dynamic road event data, and road network topology data. This data undergoes preprocessing steps, such as missing data filling, periodic feature extraction, outlier cleaning, geofence information determination, and segment embedding vector generation, to ensure data quality and consistency.

[0042] Next, spatiotemporal alignment techniques are employed. This step involves unifying data with different sampling frequencies to the same temporal granularity and aggregating discrete data into hexagonal cells using a geographic grid system, thereby achieving temporal and spatial alignment. Furthermore, a gated multimodal fusion mechanism is used to fuse features from multi-source heterogeneous data to form a fused feature matrix, which serves as input for the subsequent prediction model.

[0043] In terms of the prediction model, this paper uses a pre-trained CNN-LSTM-Transformer hybrid prediction model. This model has three main layers: a spatial feature extraction layer, a temporal modeling layer, and a multimodal interaction layer. The spatial feature extraction layer uses a multi-scale graph convolutional network to capture the complex spatial correlations of road network topology; the temporal modeling layer uses bidirectional LSTM units to capture the bidirectional dependencies of traffic flow; and the multimodal interaction layer uses the Transformer-XL architecture to fuse spatiotemporal features and process long-sequence data.

[0044] Finally, a dynamic traffic control instruction set is generated based on the prediction results. When impending traffic congestion is predicted on a critical road section, emergency strategies such as signal light adjustments and route diversion can be triggered in real time. These strategies rely on a three-level response mechanism, which triggers different actions based on the ratio of predicted traffic volume to the target road section's carrying capacity. These actions include adjusting signal light cycles, issuing detour suggestions, and forcing diversions.

[0045] This invention not only solves the problems of existing technologies, but also enables accurate prediction of traffic volume under adverse weather conditions and establishes a closed-loop mechanism for prediction and control. This is of great significance for improving the efficiency of urban traffic management and the ability to cope with severe weather.

[0046] The technical solution will be described in detail below in conjunction with specific embodiments.

[0047] Example

[0048] refer to Figure 1 , a traffic volume prediction method under adverse weather conditions, the prediction method comprising steps S100-S400:

[0049] S100, obtaining and preprocessing multi-source heterogeneous data on roads within a target area, wherein the multi-source heterogeneous data includes meteorological data, historical traffic flow data, real-time traffic sensor data, dynamic road event data, and road network topology data;

[0050] S200, performing spatiotemporal alignment on the pre-processed multi-source heterogeneous data to determine a fusion feature matrix;

[0051] S300, based on the pre-trained CNN-LSTM-Transformer hybrid prediction model, input the fusion feature matrix and output the traffic volume prediction result for the specified time period in the future;

[0052] S400: Generate a dynamic traffic control instruction set based on the traffic volume prediction result.

[0053] Through the above steps S100-S400, the present invention can achieve accurate prediction of traffic volume under adverse weather conditions and take corresponding traffic control measures based on the prediction results. In the specific implementation process, various types of data are first integrated through a multi-source heterogeneous data integration framework to ensure data quality and consistency. Subsequently, the spatiotemporal alignment technology is used to unify data with different sampling frequencies to the same time granularity, and the geographic grid system is used for spatial alignment to form a fusion feature matrix. Next, the pre-trained CNN-LSTM-Transformer hybrid prediction model is used to process the fusion feature matrix and output the traffic volume prediction results for the specified time period in the future. Finally, a dynamic traffic control instruction set is generated based on the prediction results to trigger emergency strategies such as signal light control and path diversion in real time to alleviate traffic pressure and improve the efficiency of urban traffic. The entire prediction and control process operates in a closed loop, effectively improving the level of urban traffic management and the ability to cope with severe weather.

[0054] Specifically, in step S100, the meteorological data includes precipitation, visibility, wind speed, road surface temperature, weather type code and radar reflectivity map; the historical traffic flow data includes timestamp, road section ID, average speed, traffic volume and lane occupancy; the real-time traffic sensor flow data includes geomagnetic coil detector data, microwave radar data and electronic police camera data; the dynamic road event data includes road construction plans, traffic accident reports and emergency event data; the road network topology data is a directed graph structure, including latitude and longitude coordinates, number of lanes, slope, and edge attributes include road section length, traffic capacity and real-time impedance weight.

[0055] The preprocessing of multi-source heterogeneous data includes the following methods:

[0056] The missing data in the meteorological data were filled using the spatiotemporal kriging interpolation method;

[0057] Extracting periodic features and event-related features from historical traffic flow data, where the periodic features include daily and weekly Fourier basis expansions, and the event-related features include holiday markers and morning and evening peak markers;

[0058] Perform streaming computation on real-time traffic sensor stream data using a sliding time window, and use a spatiotemporal joint outlier detection algorithm to clean outliers.

[0059] Determine the geofence information of road construction and traffic accidents contained in dynamic road event data, and calculate the impact radius through spatiotemporal propagation model;

[0060] For road network topology data, the target road segment embedding vector is generated through a multi-scale graph convolutional network.

[0061] Specifically, the generation of the target road segment embedding vector by the multi-scale graph convolutional network includes steps S101-S104:

[0062] S101, constructing an adjacency matrix based on the road network topology data, wherein the adjacency matrix represents the spatial connectivity between road sections;

[0063] S102, performing multi-scale sampling on the adjacency matrix to obtain a set of neighboring road sections of the target road section within multiple preset neighborhood radii;

[0064] S103, using a graph convolution operator to aggregate the segment attribute features of neighboring segments at each scale;

[0065] S104: Concatenate or weightedly fuse the aggregated features of different scales to generate an embedding vector of the target road segment.

[0066] In step S102 , the neighborhood radius preset in the multi-scale sampling includes at least a radius of a directly adjacent road segment, a mid-range neighborhood radius, and a long-range neighborhood radius.

[0067] In one embodiment of the present application, in step S200, performing spatiotemporal alignment on the pre-processed multi-source heterogeneous data to determine a fusion feature matrix includes the following steps S201-S203:

[0068] S201, unifying data with different sampling frequencies to the same time granularity, and aligning the time dimension of multi-source heterogeneous data using spline interpolation;

[0069] S202. Based on the H3 geographic grid system, discrete data are aggregated into hexagonal units to align the spatial dimensions of multi-source heterogeneous data;

[0070] S203, using the gated multimodal fusion mechanism to perform feature fusion on multi-source heterogeneous data, the expression is:

[0071] F fusion =g⊙F traffic +(1-g)⊙F weather ,g=σ(W g [F traffic ‖F weather ]+η·b emergency ) (1),

[0072] In expression (1), F traffic represents the traffic feature vector, F weather represents the meteorological feature vector, ‖ represents the vector concatenation operation, W g represents the learnable weight matrix, η represents the meteorological urgency scalar, and η∈[0,1], b emergency represents the emergency bias vector, and σ represents the Sigmoid activation function.

[0073] In one embodiment of the present application, in step S300, the pre-trained CNN-LSTM-Transformer hybrid prediction model is constructed according to the following structure, including:

[0074] The spatial feature extraction layer uses a multi-scale graph convolutional network, and the calculation expression is:

[0075]

[0076] In expression (2), represents the output features of the lth layer, represents the adjacency matrix, represents the degree matrix, represents the normalization operator, represents the learnable weight matrix;

[0077] The time series modeling layer captures the bidirectional dependency of traffic flow through bidirectional LSTM units, and the hidden state update expression is:

[0078]

[0079] In expression (3), x t represents the input feature at time t, represents the forward hidden state, represents the backward hidden state, represents the forward history state, Represents a backward future state;

[0080] And a multimodal interaction layer whose spatiotemporal features are fused through the Transformer-XL architecture, using relative position encoding to process long sequences:

[0081]

[0082] In expression (4), Represents the relative position encoding between positions i and j, which is used to adjust the position relationship when calculating the attention weight. i represents the target time step, j represents the source time step, and w |i-j| represents the learnable position weight vector, and d represents the feature dimension.

[0083] In one embodiment of the present application, in step S400, generating a dynamic traffic control instruction set based on the traffic volume prediction result includes:

[0084] When the predicted traffic volume exceeds the target road section's carrying capacity, a three-level response mechanism is activated. The triggering conditions and execution actions of the three-level response mechanism are as follows:

[0085] When the ratio of the predicted traffic volume to the carrying capacity of the target road section is greater than the first warning value and less than or equal to the second warning value, the traffic light cycle of the target road section is adjusted to the yellow flashing mode;

[0086] When the ratio of the predicted traffic volume to the carrying capacity of the target road section is greater than the second warning value and less than or equal to the third warning value, a suggestion to detour to the target road section is issued through the variable information board, and the weight of the target road section is reallocated;

[0087] When the ratio of the predicted traffic volume to the carrying capacity of the target road section is greater than the third warning value, the target road section is forced to be diverted, and the road network impedance matrix of the target road section is updated.

[0088] Specifically, the reallocation of the weight of the target road section includes increasing the travel time cost or impedance value of the target road section in the real-time road network impedance matrix; the forced diversion of the target road section includes issuing diversion instructions through the navigation platform, ensuring public transportation priority through the signal control system, issuing detour guidance through the variable information board, and linking the electronic police system to implement lane control.

[0089] Compared to existing technologies, this method integrates multi-source heterogeneous data, including meteorological data, historical traffic flow data, real-time traffic sensor data, dynamic road event data, and road network topology data, to capture more factors affecting traffic volume, thereby improving prediction accuracy. Furthermore, by employing spatiotemporal alignment technology and a gated multimodal fusion mechanism, this method effectively integrates multi-source heterogeneous data, providing high-quality input for prediction models.

[0090] In terms of prediction models, the pre-trained CNN-LSTM-Transformer hybrid prediction model used in this paper has powerful spatiotemporal feature extraction and long sequence processing capabilities. This model can capture the complex spatial correlations of road network topology, the bidirectional dependencies of traffic flows, and the interactive relationships between multimodal data, thereby achieving accurate predictions of future traffic volumes.

[0091] Finally, based on the dynamic traffic control instruction set generated by the prediction results, the present invention implements a closed-loop prediction and control mechanism. By triggering emergency strategies such as real-time signal light control and route diversion, the present invention can effectively alleviate traffic pressure and improve urban traffic efficiency. Furthermore, the introduction of a three-level response mechanism makes control measures more flexible and targeted, triggering different response actions based on the ratio of predicted traffic volume to the target road section's carrying capacity, thereby better addressing traffic congestion in adverse weather conditions.

[0092] Various embodiments of the systems and techniques described 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), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0093] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0094] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0095] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer 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 can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the 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 acoustic input, voice input, or tactile input).

[0096] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0097] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0098] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not a limitation herein.

[0099] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A method for predicting traffic volume under adverse weather conditions, characterized in that: The prediction method comprises: Acquire and preprocess multi-source heterogeneous data on roads within a target area, wherein the multi-source heterogeneous data includes meteorological data, historical traffic flow data, real-time traffic sensor data, dynamic road event data, and road network topology data; Perform spatiotemporal alignment on the pre-processed multi-source heterogeneous data to determine the fusion feature matrix; Based on the pre-trained CNN-LSTM-Transformer hybrid prediction model, the fused feature matrix is ​​input and the traffic volume prediction result for the specified future period is output; Generate a dynamic traffic control instruction set based on traffic volume prediction results.

2. The method for predicting traffic volume under adverse weather conditions according to claim 1, characterized in that: The meteorological data includes precipitation, visibility, wind speed, road surface temperature, weather type code and radar reflectivity map; the historical traffic flow data includes timestamp, road section ID, average speed, traffic volume and lane occupancy rate; the real-time traffic sensor flow data includes geomagnetic coil detector data, microwave radar data and electronic police camera data; the dynamic road event data includes road construction plans, traffic accident reports and emergency event data; the road network topology data is a directed graph structure, including latitude and longitude coordinates, number of lanes, slope, and edge attributes include road section length, traffic capacity and real-time impedance weight.

3. The method for predicting traffic volume under adverse weather conditions according to claim 2, characterized in that: The preprocessing of multi-source heterogeneous data includes: The missing data in the meteorological data were filled using the spatiotemporal kriging interpolation method; Extracting periodic features and event-related features from historical traffic flow data, where the periodic features include daily and weekly Fourier basis expansions, and the event-related features include holiday markers and morning and evening peak markers; Perform streaming computation on real-time traffic sensor stream data using a sliding time window, and use a spatiotemporal joint outlier detection algorithm to clean outliers. Determine the geofence information of road construction and traffic accidents contained in dynamic road event data, and calculate the impact radius through spatiotemporal propagation model; For road network topology data, the target road segment embedding vector is generated through a multi-scale graph convolutional network.

4. The method for predicting traffic volume under adverse weather conditions according to claim 3, characterized in that: Generating the target road segment embedding vector through the multi-scale graph convolutional network includes: Constructing an adjacency matrix based on the road network topology data, wherein the adjacency matrix represents the spatial connectivity between road sections; Performing multi-scale sampling on the adjacency matrix to obtain a set of neighboring road sections of the target road section within multiple preset neighborhood radii; Use graph convolution operators to aggregate the road segment attribute features of neighboring road segments at each scale; Aggregate features at different scales are concatenated or weightedly fused to generate an embedding vector for the target road segment.

5. The method for predicting traffic volume under adverse weather conditions according to claim 4, characterized in that: The neighborhood radius preset in the multi-scale sampling includes at least a radius of a directly adjacent road segment, a mid-range neighborhood radius, and a long-range neighborhood radius.

6. The method for predicting traffic volume under adverse weather conditions according to claim 4, characterized in that: The step of performing spatiotemporal alignment on the pre-processed multi-source heterogeneous data to determine a fusion feature matrix includes: Unify data with different sampling frequencies to the same time granularity and use spline interpolation to align the time dimensions of multi-source heterogeneous data; Based on the H3 geographic grid system, discrete data are aggregated into hexagonal units, aligning the spatial dimensions of multi-source heterogeneous data; The gated multimodal fusion mechanism is used to fuse features of multi-source heterogeneous data. The expression is: F fusion =g⊙F traffic +(1-g)⊙F weather ,g=σ(Wg[F traffic ‖F weather ]+η·b emergency ) (1), In expression (1), F traffic represents the traffic feature vector, F weather represents the meteorological feature vector, ‖ represents the vector concatenation operation, W g represents the learnable weight matrix, η represents the meteorological urgency scalar, and η∈[0,1], b emergency represents the emergency bias vector, and σ represents the Sigmoid activation function.

7. The method for predicting traffic volume under adverse weather conditions according to claim 6, characterized in that: The pre-trained CNN-LSTM-Transformer hybrid prediction model is constructed according to the following structure, including: The spatial feature extraction layer uses a multi-scale graph convolutional network, and the calculation expression is: In expression (2), represents the output features of the lth layer, represents the adjacency matrix, represents the degree matrix, represents the normalization operator, represents the learnable weight matrix; The time series modeling layer captures the bidirectional dependency of traffic flow through bidirectional LSTM units, and the hidden state update expression is: In expression (3), x t represents the input feature at time t, represents the forward hidden state, represents the backward hidden state, represents the forward history state, Represents a backward future state; And a multimodal interaction layer whose spatiotemporal features are fused through the Transformer-XL architecture, using relative position encoding to process long sequences: In expression (4), Represents the relative position encoding between positions i and j, which is used to adjust the position relationship when calculating the attention weight. i represents the target time step, j represents the source time step, and w |i-j| represents the learnable position weight vector, and d represents the feature dimension.

8. The method for predicting traffic volume under adverse weather conditions according to claim 2, characterized in that: Generating a dynamic traffic control instruction set based on traffic volume prediction results includes: When the predicted traffic volume exceeds the target road section's carrying capacity, a three-level response mechanism is activated. The triggering conditions and execution actions of the three-level response mechanism are as follows: When the ratio of the predicted traffic volume to the carrying capacity of the target road section is greater than the first warning value and less than or equal to the second warning value, the traffic light cycle of the target road section is adjusted to the yellow flashing mode; When the ratio of the predicted traffic volume to the carrying capacity of the target road section is greater than the second warning value and less than or equal to the third warning value, a suggestion to detour to the target road section is issued through the variable information board, and the weight of the target road section is reallocated; When the ratio of the predicted traffic volume to the carrying capacity of the target road section is greater than the third warning value, the target road section is forced to be diverted, and the road network impedance matrix of the target road section is updated.

9. The method for predicting traffic volume under adverse weather conditions according to claim 8, characterized in that: The reallocation of the weight of the target road section includes increasing the travel time cost or impedance value of the target road section in the real-time road network impedance matrix; the forced diversion of the target road section includes issuing diversion instructions through the navigation platform, ensuring public transportation priority through the signal control system, issuing detour guidance through the variable information board, and linking the electronic police system to implement lane control.

Images