Traffic decision-making system and method based on multi-source meteorological data fusion

Abstract

The invention relates to the technical field of traffic management, and discloses a traffic decision-making system and method based on multi-source meteorological data fusion, and the system comprises a data collection module which obtains meteorological station observation data, satellite remote sensing data and meteorological information in a traffic camera image in real time; the preprocessing module is used for carrying out space-time alignment and standardization processing on the meteorological information to generate preprocessed meteorological data with a unified space-time reference; the multi-source fusion module is used for performing cross-modal feature extraction and correlation analysis on the preprocessed meteorological data based on a feature level fusion model of deep learning to generate a fused meteorological feature map; and the decision generation module inputs the fused meteorological characteristic spectrum into a pre-trained traffic decision neural network model, and synchronously outputs a traffic scheduling instruction and a safety management instruction. According to the invention, efficient fusion of multi-source meteorological data can be realized, and the accuracy and effect of traffic scheduling and safety management decisions are improved, so that the traffic scheduling and safety management decisions can better cope with complex and changeable traffic scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic management, and in particular to a traffic decision system and method based on multi-source meteorological data fusion. BACKGROUND

[0002] In the field of modern traffic management, with the continuous progress of meteorological monitoring technology, there are numerous types of meteorological data. Although the existing traffic decision system utilizes some meteorological information, there are still some key problems. On the one hand, different source meteorological data, such as meteorological station observation data, satellite remote sensing data, and meteorological information in traffic camera images, are mostly processed independently, without forming an effective data fusion mechanism, resulting in insufficient and inaccurate information utilization. On the other hand, the existing system is difficult to cope with complex and variable traffic scenarios when converting meteorological data into traffic decisions, as the processing model is relatively simple, which makes the decision effect of traffic scheduling and safety management not ideal enough.

[0003] Therefore, there is an urgent need to provide a technical solution to solve the above problems. SUMMARY

[0004] To solve the above technical problems, the present application provides a traffic decision system and method based on multi-source meteorological data fusion.

[0005] In a first aspect, the present application provides a traffic decision system based on multi-source meteorological data fusion, and the technical solution of the system is as follows:

[0006] A data acquisition module is configured to acquire meteorological station observation data, satellite remote sensing data, and meteorological information in traffic camera images in real time.

[0007] A preprocessing module is configured to perform spatio-temporal alignment and standardization processing on the meteorological information to generate preprocessed meteorological data with a unified spatio-temporal reference.

[0008] A multi-source fusion module is configured to perform cross-modal feature extraction and correlation analysis on the preprocessed meteorological data based on a deep learning feature-level fusion model to generate a fusion meteorological feature map.

[0009] A decision generation module is configured to input the fusion meteorological feature map into a pre-trained traffic decision neural network model to simultaneously output traffic scheduling instructions and safety management instructions. The traffic decision neural network model is trained by historical meteorological data and optimal decision samples under corresponding traffic scenarios to realize end-to-end mapping of meteorological features to scheduling-safety instructions.

[0010] The traffic decision system based on multi-source meteorological data fusion of the present application has the following beneficial effects:

[0011] The system of the application can realize efficient fusion of multi-source meteorological data, improve the accuracy and effect of traffic scheduling and safety management decision, and better cope with complex and changeable traffic scenes.

[0012] On the basis of the above scheme, the traffic decision system based on multi-source meteorological data fusion of the application can be further improved as follows.

[0013] In an optional manner, the data acquisition module is specifically configured to:

[0014] receive the meteorological station observation data, the meteorological station observation data including temperature, humidity, wind speed and visibility measurement values;

[0015] analyze the satellite remote sensing data, the satellite remote sensing data including meteorological element inversion results in visible light bands, infrared bands and microwave bands;

[0016] analyze the traffic camera image, the traffic camera image extracting real-time visibility indicators and precipitation state characteristics through a dynamic target recognition algorithm.

[0017] In an optional manner, the preprocessing module is specifically configured to:

[0018] convert the site coordinates in the meteorological station observation data into target traffic road network coordinates and interpolate to road network grid nodes;

[0019] perform geographic projection transformation on the satellite remote sensing data to align the satellite pixel center with the road network grid center;

[0020] synchronize the time sequences of the meteorological station observation data and the satellite remote sensing data based on the GPS time stamp of the traffic camera image;

[0021] perform dimensionless normalization processing on the meteorological station interpolation data, satellite projection alignment data and traffic camera time synchronization data of the road network grid nodes to generate the preprocessed meteorological data.

[0022] In an optional manner, the feature-level fusion model includes a convolutional neural network, a long short-term memory network and a dynamic graph attention network; and the multi-source fusion module is specifically configured to:

[0023] extract spatial texture features of the satellite projection alignment data using the convolutional neural network;

[0024] build time sequence dependency relationships of the meteorological station interpolation data using the long short-term memory network;

[0025] The dynamic target feature of the traffic camera time synchronization data is established by the dynamic graph attention network, and a cross-modal association rule of the spatial texture feature and the time sequence dependency relationship is established.

[0026] The spatial texture feature, the time sequence dependency relationship, and the cross-modal association rule are fused to generate the fusion weather feature atlas.

[0027] In an optional manner, the step of establishing, by the dynamic graph attention network, the cross-modal association rule of the dynamic target feature of the traffic camera time synchronization data, the spatial texture feature, and the time sequence dependency relationship in the multi-source fusion module further includes:

[0028] A dynamic graph structure is constructed based on the topological structure of the road network grid nodes and the spatiotemporal propagation characteristics of the weather elements; wherein the graph vertex is the road network grid node, and the graph edge is generated by the propagation path of the weather elements between adjacent nodes;

[0029] The comprehensive association weight of the dynamic target feature and the spatial texture feature and the time sequence dependency feature of each road network grid node is calculated;

[0030] The spatial texture feature and the time sequence dependency relationship are aggregated according to the comprehensive association weight to generate the cross-modal association rule.

[0031] In an optional manner, the calculation of the comprehensive association weight adopts the following formula:

[0032]

[0033] In the formula, v i is the i-th dynamic target feature vector, which is derived from the traffic camera time synchronization data; s j is the spatial texture feature vector of the j-th road network grid node; α ij represents the comprehensive association weight of the i-th dynamic target feature vector and the j-th road network grid node; t j is the time sequence dependency feature vector of the j-th road network grid node; W is a trainable weight matrix, and β is a weather influence factor, represents a neighbor node set that has a weather element propagation association with the i-th dynamic target.

[0034] In an optional manner, the step of fusing the spatial texture feature, the time sequence dependency relationship, and the cross-modal association rule in the multi-source fusion module to generate the fusion weather feature atlas further includes:

[0035] The spatial texture feature vector and the time sequence dependency feature vector are tensor spliced to generate a spatiotemporal fusion feature vector.

[0036] perform a weighted aggregation operation on the spatio-temporal fusion feature vector based on the comprehensive correlation weight, to obtain a weighted aggregated feature vector;

[0037] perform a nonlinear transformation on the weighted aggregated feature vector through a multilayer perceptron network, to generate the fusion meteorological feature map.

[0038] In an optional manner, the decision generation module is specifically configured to:

[0039] identify a conduction path from a meteorological risk source to a traffic network key node based on a meteorological element propagation path and a gradient distribution in the fusion meteorological feature map;

[0040] calculate an influence weight of the meteorological risk along the conduction path on the traffic network key node through a spatio-temporal attention mechanism;

[0041] generate a collaborative decision vector of a traffic scheduling instruction branch and a safety management instruction branch based on the influence weight;

[0042] decode the collaborative decision vector to generate the traffic scheduling instruction and the safety management instruction.

[0043] In an optional manner, the expression of the influence weight is:

[0044]

[0045] ηmn= hTmUQnRn+ bσ(hTmUQnRn) where ηmn mn represents the influence weight of the meteorological risk source m on the traffic network key node n; h m is a feature vector of the meteorological risk source m, derived from the fusion meteorological feature map; h n is a topological feature vector of the traffic network key node n; U, Q, and R are trainable weight matrices, b is a bias vector, and σ represents a Sigmoid activation function; represents a meteorological element gradient vector along the meteorological element propagation path from the risk source m to the node n; γ(·) is a meteorological risk conduction coefficient function, w type is a meteorological disaster type weight vector.

[0046] In a second aspect, the present application provides a traffic decision method based on multi-source meteorological data fusion, and the technical scheme of the method is as follows:

[0047] real-time acquisition of meteorological information in meteorological station observation data, satellite remote sensing data, and traffic camera images;

[0048] spatio-temporal alignment and standardization processing of the meteorological information, to generate preprocessed meteorological data with a unified spatio-temporal reference.

[0049] The deep learning-based feature-level fusion model performs cross-modal feature extraction and correlation analysis on the preprocessed meteorological data to generate a fusion meteorological feature map;

[0050] The fusion meteorological feature map is input into a pre-trained traffic decision neural network model to synchronously output traffic scheduling instructions and safety management instructions; the traffic decision neural network model is generated by training historical meteorological data and optimal decision samples under corresponding traffic scenes, realizing end-to-end mapping of meteorological features to scheduling-safety instructions.

[0051] The traffic decision method based on multi-source meteorological data fusion has the following beneficial effects:

[0052] The method can efficiently fuse multi-source meteorological data, improve the accuracy and effect of traffic scheduling and safety management decisions, and better cope with complex and variable traffic scenes.

[0053] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the following specific embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of these drawings.

[0055] The drawings are used to illustrate the embodiments and are not considered as limiting the present application. Moreover, the same reference signs are used to represent the same parts throughout the drawings. In the drawings:

[0056] Figure 1 The structural schematic diagram of an embodiment of a traffic decision system based on multi-source meteorological data fusion of the present application;

[0057] Figure 2 The flowchart of an embodiment of a traffic decision method based on multi-source meteorological data fusion of the present application. DETAILED DESCRIPTION

[0058] Exemplary embodiments of the present application will be described in greater detail below with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it is understood that the present application can be embodied in various forms and should not be limited by the embodiments set forth herein.

[0059] Figure 1 An embodiment of a traffic decision system based on multi-source meteorological data fusion provided by the present application is shown in a structural schematic diagram. As shown in the figure, the system includes: Figure 1

[0060] A data acquisition module 110 is configured to acquire meteorological information in real time from meteorological station observation data, satellite remote sensing data, and traffic camera images.

[0061] The meteorological station observation data refers to physical quantity measurement values collected by fixed meteorological monitoring sites, including real-time quantitative data of temperature, humidity, wind speed, and visibility, for representing the local area meteorological state. The satellite remote sensing data refers to electromagnetic wave reflection or radiation data obtained by earth observation satellites through visible light band, infrared band, and microwave band sensors, and two-dimensional spatial distribution information of cloud cover, surface temperature, and precipitation intensity generated after meteorological element inversion algorithm processing. The traffic camera image refers to a video frame sequence captured in real time by monitoring equipment deployed in the traffic network, and visibility indicators and precipitation state features are extracted therefrom by a dynamic target recognition algorithm, for supplementing near-surface meteorological information. The meteorological information refers to a set of physical parameters related to weather phenomena extracted from meteorological station observation data, satellite remote sensing data, and traffic camera images, including temperature, humidity, wind speed, visibility, cloud cover, and precipitation state.

[0062] A preprocessing module 120 is configured to perform spatio-temporal alignment and standardization processing on the meteorological information, to generate preprocessed meteorological data with a unified spatio-temporal reference.

[0063] The preprocessed meteorological data refers to a standardized data set generated after spatio-temporal alignment and standardization processing of the original meteorological information, with a unified spatio-temporal reference to the target traffic network coordinate system and a synchronous time stamp, and all physical quantities have consistent dimensions.

[0064] A multi-source fusion module 130 is configured to perform cross-modal feature extraction and correlation analysis on the preprocessed meteorological data based on a deep learning feature-level fusion model, to generate a fusion meteorological feature map.

[0065] ​The feature-level fusion model refers to a multi-source data processing architecture based on deep learning, which uses a convolutional neural network, a long short-term memory network, and a dynamic graph attention network to collaboratively realize cross-modal feature extraction and correlation analysis. The fusion meteorological feature map refers to a two-dimensional matrix data structure generated by the feature-level fusion model, wherein each grid cell stores a meteorological risk intensity value after cross-modal correlation, which is used to represent the meteorological threat distribution of the entire road network.

[0066] The decision generation module 140 is configured to input the fusion meteorological feature map into a pre-trained traffic decision neural network model, and synchronously output traffic scheduling instructions and safety management instructions; wherein the traffic decision neural network model is generated by training historical meteorological data and optimal decision samples in corresponding traffic scenes, and realizes end-to-end mapping of meteorological features to scheduling-safety instructions.

[0067] The traffic decision neural network model refers to an end-to-end mapping network generated by training historical meteorological data and optimal decision samples, the input is the fusion meteorological feature map, and the output is a collaborative decision vector of traffic scheduling instructions and safety management instructions. The traffic scheduling instruction refers to a dynamic control strategy generated by the decision neural network model, which includes signal timing adjustment, path induction scheme, and transportation capacity scheduling command, and is used to optimize the road network traffic efficiency. The safety management instruction refers to a risk disposal strategy generated by the decision neural network model, which includes emergency resource deployment, evacuation channel control, and warning area demarcation command, and is used to reduce the probability of traffic accidents caused by meteorological disasters.

[0068] The technical scheme of the embodiment can realize efficient fusion of multi-source meteorological data, improve the accuracy and effect of traffic scheduling and safety management decisions, and better cope with complex and variable traffic scenes.

[0069] In an optional manner, the data acquisition module 110 is specifically configured to:

[0070] Receive the meteorological station observation data, which includes temperature, humidity, wind speed, and visibility measurement values.

[0071] Parse the satellite remote sensing data, which includes meteorological element inversion results in visible light, infrared, and microwave bands.

[0072] Analyze the traffic camera image, which extracts real-time visibility indicators and precipitation state features through a dynamic target recognition algorithm.

[0073] In the above optional manner, the specific functions of the data collection module are further specified, so that it can receive weather station observation data, analyze satellite remote sensing data, and analyze weather information in traffic camera images, thereby ensuring that the collected weather data is comprehensive and targeted, providing more accurate basic information for subsequent processing and decision-making.

[0074] In an optional manner, the preprocessing module 120 is specifically configured to:

[0075] Convert the station coordinates in the weather station observation data into the target traffic road network coordinate system and interpolate to the road network grid nodes.

[0076] Wherein, the station coordinates refer to the geographical position identification of the weather monitoring station in the latitude and longitude coordinate system, which needs to be converted into the target traffic road network plane coordinate system in the preprocessing stage. The road network grid node refers to the grid unit center point of the traffic road network divided according to the preset resolution, which is used as the spatial reference anchor point for weather data interpolation and feature fusion.

[0077] Specifically, the station coordinates in the weather station observation data are converted from the original latitude and longitude coordinate system to the target traffic road network plane rectangular coordinate system, and the discrete station data is interpolated to the road network grid node center position according to the converted coordinate position using the Kriging interpolation algorithm.

[0078] Performing geographic projection transformation on the satellite remote sensing data aligns the satellite pixel center with the road network grid center.

[0079] Specifically, Lambert equal-area conic projection transformation is performed on the original geographic projection coordinate system of the satellite remote sensing data, so that the satellite pixel center point and the road network grid node center point are aligned in space.

[0080] Synchronize the time series of the weather station observation data and the satellite remote sensing data based on the GPS timestamp of the traffic camera image.

[0081] Specifically, the GPS timestamp embedded in the traffic camera image is extracted as a time reference, and the time series of the weather station observation data and the satellite remote sensing data are synchronized to this reference time point by a cubic spline interpolation method.

[0082] Perform dimensionless normalization processing on the weather station interpolation data, satellite projection alignment data, and traffic camera time synchronization data of the road network grid nodes to generate the preprocessed weather data.

[0083] Specifically, the meteorological station interpolation data, the satellite projection alignment data and the traffic camera time synchronization data of the road network grid node position are respectively subjected to dimension normalization processing, wherein the temperature data is normalized to the interval of 0 to 1 by adopting the Celsius temperature scale, the wind speed data is normalized to the interval of 0 to 1 by adopting the unit of meters per second, and the visibility data is normalized to the interval of 0 to 1 by adopting the unit of meters, and finally the pretreatment meteorological data with unified space-time reference is generated.

[0084] In the optional manner described above, the meteorological data of different sources can be further integrated under the unified space-time reference, thereby improving the accuracy and consistency of the data and laying a foundation for subsequent fusion processing.

[0085] In an optional manner, the feature-level fusion model comprises a convolutional neural network, a long short-term memory network and a dynamic graph attention network; and the multi-source fusion module 130 is specifically configured to:

[0086] The spatial texture features of the satellite projection alignment data are extracted by using the convolutional neural network.

[0087] The convolutional neural network refers to a deep learning model for extracting spatial texture features of satellite remote sensing data, and identifies two-dimensional patterns such as cloud cluster morphology and temperature gradient through a convolution kernel scan. The spatial texture features refer to two-dimensional feature vectors extracted from satellite data by the convolutional neural network, representing spatial patterns such as cloud distribution and surface thermal anomaly.

[0088] Specifically, the convolutional neural network is used to process the satellite projection alignment data, and cloud pattern texture features and surface temperature gradient features are extracted through multiple convolution layers and pooling layers to generate spatial texture features (vectors).

[0089] The time series dependency relationship of the meteorological station interpolation data is constructed by using the long short-term memory network.

[0090] The long short-term memory network refers to a recurrent neural network for modeling the time series dependency relationship of meteorological station data, and captures the time series evolution law of temperature, humidity and other parameters through a gating mechanism. The time series dependency relationship refers to a vectorized feature extracted from meteorological station data by the long short-term memory network, representing the dynamic law of temperature, humidity and other parameters changing over time.

[0091] Specifically, the long short-term memory network is used to process the meteorological station interpolation data, and the time series evolution law of temperature and humidity is captured through an input gate, a forget gate and an output gate to construct a time series dependency feature relationship (vector).

[0092] The dynamic target features of the traffic camera time synchronization data are established by using the dynamic graph attention network, and the cross-modal association rules of the spatial texture features and the time series dependency relationship are established.

[0093] The dynamic graph attention network refers to a graph structure neural network constructed based on a road network topology, and the correlation weight of the dynamic target feature of the traffic camera and the spatial / temporal feature is calculated through an attention mechanism. The dynamic target feature refers to a vectorized parameter extracted from the traffic camera video stream, including a vehicle trajectory offset degree, a pedestrian contour blur degree, and other visual indicators reflecting visibility and precipitation intensity. The cross-modal correlation rule refers to a weight matrix generated by the dynamic graph attention network, quantifying the coupling strength of the dynamic target feature and the spatial texture feature and the time sequence dependent feature.

[0094] Specifically, the traffic camera time synchronization data is processed by the dynamic graph attention network, and the attention weight of the dynamic target feature and the spatial texture feature vector and the time sequence dependent feature vector is calculated based on the road network grid node as the graph vertex, and the cross-modal correlation rule (matrix) is generated.

[0095] The spatial texture feature, the time sequence dependent relationship, and the cross-modal correlation rule are fused to generate the fusion meteorological feature map.

[0096] Specifically, the spatial texture feature vector and the time sequence dependent feature vector are tensor spliced to form a spatio-temporal fusion feature vector, the spatio-temporal fusion feature vector is subjected to a weighted aggregation operation based on the cross-modal correlation rule matrix, and then subjected to a nonlinear transformation through a multilayer perceptron network, and finally the fusion meteorological feature map is output.

[0097] In the above optional manner, the components of the feature-level fusion model and the specific working mode of the multi-source fusion module are further described. Through the synergistic effect of the convolutional neural network, the long short-term memory network, and the dynamic graph attention network, the meteorological data features of different modalities can be fully extracted and associated to generate a fusion meteorological feature map, thereby improving the quality and depth of meteorological data fusion.

[0098] In an optional manner, the step of establishing, by the dynamic graph attention network in the multi-source fusion module 130, the cross-modal correlation rule of the dynamic target feature of the traffic camera time synchronization data and the spatial texture feature and the time sequence dependent relationship further includes:

[0099] A dynamic graph structure is constructed based on the topological structure of the road network grid node and the spatio-temporal propagation characteristics of meteorological elements.

[0100] wherein the graph vertex is a road network grid node, and the graph edge is generated by the propagation path of the meteorological element between adjacent nodes. The topological structure refers to the spatial connection relationship between the road network grid nodes, which is determined by the road physical connectivity, and is used to construct the graph edge of the meteorological element propagation. The meteorological element space-time propagation characteristic refers to the diffusion law of parameters such as temperature and humidity between the road network grid nodes, which is determined by the wind speed, wind direction and terrain slope, and is used to generate the dynamic graph edge. The meteorological element refers to the physical parameter affecting traffic safety, including temperature, humidity, wind speed, visibility, precipitation intensity and road surface friction coefficient.

[0101] Specifically, the spatial adjacency topology of the road network grid node is determined according to the physical connection relationship of the target traffic road network, and the diffusion law of the meteorological element between adjacent nodes is generated by combining the wind speed, wind direction and terrain slope parameters, to generate a dynamic graph structure with the road network grid node as the graph vertex and the meteorological element propagation path as the graph edge.

[0102] The comprehensive correlation weight of the dynamic target feature and the spatial texture feature and the time sequence dependent feature of each road network grid node is calculated.

[0103] The spatial texture feature and the time sequence dependent relationship are aggregated according to the comprehensive correlation weight to generate the cross-modal correlation rule.

[0104] Specifically, the spatial texture feature vector and the time sequence dependent feature vector of the road network grid node are executed by a weighted sum operation according to the comprehensive correlation weight, and the cross-modal correlation rule matrix representing the coupling relationship between the dynamic target and the meteorological element space-time feature is aggregated and generated.

[0105] In the above optional manner, the implementation steps of the dynamic graph attention network in the multi-source fusion module are further refined, including constructing a dynamic graph structure, calculating a comprehensive correlation weight, and aggregating features to generate a cross-modal correlation rule, so that the correlation analysis of meteorological data is more accurate and detailed, and the system's grasp of the meteorological element space-time propagation characteristics is enhanced.

[0106] In an optional manner, the calculation of the comprehensive correlation weight uses the following formula:

[0107]

[0108] In the formula, v i is the i th dynamic target feature vector, which is derived from the traffic camera time synchronization data; s j is the spatial texture feature vector of the j th road network grid node; a ij represents the comprehensive correlation weight of the i th dynamic target feature vector and the j th road network grid node; t j is the time sequence dependent feature vector of the j th road network grid node; W is a trainable weight matrix, and β is a meteorological influence factor, denotes a set of neighbor nodes associated with the i-th dynamic target in terms of meteorological element propagation.

[0109] It should be noted that the calculation formula of the comprehensive correlation weight performs linear transformation on the splicing result of the dynamic target feature vector v i , the spatial texture feature vector s j and the time series dependent feature vector t j through the trainable weight matrix W, which can effectively encode the potential correlation between multi-modal features. The transformation result is processed by the LeakyReLU activation function to avoid the loss of negative feature information and alleviate the gradient vanishing problem. The meteorological influence factor β is introduced as an adjustable parameter to dynamically scale the correlation calculation value according to the actual meteorological disaster intensity, enhancing the adaptability of the model to extreme weather. The scaled activation value is normalized by the Softmax function to ensure that the output weight α ij satisfies the probability distribution characteristics, and the denominator is limited to the calculation within the neighbor node set , which conforms to the physical law of meteorological element propagation along the local topology of the road network.

[0110] In the above optional manner, a calculation formula of the comprehensive correlation weight is further provided, which can quantify the correlation degree between the dynamic target features and the road network grid node features, making the correlation analysis of meteorological data more scientific and operable, and further improving the accuracy and reliability of the fused meteorological feature map.

[0111] In an optional manner, the step of generating the fused meteorological feature map by fusing the spatial texture features, the time series dependent relationship and the cross-modal correlation rule in the multi-source fusion module 130 further includes:

[0112] Tensor splicing the spatial texture feature vector and the time series dependent feature vector to generate a spatio-temporal fusion feature vector.

[0113] Among them, the spatio-temporal fusion feature vector refers to a high-dimensional vector generated by tensor splicing the spatial texture feature vector and the time series dependent feature vector, which retains the spatial and temporal dimension information of the original data.

[0114] Performing a weighted aggregation operation on the spatio-temporal fusion feature vector based on the comprehensive correlation weight to obtain a weighted aggregated feature vector.

[0115] Among them, the weighted aggregated feature vector refers to a reduced dimension vector generated by performing a weighted summation operation on the spatio-temporal fusion feature vector according to the comprehensive correlation weight, which strengthens the feature contribution of the high correlation area.

[0116] The weighted aggregated feature vectors are nonlinearly transformed using a multilayer perceptron network to generate the fused meteorological feature map.

[0117] Among them, the multilayer perceptron network refers to a deep learning model composed of fully connected layers and nonlinear activation functions, which is used to map weighted aggregated features into a fused meteorological feature map.

[0118] Among the above optional methods, the generation process of the fused meteorological feature map is further described in detail, including the generation of spatiotemporal fusion feature vectors, weighted aggregation operations, and nonlinear transformation of multilayer perceptron networks. Different features are effectively integrated and transformed to finally obtain the fused meteorological feature map, which improves the depth and breadth of feature fusion and provides richer information support for decision generation.

[0119] In an alternative embodiment, the decision generation module 140 is specifically used for:

[0120] Based on the propagation paths and gradient distribution of meteorological elements in the fused meteorological feature map, the transmission paths from meteorological risk sources to key nodes of the transportation network are identified.

[0121] Among them, the propagation path and gradient distribution of meteorological elements refer to the two-dimensional vector field stored in the fused meteorological feature map, which includes the location of the risk source, the direction of diffusion, and the rate of change of meteorological parameters in adjacent grids. Meteorological risk refers to the probability estimate of traffic accidents caused by extreme weather such as heavy rain and dense fog. Meteorological risk source refers to the grid cell in the fused meteorological feature map whose meteorological risk value exceeds the threshold. Key nodes of the traffic network refer to the grid cells at accident-prone locations such as overpasses, tunnel entrances, and sharp bends. The propagation path refers to the optimal diffusion trajectory from the meteorological risk source to the key node.

[0122] The impact weight of meteorological risk on key nodes of the transportation network along the transmission path is calculated using a spatiotemporal attention mechanism.

[0123] Among them, the impact weight refers to the degree of threat posed by meteorological risk sources to key nodes, which is used to prioritize the generation of instructions.

[0124] Based on the aforementioned influence weights, a collaborative decision vector is generated between the traffic dispatch instruction branch and the safety management instruction branch.

[0125] Among them, the collaborative decision vector refers to the latent space vector generated through the spatiotemporal attention mechanism, which simultaneously encodes the decision constraints of traffic scheduling and safety management, and is separated into specific instructions by the decoder.

[0126] The traffic dispatch instructions and the safety management instructions are generated based on the collaborative decision vector.

[0127] Among the above-mentioned optional methods, the specific workflow of the decision generation module is further clarified. Based on the fusion of meteorological feature maps, the meteorological risk transmission path is identified, the influence weight is calculated and a collaborative decision vector is generated, and finally, traffic dispatch and safety management instructions are decoded and generated. This realizes the efficient conversion from meteorological features to traffic decisions and improves the accuracy and timeliness of traffic decisions.

[0128] In one alternative approach, the expression for the influencing weights is:

[0129]

[0130] In the formula, η mn h represents the weight of the impact of meteorological risk source m on key nodes n of the transportation network; m h is the feature vector of meteorological risk source m, derived from the fused meteorological feature map; n Let be the topological feature vector of the key node n in the transportation network; U, Q, R are trainable weight matrices, b is the bias vector, and σ represents the Sigmoid activation function. Let γ(·) represent the gradient vector of meteorological elements from risk source m to node n along the meteorological element propagation path; γ(·) is the meteorological risk transmission coefficient function. w type This is the weight vector for meteorological disaster types.

[0131] It should be noted that the expression affecting the weights is derived by applying the trainable weight matrices Q and R and the bias vector b to the meteorological risk source feature vector h. m With the topological feature vector h of the key node n Performing a linear transformation and then compressing the data to the [-1,1] interval using the tanh function effectively extracts the spatial semantic associations between risk sources and nodes. An attention score is applied to the compressed result using a trainable vector U, and then the score is mapped to a basic impact probability in the [0,1] interval using the sigmoid function σ, which aligns with the probabilistic characteristics of risk transmission. A meteorological risk transmission coefficient function is then introduced. Explicitly embedding meteorological element gradient vectors Physical information, among which w represents the intensity of change of meteorological parameters along the transmission path. type The risk contribution of different disaster types is weighted, resulting in a final impact weight η. mn It combines data learning with physical interpretability.

[0132] Among the above-mentioned optional methods, an expression for the influence weight is further given. This expression can accurately calculate the degree of influence of meteorological risk sources on key nodes of the transportation network, providing a quantitative basis for the decision generation module, making traffic dispatch instructions and safety management instructions more scientific and reasonable, and enhancing the system's ability to respond to meteorological risks.

[0133] Figure 2 A flowchart of an embodiment of a traffic decision method based on multi-source meteorological data fusion provided by the present application is shown. As shown, it includes the following steps: Figure 2

[0134] S1, real-time acquisition of meteorological station observation data, satellite remote sensing data and meteorological information in traffic camera images;

[0135] S2, spatio-temporal alignment and standardization processing of the meteorological information, to generate pre-processed meteorological data with unified spatio-temporal reference;

[0136] S3, cross-modal feature extraction and correlation analysis of the pre-processed meteorological data based on a deep learning-based feature-level fusion model, to generate a fusion meteorological feature map;

[0137] S4, input of the fusion meteorological feature map into a pre-trained traffic decision neural network model, to synchronously output traffic scheduling instructions and safety management instructions; wherein the traffic decision neural network model is generated by training historical meteorological data and optimal decision samples under corresponding traffic scenarios, to realize end-to-end mapping of meteorological features to scheduling-safety instructions.

[0138] The technical solution of the embodiment can realize efficient fusion of multi-source meteorological data, improve the accuracy and effectiveness of traffic scheduling and safety management decisions, and better cope with complex and variable traffic scenarios.

[0139] In addition, the system provided by the above embodiment, when realizing its functions, is only exemplified by the division of the above-mentioned functional modules. In actual application, the above-mentioned functions can be completed by different functional modules according to needs, that is, the system is divided into different functional modules according to actual conditions to complete all or part of the above-described functions. In addition, the system and method embodiments provided by the above embodiments belong to the same concept, and the specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0140] The above description is only the preferred embodiment of the present application and the explanation of the applied technical principles. Those skilled in the art should understand that the disclosed range of the present application is not limited to the technical solutions formed by the specific combination of the above technical features, and also covers other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the disclosed concept. For example, the above features are replaced with the technical features disclosed in the present application (but not limited to) having similar functions to form technical solutions.

[0141] ​It should be noted that the terms "first", "second", and the like in the description and claims of this application are used for distinguishing between similar objects and are not necessarily used to describe a particular sequential or chronological order. The use of the terms first, second, and the like in the description and claims of this application is to distinguish between analogous objects, not to necessarily describe a particular sequential or chronological order, unless otherwise expressly indicated by the context.

[0142] Although the embodiments of the present application have been shown and described above, it should be understood by those ordinary skilled in the art that the above-mentioned embodiments are exemplary and cannot be construed as limiting the present application, and those ordinary skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A traffic decision-making system based on multi-source meteorological data fusion, characterized in that, The system includes: The data acquisition module is used to acquire meteorological information in real time from meteorological station observation data, satellite remote sensing data, and traffic camera images; The preprocessing module is used to perform spatiotemporal alignment and standardization on the meteorological information to generate preprocessed meteorological data with a unified spatiotemporal reference. The multi-source fusion module is used to perform cross-modal feature extraction and correlation analysis on the preprocessed meteorological data based on a feature-level fusion model of deep learning, and generate a fused meteorological feature map. The decision generation module is used to input the fused meteorological feature map into a pre-trained traffic decision neural network model and simultaneously output traffic dispatch instructions and safety management instructions. The traffic decision neural network model is generated by training with historical meteorological data and optimal decision samples under corresponding traffic scenarios to achieve end-to-end mapping from meteorological features to dispatch and safety instructions.

2. The traffic decision-making system based on multi-source meteorological data fusion according to claim 1, characterized in that, The data acquisition module is specifically used for: Receive the observation data from the weather station, which includes measurements of temperature, humidity, wind speed, and visibility. The satellite remote sensing data is analyzed, and the satellite remote sensing data includes the meteorological element inversion results of the visible light band, infrared band and microwave band; The traffic camera images are analyzed, and real-time visibility indicators and precipitation status characteristics are extracted from the traffic camera images using a dynamic target recognition algorithm.

3. The traffic decision-making system based on multi-source meteorological data fusion according to claim 2, characterized in that, The preprocessing module is specifically used for: The station coordinates in the meteorological station observation data are converted into the target traffic network coordinate system and interpolated to the network grid nodes; Perform a geographic projection transformation on the satellite remote sensing data to align the center of the satellite pixels with the center of the road network grid; The time series of the meteorological station observation data and the satellite remote sensing data are synchronized based on the GPS timestamps of the traffic camera images. The preprocessed meteorological data is generated by performing dimensional normalization processing on the meteorological station interpolation data, satellite projection alignment data, and traffic camera time synchronization data of the road network grid nodes.

4. The traffic decision-making system based on multi-source meteorological data fusion according to claim 3, characterized in that, The feature-level fusion model includes: a convolutional neural network, a long short-term memory network, and a dynamic graph attention network; the multi-source fusion module is specifically used for: The convolutional neural network is used to extract the spatial texture features of the satellite projection alignment data; The time-series dependencies of the weather station interpolation data are constructed using the Long Short-Term Memory network. The dynamic graph attention network is used to establish cross-modal association rules between the dynamic target features, spatial texture features, and time series dependencies of the traffic camera time synchronization data; The fused meteorological feature map is generated by fusing the spatial texture features, the time series dependencies, and the cross-modal association rules.

5. The traffic decision-making system based on multi-source meteorological data fusion according to claim 4, characterized in that, The step in the multi-source fusion module to establish cross-modal association rules between the dynamic target features, spatial texture features, and temporal series dependencies of the traffic camera time-synchronized data through the dynamic graph attention network further includes: A dynamic graph structure is constructed based on the topology of the road network grid nodes and the spatiotemporal propagation characteristics of meteorological elements; wherein, the graph vertices are the road network grid nodes, and the graph edges are generated by the propagation paths of meteorological elements between adjacent nodes; Calculate the combined association weight between the dynamic target features and the spatial texture features and temporal series dependency features of each road network grid node; The spatial texture features and the time series dependency are aggregated based on the comprehensive association weight to generate the cross-modal association rule.

6. The traffic decision-making system based on multi-source meteorological data fusion according to claim 5, characterized in that, The comprehensive correlation weight is calculated using the following formula: In the formula, v i The feature vector of the i-th dynamic target originates from the time-synchronized data of the traffic camera; s j Let α be the spatial texture feature vector of the j-th road network grid node; ij The t represents the combined association weight between the i-th dynamic target feature vector and the j-th road network grid node; j Let j be the time-series dependent feature vector of the j-th road network grid node; W is a trainable weight matrix, and β is a meteorological influence factor. This represents the set of neighboring nodes that are associated with the propagation of meteorological elements with the i-th dynamic target.

7. The traffic decision-making system based on multi-source meteorological data fusion according to claim 6, characterized in that, The step of fusing the spatial texture features, the time series dependencies, and the cross-modal association rules in the multi-source fusion module to generate the fused meteorological feature map further includes: The spatial texture feature vector and the time series dependent feature vector are tensor concatenated to generate a spatiotemporal fusion feature vector. Based on the comprehensive correlation weight, a weighted aggregation operation is performed on the spatiotemporal fusion feature vector to obtain the weighted aggregated feature vector; The weighted aggregated feature vectors are nonlinearly transformed using a multilayer perceptron network to generate the fused meteorological feature map.

8. The traffic decision-making system based on multi-source meteorological data fusion according to claim 7, characterized in that, The decision generation module is specifically used for: Based on the propagation paths and gradient distribution of meteorological elements in the fused meteorological feature map, the transmission paths from meteorological risk sources to key nodes of the transportation network are identified. The impact weight of meteorological risk on key nodes of the transportation network along the transmission path is calculated using a spatiotemporal attention mechanism. Based on the aforementioned influence weights, a collaborative decision vector is generated between the traffic dispatch instruction branch and the safety management instruction branch; The traffic dispatch instructions and the safety management instructions are generated based on the collaborative decision vector.

9. The traffic decision-making system based on multi-source meteorological data fusion according to claim 8, characterized in that, The expression for the influencing weight is: In the formula, η mn h represents the weight of the impact of meteorological risk source m on key nodes n of the transportation network; m h is the feature vector of meteorological risk source m, derived from the fused meteorological feature map; n Let be the topological feature vector of the key node n in the transportation network; U, Q, R are trainable weight matrices, b is the bias vector, and σ represents the Sigmoid activation function. Let γ(·) represent the gradient vector of meteorological elements from risk source m to node n along the meteorological element propagation path; γ(·) is the meteorological risk transmission coefficient function. w type This is the weight vector for meteorological disaster types.

10. A traffic decision-making method based on multi-source meteorological data fusion, characterized in that, The method includes: Real-time acquisition of meteorological information from weather station observation data, satellite remote sensing data, and traffic camera images; The meteorological information is spatiotemporally aligned and standardized to generate preprocessed meteorological data with a unified spatiotemporal reference. Based on a feature-level fusion model using deep learning, cross-modal feature extraction and correlation analysis are performed on the preprocessed meteorological data to generate a fused meteorological feature map. The fused meteorological feature map is input into a pre-trained traffic decision neural network model, which simultaneously outputs traffic dispatch instructions and safety management instructions. The traffic decision neural network model is generated by training with historical meteorological data and optimal decision samples under corresponding traffic scenarios, realizing end-to-end mapping from meteorological features to dispatch and safety instructions.

Images