Traffic weather information system based on artificial intelligence

Abstract

The invention relates to the technical field of meteorological detection, in particular to a traffic meteorological information system based on artificial intelligence, and the system comprises a node structure modeling module which is based on key road sections, intersections and monitoring points in a traffic area, classifies and integrates relation parameters between adjacent nodes by setting spatial positions and meteorological parameters of the nodes, and obtains a node structure modeling module; and generating a traffic node structure model. According to the invention, the meteorological association graph is deeply analyzed through the graph convolutional network, the meteorological parameter deviation value of the high-risk node is extracted layer by layer, and the risk level is generated by combining deviation comparison, so that the deviation of the meteorological parameters is screened and evaluated layer by layer, and the risk detection is ensured to be accurate and hierarchical. The method further refines the judgment of high-risk nodes through a density clustering algorithm, achieves the dynamic progressive risk level judgment through Gaussian distribution, forms the precise analysis of a deviation value, marks the risk levels of the nodes, and improves the fineness and adaptability of risk assessment.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological detection, and in particular to a traffic meteorological information system based on artificial intelligence. Background Art

[0002] The field of meteorological detection technology aims to achieve a comprehensive understanding and accurate prediction of meteorological conditions through the monitoring and data collection of various meteorological parameters. Through real-time and accurate meteorological data acquisition and analysis, it provides scientific basis and decision-making support for weather forecasting, disaster warning, agricultural production, traffic management, etc., ensuring the improvement of safety and efficiency.

[0003] The main purpose of the AI-based traffic and meteorological information system is to provide real-time information and early warning services to traffic management departments and travelers through real-time monitoring and data analysis of weather changes, road conditions, and traffic flow, thereby optimizing traffic flow, improving road safety, and reducing the incidence of traffic accidents, and achieving accurate prediction and intelligent control of future traffic and meteorological conditions.

[0004] In the existing technology of monitoring and analyzing meteorological parameters and traffic data, there is a lack of hierarchical and detailed dynamic assessment of meteorological risks for specific sections and intersections. The spatial correlation of meteorological parameters is fragmented, making it difficult to establish a continuous risk gradient, which affects the refinement of the assessment. There is a lack of in-depth layer-by-layer screening of deviation values ​​of risk nodes, resulting in a lack of accuracy in detecting abnormal situations. In addition, when dynamic and real-time updating of data cannot be achieved, the assessment results of high-risk nodes lag behind, and early warning responses cannot be made in a timely manner, increasing potential safety risks. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a traffic weather information system based on artificial intelligence.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: A traffic weather information system based on artificial intelligence comprises:

[0007] Node structure modeling module: Based on the key sections, intersections and monitoring points in the traffic area, by setting the spatial location and meteorological parameters of the nodes, the associated road types and meteorological characteristics of each node are classified and generated, and the structural model is formed by integrating the parameters of the adjacent relationship between nodes to establish the traffic node structure model;

[0008] Meteorological correlation graph generation module: based on the traffic node structure model, extracts meteorological characteristics between nodes, sets the correlation strength between nodes, increases the influence of adjacent nodes layer by layer through weighted operation, and obtains a meteorological correlation weighted graph;

[0009] Regional risk index calculation module: Based on the meteorological correlation weighted graph, a graph convolutional network is used to extract the meteorological parameter deviation values ​​of high-risk nodes, and risk level screening is established by calculating and comparing parameter deviations layer by layer. The weighted parameters are called again to obtain the risk level index graph of each node;

[0010] Meteorological risk assessment module: Based on the regional risk index map, a density clustering algorithm is used to call node meteorological parameters, and the deviation value is determined by Gaussian distribution. At the same time, the risk mark and parameter exceeding level of high-risk nodes are determined by dynamic progressive comparison, and a high-risk node level mark table is generated;

[0011] Dynamic model adjustment module: based on the high-risk node level marking table, calling the latest observation data, comparing and adjusting the correlation strength between the latest data and the marked node parameters, reconfiguring the weight influence between nodes, and generating a real-time updated risk weighted model;

[0012] Credible warning screening module: Based on the real-time updated risk weighted model, high confidence nodes are marked according to the confidence threshold, and the screened node levels are recorded in sequence to form a credible warning event list.

[0013] As a further solution of the present invention, the node structure modeling module includes a node position setting submodule, an associated road classification submodule, and a node structure integration submodule, wherein:

[0014] Node location setting submodule: Based on the key sections, intersections and monitoring points in the traffic area, the spatial coordinate information and meteorological parameters of each location are extracted, the coordinate system is standardized, the precise location of each monitoring point is confirmed, and the error offset in the data is adjusted to generate node location features;

[0015] Associated road classification submodule: Based on the node location characteristics, analyze the physical connection relationship of the roads where each node is located, divide the road grade and function type, classify and mark the road type and meteorological characteristics of each node, confirm the associated attributes between roads, and generate road meteorological classification characteristics;

[0016] Node structure integration submodule: based on the road weather classification features, extract the spatial distance parameters between each node and the adjacent nodes, confirm the direct connection relationship between the nodes, integrate the adjacent relationship of all nodes, establish a complete node structure model, and form a traffic node structure model;

[0017] The traffic node structure model includes node location coordinates, associated road types, and meteorological characteristic parameter sets.

[0018] As a further solution of the present invention, the meteorological association graph generation module includes a meteorological feature extraction submodule, a node association strength setting submodule, and a meteorological weighted graph generation submodule, wherein:

[0019] Meteorological feature extraction submodule: based on the traffic node structure model, extracts meteorological feature data between nodes, analyzes meteorological factors corresponding to each node, sorts out the impact of meteorological features on road conditions, forms a meteorological feature set for each node, and generates meteorological features between nodes;

[0020] Node association strength setting submodule: based on the meteorological characteristics between the nodes, calculate the mutual influence strength between the nodes, analyze the association of each pair of nodes, generate a quantized association strength value, organize it into a matrix data of association strength, and generate node association strength parameters;

[0021] Meteorological weighted graph generation submodule: based on the node association strength parameter, weighted calculation of meteorological characteristics of adjacent nodes is performed layer by layer, node influence is superimposed, a meteorologically associated network structure is generated, a weighted relationship graph of all nodes is obtained, and a meteorologically associated weighted graph is generated;

[0022] The meteorological correlation weighted graph includes wind speed weighted value, temperature correlation value and humidity influencing factor.

[0023] As a further solution of the present invention, the regional risk index calculation module includes a high-risk node extraction submodule, a risk level screening submodule, and a node risk index generation submodule, wherein:

[0024] High-risk node extraction submodule: Based on the meteorological correlation weighted graph, a graph convolutional network is used to extract nodes with obvious meteorological parameter fluctuations. By analyzing the meteorological parameter variation range of each node, nodes with large meteorological parameter deviation values ​​are screened, the threshold range of parameter deviation is confirmed, and the data characteristics of high-deviation nodes are identified to form a high-risk node deviation set;

[0025] Risk level screening submodule: based on the high-risk node deviation set, compare the meteorological parameter deviation values ​​of the nodes layer by layer, classify the deviation data of each node according to the risk level standard, divide and mark the risk level of each node, and generate a risk level marking set;

[0026] Node risk index generation submodule: Based on the risk level tag set and combined with the weight parameters of the meteorological association weighted graph, the marked risk level is numerically calculated, the risk index value of each node is synthesized, the quantitative expression of the node risk is established, and the node risk level index graph is obtained;

[0027] Among them, the regional risk index map includes the cumulative value of risk deviation, node risk level, and regional risk weight distribution.

[0028] As a further solution of the present invention, the graph convolutional network is based on the formula:

[0029]

[0030] Where: H (l+1) is the feature representation of node i in the l+1 layer, σ is the activation function, is the feature representation of neighbor node j at layer l, W (l) is the weight matrix of the lth layer, α represents the weight coefficient related to the fluctuation deviation of meteorological parameters, ΔP i represents the meteorological parameter deviation value of node i, β represents the weight coefficient of the influence of neighboring nodes, C j represents the fluctuation characteristics of the meteorological parameters of neighbor node j, γ represents the weight coefficient of the complexity of the node meteorological environment, M i Represents the complexity factor of the meteorological environment where node i is located.

[0031] As a further scheme of the present invention, the threshold range of the parameter deviation includes an upper limit of temperature deviation, an upper limit of humidity deviation and a lower limit of wind speed deviation, which are used to identify nodes with abnormal fluctuations in meteorological parameters. The risk level standard specifically includes a low-risk deviation range, a medium-risk deviation range and a high-risk deviation range, which are used to hierarchically classify the meteorological parameter deviation data of each node.

[0032] As a further solution of the present invention, the meteorological risk assessment module includes a meteorological parameter calling submodule, a risk deviation judgment submodule, and a high-risk node marking submodule, wherein:

[0033] Meteorological parameter calling submodule: Based on the node risk level index graph, a density clustering algorithm is used to extract the real-time meteorological parameter data of each node, and the current deviation value of each node is determined by comparing it with the benchmark range data, and the deviation value set of each node is calculated to form a node meteorological deviation value set;

[0034] Risk deviation judgment submodule: based on the node meteorological deviation value set, compare the deviation value range of each node, calculate and determine the degree of deviation exceeding the benchmark, divide the deviation level and mark the risk level, and generate a deviation level judgment set;

[0035] High-risk node marking submodule: Based on the deviation level determination set, identify the high-risk mark of each node, record the risk level and parameter exceeding level, classify the node as high risk, and form a high-risk node level marking table including risk marks and levels;

[0036] The high-risk node level marking table includes risk node marking, risk level label, and parameter out-of-range category.

[0037] As a further solution of the present invention, the density clustering algorithm is according to the formula:

[0038]

[0039] in: represents the adjusted deviation distance of node i, k represents the number of selected neighboring nodes, P i represents the real-time meteorological parameter value vector of node i, P j represents the meteorological parameter value vector of node j adjacent to node i, ∥P i -P j ∥ represents the Euclidean distance of the difference in meteorological parameters between nodes i and j, T i The current temperature value of node i, R i is the rainfall data of node i, V i is the wind speed value of node i, B i is the air pressure value of node i, w 1 、w 2 、w 3 、w 4 is the weight coefficient of each meteorological parameter.

[0040] As a further solution of the present invention, the dynamic model adjustment module includes a data association adjustment submodule, a node weight configuration submodule, and a risk model update submodule, wherein:

[0041] Data association adjustment submodule: based on the high-risk node level marking table, call the latest observation data, extract the latest observation parameters of the marked node, calculate the deviation value between the latest data and the historical marking, confirm the changes within the deviation range by comparing item by item, and adjust the association strength between the marked node and other nodes to generate an updated association strength set;

[0042] Node weight configuration submodule: Based on the updated association strength set, adjust the weight distribution between nodes, extract the latest association strength value of each node, analyze the distribution ratio of association strength in the network, and reconfigure the weight influence between nodes to ensure that the weight conforms to the latest data relationship, and obtain the node weight configuration table;

[0043] Risk model update submodule: Based on the node weight configuration table, the latest node weights and association relationships are integrated, and weighted calculation of node weights is performed. The latest weighted information between nodes is integrated into the overall network model, and a weighted influence model between current nodes is generated to obtain a real-time risk weighted model.

[0044] The real-time updated risk weighted model includes the node weight adjustment value, the updated risk intensity, and the inter-node correction coefficient.

[0045] As a further solution of the present invention, the credible warning screening module includes a confidence screening submodule, a node level sorting submodule, and a warning event recording submodule, wherein:

[0046] Confidence screening submodule: based on the real-time risk weighted model, set the confidence screening threshold, compare the confidence values ​​of each node, screen out nodes that meet the high confidence conditions, mark the high confidence nodes, and generate a high confidence node set;

[0047] Node ranking submodule: based on the high confidence node set, the nodes are ranked according to their risk levels, the risk level values ​​of the high confidence nodes are extracted, the nodes are arranged in descending order, a ranking list of the nodes is constructed, and a node ranking table is obtained;

[0048] Warning event recording submodule: Based on the node ranking table, the sorted high-confidence nodes are recorded one by one, and the risk level and marking information of each node are sorted into warning event records to form a credible warning event list including the node risk level;

[0049] The credible warning event list includes credible warning node numbers, warning level identifiers, and confidence threshold classifications.

[0050] Compared with the prior art, the advantages and positive effects of the present invention are:

[0051] 1. In the present invention, the meteorological correlation graph is deeply analyzed through the graph convolution network, the meteorological parameter deviation values ​​of high-risk nodes are extracted layer by layer, and the risk level is generated by combining the deviation comparison, so that the deviation of the meteorological parameters is screened and evaluated layer by layer to ensure that the risk detection is accurate and hierarchical;

[0052] 2. In the present invention, the determination of high-risk nodes is further refined through the density clustering algorithm, and the dynamic progressive risk level determination is realized through the Gaussian distribution, so as to form an accurate analysis of the deviation value, mark the risk level of the node, and improve the precision and adaptability of risk assessment;

[0053] 3. In the present invention, high-confidence nodes are screened out through confidence thresholds and a credible warning list is generated, which improves the accuracy and real-time performance of meteorological risk assessment and makes the prediction more practical and reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a system flow chart of the present invention;

[0055] Figure 2 It is a schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0057] Embodiment 1

[0058] See also Figure 1 The present invention provides a technical solution: a traffic weather information system based on artificial intelligence comprises:

[0059] Node structure modeling module: Based on the key sections, intersections and monitoring points in the traffic area, by setting the spatial location and meteorological parameters of the nodes, the associated road types and meteorological characteristics of each node are classified and generated, and the structural model is formed by integrating the parameters of the adjacent relationship between nodes to establish the traffic node structure model;

[0060] Meteorological correlation map generation module: Based on the traffic node structure model, the meteorological characteristics between nodes are extracted, the correlation strength between nodes is set, and the influence of adjacent nodes is increased layer by layer through weighted operation to obtain a meteorological correlation weighted map;

[0061] Regional risk index calculation module: Based on the meteorological correlation weighted graph, the graph convolutional network is used to extract the meteorological parameter deviation values ​​of high-risk nodes. By calculating and comparing the parameter deviations layer by layer, risk level screening is established, and the weighted parameters are called again to obtain the risk level index graph of each node;

[0062] Meteorological risk assessment module: Based on the regional risk index map, the density clustering algorithm is used to call the node meteorological parameters, and the deviation value is determined through Gaussian distribution. At the same time, the risk mark and parameter exceeding level of high-risk nodes are determined through dynamic progressive comparison, and a high-risk node level mark table is generated;

[0063] Dynamic model adjustment module: Based on the high-risk node level marking table, call the latest observation data, compare and adjust the correlation strength between the latest data and the marked node parameters, reconfigure the weight influence between nodes, and generate a real-time updated risk weighted model;

[0064] Credible warning screening module: Based on the real-time updated risk-weighted model, high-confidence nodes are marked according to the confidence threshold, and the screened node levels are recorded in sequence to form a credible warning event list.

[0065] See also Figure 2 The node structure modeling module includes a node position setting submodule, an associated road classification submodule, and a node structure integration submodule, among which:

[0066] Node location setting submodule: Based on the key sections, intersections and monitoring points in the traffic area, the spatial coordinate information and meteorological parameters of each location are extracted, the coordinate system is standardized, the precise location of each monitoring point is confirmed, and the error offset in the data is adjusted to generate node location features;

[0067] Associated road classification submodule: Based on the node location characteristics, analyze the physical connection relationship of the roads where each node is located, divide the road grade and functional type, classify and mark the road type and meteorological characteristics of each node, confirm the associated attributes between roads, and generate road meteorological classification characteristics;

[0068] Node structure integration submodule: Based on the road meteorological classification characteristics, the spatial distance parameters between each node and the adjacent nodes are extracted, the direct connection relationship between the nodes is confirmed, the adjacent relationship of all nodes is integrated, and a complete node structure model is established to form a traffic node structure model;

[0069] Node location setting submodule: Based on the key sections, intersections and monitoring points in the traffic area, the spatial coordinate information and meteorological parameters of each location are extracted, the spatial coordinates are standardized through the three-dimensional coordinate conversion method, the Hough transform algorithm is used to calculate the exact position of each monitoring point in the specified coordinate system, and the least squares method is used to adjust the error offset of the coordinate information. By setting the threshold error limit of no more than 0.1 meters, the data offset is minimized and the node location features are generated;

[0070] Associated road classification submodule: Based on the node location characteristics, the graph connectivity analysis algorithm is used to identify the physical connection relationship of each node, and the road grade is distinguished according to the road physical connection parameters. The hierarchical labeling method is used to divide the road function type of each node, and the pre-set road grade coding table in the database is called to classify and code the road type of each node as main road, secondary road, and branch road. The meteorological parameters such as wind speed, temperature, and humidity of each road are extracted through the meteorological feature standardization algorithm, and the minimum and maximum normalization method is used to standardize the meteorological parameters to the range of 0 to 1 to generate the road meteorological classification characteristics of each node;

[0071] Node structure integration submodule: Based on the road weather classification characteristics, the adjacency matrix method is used to extract the spatial distance parameters between each node and the adjacent nodes. The physical distance between nodes is calculated using the Euclidean distance formula. The threshold is set to be within 3 kilometers for direct connection. The connection relationship of the nodes is stored using a sparse matrix. All nodes are traversed in turn and the direct connection relationship between adjacent nodes is recorded. The adjacency relationship of each node is summarized into a unified data table through a data integration algorithm to generate a complete node structure model and form a traffic node structure model.

[0072] Among them, the traffic node structure model includes node location coordinates, associated road types, and meteorological characteristic parameter sets.

[0073] See also Figure 2 The meteorological correlation graph generation module includes a meteorological feature extraction submodule, a node correlation strength setting submodule, and a meteorological weighted graph generation submodule, wherein:

[0074] Meteorological feature extraction submodule: Based on the traffic node structure model, extract the meteorological feature data between nodes, analyze the meteorological factors corresponding to each node, sort out the impact of meteorological features on road conditions, form a meteorological feature set for each node, and generate meteorological features between nodes;

[0075] Node association strength setting submodule: Based on the meteorological characteristics between nodes, calculate the mutual influence strength between nodes, analyze the correlation of each pair of nodes, generate quantitative association strength values, organize them into matrix data of association strength, and generate node association strength parameters;

[0076] Meteorological weighted graph generation submodule: Based on the node association strength parameters, the meteorological characteristics of adjacent nodes are weighted and calculated layer by layer, the node influence is superimposed, the network structure of meteorological association is generated, the weighted relationship graph of all nodes is obtained, and the meteorological association weighted graph is generated;

[0077] Meteorological feature extraction submodule: Based on the traffic node structure model, the meteorological feature data between nodes are extracted, and the characteristic dimension of the meteorological data is reduced by principal component analysis. By setting the number of principal components with a variance contribution rate of 90%, the main meteorological features of each node are obtained and redundant dimensions are eliminated. The meteorological feature data of each node are integrated according to wind speed, rainfall, and temperature to generate a meteorological feature vector. The meteorological feature data are standardized in the range of 0 to 1 using the normalization method, and the meteorological data values ​​are mapped to a standard scale to form a meteorological feature set for each node, and the meteorological features between nodes are generated;

[0078] Node association strength setting submodule: Based on the meteorological characteristics between nodes, the cosine similarity algorithm is used to calculate the similarity of the meteorological feature vectors of each pair of nodes. The mutual influence intensity of the meteorological characteristics between nodes is calculated through the similarity formula. The similarity above the threshold of 0.8 is set as a strong association. The cosine similarity is converted into an intensity coefficient to generate a quantized association strength value. The intensity coefficient of each node pair is sorted into a matrix format to generate matrix data of the association strength and generate node association strength parameters.

[0079] Meteorological weighted graph generation submodule: Based on the node association strength parameter, the meteorological characteristics of adjacent nodes are weightedly calculated layer by layer, and the weighted average algorithm is used for weighted calculation. The weight coefficient of each layer of adjacent nodes is set to decay to 0.5 times that of the previous layer. By superimposing the influence value of the node on the adjacent node layer by layer, the weighted influence between all nodes is organized into a network structure, and the weighted relationship graph of all nodes is obtained to generate a meteorological association weighted graph;

[0080] Among them, the meteorological correlation weighted map includes wind speed weighted value, temperature correlation value, and humidity influencing factor.

[0081] See also Figure 2The regional risk index calculation module includes a high-risk node extraction submodule, a risk level screening submodule, and a node risk index generation submodule, among which:

[0082] High-risk node extraction submodule: Based on the meteorological correlation weighted graph, a graph convolutional network is used to extract nodes with obvious meteorological parameter fluctuations. By analyzing the meteorological parameter variation range of each node, nodes with large meteorological parameter deviation values ​​are screened, the threshold range of parameter deviation is confirmed, and the data characteristics of high-deviation nodes are identified to form a high-risk node deviation set;

[0083] Risk level screening submodule: Based on the high-risk node deviation set, the meteorological parameter deviation values ​​of the nodes are compared layer by layer, the deviation data of each node is hierarchically classified according to the risk level standard, the risk level of each node is divided and marked, and a risk level marking set is generated;

[0084] Node risk index generation submodule: Based on the risk level tag set and combined with the weight parameters of the meteorological correlation weighted graph, the marked risk level is numerically calculated, the risk index value of each node is synthesized, the quantitative expression of the node risk is established, and the node risk level index graph is obtained;

[0085] High-risk node extraction submodule: Based on the meteorological correlation weighted graph, the graph convolutional network is used to convolve the meteorological parameters of each node, extract the nodes with obvious meteorological parameter fluctuations, and aggregate the meteorological parameters layer by layer. By setting the number of convolution layers to 3 and the learning rate to 0.01, the node feature matrix and the adjacency matrix are used as inputs, and the aggregated features of each node are extracted through recursive calculation. The network is trained using the mean square error as the loss function. During the training process, the parameter deviation threshold range is set to 0.1, and the nodes with meteorological parameter deviation values ​​greater than the set threshold are screened out. The high-deviation data features of the nodes are identified to form a high-risk node deviation set;

[0086] Risk level screening submodule: Based on the high-risk node deviation set, the meteorological parameter deviation values ​​of the nodes are compared layer by layer, and the node deviation data is hierarchically classified using a hierarchical clustering algorithm. First, the meteorological parameter deviation value vectors of each high-risk node are extracted, and the difference data between the nodes are aggregated and grouped according to the risk level standard. The three risk categories of low, medium and high are divided in turn, and labels are assigned to mark the risk level of each node. The hierarchical clustering results are converted into a list form to generate a risk level label set;

[0087] Node risk index generation submodule: Based on the risk level tag set and combined with the weight parameters of the meteorological correlation weighted graph, the marked risk level is numerically calculated, and the risk index value of each node is synthesized using a weighted accumulation algorithm. First, the risk level of each node is assigned 1, 2, and 3 according to the level intensity, and the risk value of the marked node is multiplied by the corresponding weight parameter, and weighted accumulation is performed layer by layer to obtain the final risk index value. The risk index table is generated by traversing all nodes, and a quantitative expression of node risk is established to obtain a node risk level index graph;

[0088] Among them, the regional risk index map includes the cumulative value of risk deviation, node risk level, and regional risk weight distribution.

[0089] Graph convolutional network, according to the formula:

[0090]

[0091] Where: H (l+1) is the feature representation of node i in the l+1 layer, σ is the activation function, is the feature representation of neighbor node j at layer l, W (l) is the weight matrix of the lth layer, α represents the weight coefficient related to the fluctuation deviation of meteorological parameters, ΔP i represents the meteorological parameter deviation value of node i, β represents the weight coefficient of the influence of neighboring nodes, C j represents the fluctuation characteristics of the meteorological parameters of neighbor node j, γ represents the weight coefficient of the complexity of the node meteorological environment, M i Represents the complexity factor of the meteorological environment where node i is located;

[0092] Execution process: For node i, select the feature representation of each neighbor node j from the neighbor node set N(i) And with the weight matrix W of the current layer (l) Multiply, realize linear transformation of features through matrix operation, obtain the basic features of the l+1th layer nodes, and then aggregate the feature representation of each neighbor node with the feature representation of node i, through Normalize the degree of neighbor node features to ensure that node degree differences do not affect the results, balance the influence of highly connected nodes and low-degree connected nodes in the features, and introduce the weather fluctuation deviation value ΔP i The product of coefficient α represents the weather fluctuation of node i and adjusts its influence weight in the feature. The specific value of α can be obtained by optimizing the training set, and then combined with the weather fluctuation feature C of the neighboring node j The product of the coefficient β quantifies the impact of the meteorological fluctuations of the neighboring nodes on the target node i, and the complexity factor M of the meteorological environment where the node i is located is iMultiplying it with the corresponding weight coefficient γ reflects the impact of environmental complexity on the identification of high-risk nodes. The higher the environmental complexity, the more significant the feature deviation of the node. Finally, all the above factors are superimposed, and the result range is limited by the activation function σ. The output is the feature representation H of node i in the l+1 layer. (l+1) , and obtain the high-risk node deviation set.

[0093] The threshold range of parameter deviation includes the upper limit of temperature deviation, the upper limit of humidity deviation and the lower limit of wind speed deviation, which are used to identify nodes with abnormal fluctuations in meteorological parameters. The risk level standards specifically include low-risk deviation range, medium-risk deviation range and high-risk deviation range, which are used to hierarchically classify the meteorological parameter deviation data of each node.

[0094] See also Figure 2 The meteorological risk assessment module includes a meteorological parameter calling submodule, a risk deviation judgment submodule, and a high-risk node marking submodule, among which:

[0095] Meteorological parameter calling submodule: Based on the node risk level index graph, the density clustering algorithm is used to extract the real-time meteorological parameter data of each node, and the current deviation value of each node is determined by comparing it with the benchmark range data, and the deviation value set of each node is calculated to form a node meteorological deviation value set;

[0096] Risk deviation judgment submodule: Based on the node meteorological deviation value set, compare the deviation value range of each node, calculate and determine the degree to which the deviation exceeds the benchmark, divide the deviation level and mark the risk level, and generate a deviation level judgment set;

[0097] High-risk node marking submodule: Based on the deviation level judgment set, it identifies the high-risk mark of each node, records the risk level and parameter exceeding level, classifies the node as high-risk, and forms a high-risk node level marking table containing risk marks and levels;

[0098] Meteorological parameter calling submodule: Based on the node risk level index graph, the density clustering algorithm is used to extract the real-time meteorological parameter data of each node, the density threshold is set to 5 and the minimum number of samples is set to 10, the node meteorological parameter data is used to construct the feature vector space, the similarity between nodes is calculated by cosine similarity, the meteorological parameter data set with higher density is screened out, the node parameters that meet the density standard are extracted, the pre-set meteorological benchmark range data table is called, the real-time meteorological parameters are compared with the benchmark data one by one, the real-time deviation value of each node is calculated, and the deviation value is recorded in the deviation set to form a node meteorological deviation value set;

[0099] Risk deviation judgment submodule: Based on the node meteorological deviation value set, the meteorological deviation value range of each node is compared, and the deviation data is graded using the threshold classification method. The deviation threshold standard is set to risk levels 1, 2, and 3. The degree of excess is determined by comparing the deviation value with the standard value. The risk level of each node is calculated and marked in turn, and a risk label is generated. The matrix is ​​recorded by corresponding the risk label to the meteorological deviation value to form a deviation level judgment set.

[0100] High-risk node marking submodule: Based on the deviation level judgment set, identify and mark the high-risk status of each node, use Boolean logic to determine whether the node is high-risk, assign high-risk marks to nodes with deviation levels of 3 and above, call the node marking table to compare and generate node data with parameters exceeding the level, traverse the nodes and classify the records according to their high-risk status, organize the risk levels and parameter exceeding levels of high-risk nodes into a list, and generate a high-risk node level marking table containing node risk level marks and risk marks;

[0101] Among them, the high-risk node level marking table includes risk node marking, risk level label, and parameter out-of-range category.

[0102] Density clustering algorithm, according to the formula:

[0103]

[0104] in: represents the adjusted deviation distance of node i, k represents the number of selected neighboring nodes, P i represents the real-time meteorological parameter value vector of node i, P j represents the meteorological parameter value vector of node j adjacent to node i, ∥P i -P j ∥ represents the Euclidean distance of the difference in meteorological parameters between nodes i and j, T i The current temperature value of node i, R i is the rainfall data of node i, V i is the wind speed value of node i, B i is the air pressure value of node i, w 1 、w 2 、w 3 、w 4 is the weight coefficient of each meteorological parameter;

[0105] Execution process: First, collect the real-time meteorological parameter data of each node to form a vector containing the meteorological parameters of each node, where the meteorological parameter value of node i is recorded as P i , the neighboring nodes of node i are denoted as P j, select k neighboring nodes closest to node i as the calculation basis, and quantify the basic deviation distance between nodes by calculating the difference in meteorological parameters between node i and each neighboring node. In order to enhance the adaptability to different meteorological conditions, the temperature value T of the node is introduced respectively. i , rainfall R i , wind speed V i and air pressure B i , apply different weights to adjust the impact in the deviation calculation, where the weight coefficient w 1 、w 2 、w 3 、w 4 The weights corresponding to the difference in meteorological parameters, temperature, rainfall, wind speed and air pressure are respectively determined by analyzing a large amount of historical meteorological data. First, the correlation analysis of the influence of different meteorological parameters is carried out, and the weights of each parameter are preliminarily estimated based on the correlation results. Then, the principal component analysis method is used for fitting. Finally, the actual data is verified and optimized to determine the final weight value. Finally, all the calculation results are summed up and the average of the total is taken to obtain the adjusted deviation distance of node i. Used to evaluate the deviation between real-time meteorological conditions and the benchmark range.

[0106] See also Figure 2 The dynamic model adjustment module includes a data association adjustment submodule, a node weight configuration submodule, and a risk model update submodule, wherein:

[0107] Data association adjustment submodule: Based on the high-risk node level marking table, call the latest observation data, extract the latest observation parameters of the marked nodes, calculate the deviation value between the latest data and the historical markings, confirm the changes within the deviation range by comparing them item by item, and adjust the association strength between the marked nodes and other nodes to generate an updated association strength set;

[0108] Node weight configuration submodule: Based on the updated association strength set, adjust the weight distribution between nodes, extract the latest association strength value of each node, analyze the distribution ratio of association strength in the network, and reconfigure the weight influence between nodes to ensure that the weight conforms to the latest data relationship, and obtain the node weight configuration table;

[0109] Risk model update submodule: Based on the node weight configuration table, integrate the latest node weights and associations, perform weighted calculations on node weights, integrate the latest weighted information between nodes into the overall network model, generate a weighted impact model between current nodes, and obtain a real-time risk weighted model;

[0110] Data association adjustment submodule: Based on the high-risk node level marking table, the latest observation data is called, and the latest observation parameters of the marked nodes are extracted through the dynamic time warping algorithm. The window width is set to 5, and the Euclidean distance is used as the distance metric. The latest observation data is aligned with the historical data item by item, and the deviation value between the two is calculated. The deviation threshold is set to 0.2. By comparison, it is determined whether the deviation of each node is within the threshold range. The change in the association strength between the marked node and the surrounding nodes is gradually analyzed. The association weights of the nodes whose changes exceed the threshold are reset to generate an updated association strength set.

[0111] Node weight configuration submodule: Based on the updated association strength set, the hierarchical analysis method is used to adjust the weight distribution between nodes, extract the latest association strength value of each node, divide the strength data into primary factors and secondary factors according to priority, and construct a pairwise comparison matrix. The weight of the primary factor is set to 0.6 and the weight of the secondary factor is set to 0.4. The relative weight of each node is calculated using the eigenvalue. The consistency test ensures that the weight distribution is reasonable, and the calculated weight value is filled into the node weight matrix to obtain the node weight configuration table;

[0112] Risk model update submodule: Based on the node weight configuration table, the latest node weights and associations are integrated, and the weighted accumulation algorithm is used to accumulate the weighted values ​​between nodes layer by layer. By merging the latest weighted data between nodes layer by layer, the influence value of the node weight in the overall network is calculated, and the node weight is added to the overall risk calculation according to a certain proportion. The updated weighted relationship between nodes is summarized into a complete network structure, and a real-time risk weighted model between the current nodes is generated;

[0113] Among them, the real-time updated risk weighting model includes the node weight adjustment value, the updated risk intensity, and the inter-node correction coefficient.

[0114] See also Figure 2 ,The credible warning screening module includes a confidence screening submodule, a node level sorting submodule, and a warning event recording submodule, among which:

[0115] Confidence screening submodule: Based on the real-time risk weighted model, the confidence screening threshold is set, the confidence values ​​of each node are compared, the nodes that meet the high confidence conditions are screened out, the high confidence nodes are marked, and a high confidence node set is generated;

[0116] Node ranking submodule: Based on the high-confidence node set, the nodes are ranked according to their risk levels, the risk level values ​​of the high-confidence nodes are extracted, the nodes are arranged in descending order, and a ranking list of the nodes is constructed to obtain a node ranking table;

[0117] Warning event recording submodule: Based on the node ranking table, the high-confidence nodes after ranking are recorded one by one, and the risk level and marking information of each node are sorted into warning event records to form a list of credible warning events including node risk levels;

[0118] Confidence screening submodule: Based on the real-time risk weighted model, the confidence screening threshold is set, and the Bayesian confidence interval algorithm is used to compare the confidence values ​​of each node. First, the confidence interval of the node data is extracted by setting the confidence threshold to 0.95, and the upper and lower limits of the confidence interval of each node are calculated. The log-likelihood estimation method is used to screen nodes with confidence higher than the set threshold, and the qualified nodes are marked as high-confidence nodes. The marked high-confidence nodes are generated and stored to obtain a high-confidence node set;

[0119] Node level sorting submodule: Based on the high-confidence node set, the nodes are sorted according to their risk levels. The risk level values ​​of high-confidence nodes are extracted and sorted using a quick sorting algorithm. The risk levels of high-confidence nodes are sorted in descending order. First, the hub point is set as the base node for sorting. The risk level value of each node is compared to partition. The nodes with higher levels are placed at the front of the sorting. The sorting results are stored in an ordered list, and a node level sorting list is constructed to obtain a node level sorting table.

[0120] Warning event recording submodule: Based on the node ranking table, record the sorted high-confidence nodes one by one, integrate the risk level and tag information of each node in the sorting order, use the structured data recording method to input the structured entries containing node ID, risk level and confidence label line by line, index and store the table data, form and export the credible warning event list;

[0121] Among them, the trusted warning event list includes the trusted warning node number, warning level identification, and confidence threshold classification.

[0122] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A traffic weather information system based on artificial intelligence, characterized in that: The system comprises: Node structure modeling module: Based on the key sections, intersections and monitoring points in the traffic area, by setting the spatial location and meteorological parameters of the nodes, the associated road types and meteorological characteristics of each node are classified and generated, and the structural model is formed by integrating the parameters of the adjacent relationship between nodes to establish the traffic node structure model; Meteorological correlation graph generation module: based on the traffic node structure model, extracts meteorological characteristics between nodes, sets the correlation strength between nodes, increases the influence of adjacent nodes layer by layer through weighted operation, and obtains a meteorological correlation weighted graph; Regional risk index calculation module: Based on the meteorological correlation weighted graph, a graph convolutional network is used to extract the meteorological parameter deviation values ​​of high-risk nodes, and risk level screening is established by calculating and comparing parameter deviations layer by layer. The weighted parameters are called again to obtain the risk level index graph of each node; Meteorological risk assessment module: Based on the regional risk index map, a density clustering algorithm is used to call node meteorological parameters, and the deviation value is determined by Gaussian distribution. At the same time, the risk mark and parameter exceeding level of high-risk nodes are determined by dynamic progressive comparison, and a high-risk node level mark table is generated; Dynamic model adjustment module: based on the high-risk node level marking table, calling the latest observation data, comparing and adjusting the correlation strength between the latest data and the marked node parameters, reconfiguring the weight influence between nodes, and generating a real-time updated risk weighted model; Credible warning screening module: Based on the real-time updated risk weighted model, high confidence nodes are marked according to the confidence threshold, and the screened node levels are recorded in sequence to form a credible warning event list.

2. The traffic weather information system based on artificial intelligence according to claim 1 is characterized in that: The traffic node structure model includes node location coordinates, associated road types, and meteorological characteristic parameter sets.

3. The traffic weather information system based on artificial intelligence according to claim 1 is characterized in that: The meteorological correlation weighted graph includes a wind speed weighted value, a temperature correlation value, and a humidity influencing factor.

4. The traffic weather information system based on artificial intelligence according to claim 1 is characterized in that: The regional risk index calculation module includes a high-risk node extraction submodule, a risk level screening submodule, and a node risk index generation submodule, wherein: High-risk node extraction submodule: Based on the meteorological correlation weighted graph, a graph convolutional network is used to extract nodes with obvious meteorological parameter fluctuations. By analyzing the meteorological parameter variation range of each node, nodes with large meteorological parameter deviation values ​​are screened, the threshold range of parameter deviation is confirmed, and the data characteristics of high-deviation nodes are identified to form a high-risk node deviation set; Risk level screening submodule: based on the high-risk node deviation set, compare the meteorological parameter deviation values ​​of the nodes layer by layer, classify the deviation data of each node according to the risk level standard, divide and mark the risk level of each node, and generate a risk level marking set; Node risk index generation submodule: Based on the risk level tag set and combined with the weight parameters of the meteorological association weighted graph, the marked risk level is numerically calculated, the risk index value of each node is synthesized, the quantitative expression of the node risk is established, and the node risk level index graph is obtained; Among them, the regional risk index map includes the cumulative value of risk deviation, node risk level, and regional risk weight distribution.

5. The traffic weather information system based on artificial intelligence according to claim 1 is characterized in that: The graph convolutional network, according to the formula: Where: H (l+1) is the feature representation of node i in the l+1 layer, σ is the activation function, is the feature representation of neighbor node j at layer l, W (l) is the weight matrix of the lth layer, α represents the weight coefficient related to the fluctuation deviation of meteorological parameters, ΔP i represents the meteorological parameter deviation value of node i, β represents the weight coefficient of the influence of neighboring nodes, C j represents the fluctuation characteristics of the meteorological parameters of neighbor node j, γ represents the weight coefficient of the complexity of the node meteorological environment, M i Represents the complexity factor of the meteorological environment where node i is located.

6. The traffic weather information system based on artificial intelligence according to claim 4 is characterized in that: The threshold range of the parameter deviation includes the upper limit of temperature deviation, the upper limit of humidity deviation and the lower limit of wind speed deviation, which are used to identify nodes with abnormal fluctuations in meteorological parameters. The risk level standard specifically includes a low-risk deviation range, a medium-risk deviation range and a high-risk deviation range, which are used to hierarchically classify the meteorological parameter deviation data of each node.

7. The traffic weather information system based on artificial intelligence according to claim 1 is characterized in that: The meteorological risk assessment module includes a meteorological parameter calling submodule, a risk deviation judgment submodule, and a high-risk node marking submodule, wherein: Meteorological parameter calling submodule: Based on the node risk level index graph, a density clustering algorithm is used to extract the real-time meteorological parameter data of each node, and the current deviation value of each node is determined by comparing it with the benchmark range data, and the deviation value set of each node is calculated to form a node meteorological deviation value set; Risk deviation judgment submodule: based on the node meteorological deviation value set, compare the deviation value range of each node, calculate and determine the degree of deviation exceeding the benchmark, divide the deviation level and mark the risk level, and generate a deviation level judgment set; High-risk node marking submodule: Based on the deviation level determination set, identify the high-risk mark of each node, record the risk level and parameter exceeding level, classify the node as high risk, and form a high-risk node level marking table including risk marks and levels; The high-risk node level marking table includes risk node marking, risk level label, and parameter out-of-range category.

8. The traffic weather information system based on artificial intelligence according to claim 1 is characterized in that: The density clustering algorithm is based on the formula: in: represents the adjusted deviation distance of node i, k represents the number of selected neighboring nodes, P i represents the real-time meteorological parameter value vector of node i, P j represents the meteorological parameter value vector of node j adjacent to node i, ∥P i -P j ∥ represents the Euclidean distance of the difference in meteorological parameters between nodes i and j, T i The current temperature value of node i, R i is the rainfall data of node i, V i is the wind speed value of node i, B i is the air pressure value of node i, w1, w2, w3, w4 are the weight coefficients of each meteorological parameter.

9. The traffic weather information system based on artificial intelligence according to claim 1, characterized in that: The real-time updated risk weighted model includes a node weight adjustment value, an updated risk intensity, and an inter-node correction coefficient.

10. The traffic weather information system based on artificial intelligence according to claim 1, characterized in that: The credible warning event list includes credible warning node numbers, warning level identifiers, and confidence threshold classifications.

Images