Intelligent traffic information acquisition and early warning system

Abstract

The invention relates to the technical field of intelligent traffic, in particular to an intelligent traffic information acquisition and early warning system, which comprises a traffic data acquisition module, a feature extraction and fusion module, a health index calculation module and a traffic fault diagnosis and early warning module, synchronizing the data by using a space-time alignment algorithm; the feature extraction and fusion module extracts traffic safety information features and splices the traffic safety information features to obtain a comprehensive feature vector; the health index calculation module constructs a Hybrid-PINN model and introduces a traffic flow dynamics equation to calculate a traffic safety health index; and the traffic fault diagnosis and early warning module identifies vehicle and road faults through the ST-GCN network to obtain fault types, and performs graded early warning according to the fault types and the traffic safety and health indexes. Therefore, the problems of single monitoring dimension, lack of physical constraints, insufficient spatial-temporal feature utilization in fault diagnosis and the like in the prior art are solved.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, and particularly relates to an intelligent transportation information collection and early warning system. Background Art

[0002] With the acceleration of the urbanization process and the continuous growth of the motor vehicle ownership, it is crucial to ensure the safety of traffic operation. The traditional traffic safety monitoring system mainly relies on manual inspection and single sensor to collect data, and there are problems such as limited monitoring range, poor real-time performance, and lagged fault diagnosis.

[0003] In the prior art, although some systems can collect traffic safety data, there are obvious deficiencies in data processing and analysis. On the one hand, the existing systems usually process road monitoring data and vehicle operation data independently, lacking an effective cross-system data fusion mechanism; on the other hand, the traditional CNN / LSTM networks ignore the traffic flow dynamics law, are difficult to model the spatial topology relationship of the system, make insufficient use of spatio-temporal features, and are difficult to comprehensively capture the complex features of the traffic system. In addition, the existing fault early warning mechanism is mostly based on fixed threshold judgment, lacking an in-depth understanding of the dynamic characteristics and physical laws of the traffic system, prone to false alarms or missed alarms, and difficult to meet the intelligent and precise traffic safety operation and maintenance requirements. Summary of the Invention

[0004] This application provides an intelligent transportation information collection and early warning system to solve the problems in the prior art such as single monitoring dimension, lack of physical constraints, and insufficient utilization of spatio-temporal features in fault diagnosis.

[0005] An embodiment of the first aspect of the present application provides an intelligent transportation information collection and early warning system, including: a traffic data collection module, a feature extraction and fusion module, a health index calculation module, and a traffic fault diagnosis and early warning module. Among them, the traffic data collection module is used to collect traffic safety information and synchronize traffic safety data using a spatio-temporal alignment algorithm. Among them, the traffic safety information includes road health information, vehicle operation information, and environmental information; the feature extraction and fusion module extracts road health information features, vehicle operation information features, and environmental information features respectively, and splices the road information features, vehicle operation information features, and environmental information features to obtain a comprehensive feature vector. Among them, the road information features include road surface health, traffic flow, and lane occupancy rate. The vehicle operation features include vehicle speed, acceleration, sudden braking / lane-changing behavior, and GPS trajectory. The environmental information features include weather and visibility; the health index calculation module calculates the traffic safety health index by constructing a Hybrid-PINN model and introducing a traffic flow dynamics equation. If the traffic safety health index is lower than the target threshold, a fault diagnosis is performed; the traffic fault diagnosis and early warning module identifies vehicle-road faults by learning the correlation features of road states and vehicle operation states through an ST-GCN network, obtains the fault type, and performs hierarchical early warning according to the fault type and the traffic safety health index.

[0006] In a preferred embodiment, the traffic data collection module is used to collect traffic safety information, and the specific steps are as follows: A1. Traffic information collection: Deploy distributed fiber optic sensors every 300m along the road, set a unique sensor identifier, set the fixed sampling frequency of the sensors, and obtain the road stress distribution obtained by the sensors ; Use a high-definition camera in cooperation with a laser light source to obtain road surface images, and obtain road crack information I(t) and pothole information L(t) through image processing; obtain traffic flow Q and lane occupancy rate O through geomagnetic sensors; obtain vehicle real-time speed, acceleration, braking state, and GPS position through in-vehicle terminal data; obtain data on collection temperature, visibility, wind speed, and rainfall through the nearest weather station; A2. Data spatio-temporal synchronization: Stamp all data with a unified time stamp, encapsulate the data obtained by high-frequency sensors into a sliding window, generate synchronized data points for the data obtained by low-frequency sensors through linear interpolation to achieve time alignment, map the vehicle position P(t) to a unified road network coordinate system, and determine the spatial distance d(t) between the vehicle and the nearest road monitoring point. When d(t) is less than the target threshold, establish data association; align the road data O and vehicle data T through a spatio-temporal alignment function, and the alignment function is where, is the spatial association threshold, is the time - related threshold, is the result of mapping the vehicle position to the road network coordinate system, represents the position of the K - th road sensor; calculate the correlation coefficient R of the aligned data to verify the alignment effect, requiring R≥0.9, where, , where Cov is the covariance, is the standard deviation of the road data, is the standard deviation of the vehicle data.

[0007] In a preferred embodiment, the feature extraction and fusion module obtains a comprehensive feature vector by separately extracting the road health information features, vehicle operation information features, and environmental information features, and splicing the road information features, vehicle operation information features, and environmental information features. The specific steps are as follows: B1. Road information feature extraction: For the stress extract the optical fiber stress features, including time - domain features and frequency - domain features. The time - domain features include the mean value, standard deviation, and peak factor CF, and the frequency - domain features include the spectrum S(f), extract the main frequency , and the frequency - band energy ratio ; for the road crack information I(t), extract the crack features, and the crack features include the crack length , width and depth ; for the pothole information L(t), extract the features, and the pothole features include the pothole length , width and depth ; splice the extracted optical fiber stress features, crack features, and pothole features to obtain the road information feature ; B2. Vehicle operation information feature extraction: Extract the vehicle operation features within 15 seconds of the vehicle through in - vehicle terminal data, including the average speed, number of hard brakes, and lane - changing frequency. Calculate the traffic flow features through geomagnetic data, and the traffic flow features include traffic flow Q, traffic density M, and lane occupancy rate O; splice the extracted vehicle operation features and traffic flow features to obtain the vehicle operation feature ; B3. Environmental information feature extraction: Extract the environmental information features according to the data obtained from the weather station to obtain the environmental information feature , and the environmental information features are temperature, visibility, wind speed, and rainfall data; B4. Feature fusion: Standardize the road information feature and the vehicle operation feature and the environmental information feature , and directly splice the standardized feature vectors according to the dimensions to construct a comprehensive feature vector , where .

[0008] In a preferred embodiment, the health index calculation module calculates the traffic safety health index by constructing a Hybrid-PINN model and introducing a traffic flow dynamics equation. The specific steps are as follows: C1. Feature learning: Input the comprehensive feature vector into the model, perform non-linear transformation through a multi-layer LeakyReLU activation function, mine the relationships between features, and enhance the feature propagation ability through residual connections. Among them, for the l-th layer, the calculation formula is: , , where L is the number of network layers, is the output feature vector of the l-th layer, is the weight matrix of the l-th layer, is the bias vector of the l-th layer, is the output of the previous layer; C2. Physical constraint embedding: Introduce a traffic flow dynamics equation to describe the movement law of vehicles in traffic flow and their interaction relationship with the road environment. Among them, the traffic flow dynamics equation is the LWR equation, and the formula is , where is the traffic density, which refers to the number of vehicles per unit length of the road at time t and position x, is the traffic flow, which refers to the number of vehicles passing through a certain section per unit time at time t and position x; Take the traffic flow dynamics equation as a constraint condition and integrate it into the model by minimizing the residual term; C3. Health index calculation: Map the health index in the health index mapping network, and generate a standardized health index through a Sigmoid activation function in the output layer network. Among them, the health index mapping formula is , where and are the weight matrix and bias vector of the health index mapping network, is the output of the previous layer network; The formula for generating the standardized health index is , where is the Sigmoid activation function, which maps the output value to the range of [0, 100].

[0009] In a preferred embodiment, if the traffic safety and health index is lower than the target threshold, fault diagnosis is performed, including: grading the health status according to the traffic safety and health index. Specifically, if the traffic safety and health index is greater than or equal to 90, the health status is marked as Excellent; otherwise, if the traffic safety and health index is greater than or equal to 80, the health status is marked as Good; otherwise, if the traffic safety and health index is greater than or equal to 60, the health status is marked as Attention; otherwise, if the traffic safety and health index is greater than or equal to 40, the health status is marked as Warning; otherwise, the health status is marked as Danger. When the traffic safety and health index is less than 60, fault diagnosis is performed.

[0010] In a preferred embodiment, the Hybrid-PINN model consists of an input layer, a hidden layer, and an output layer. Among them, the input layer receives the comprehensive feature vector output by the feature extraction and fusion module , which includes road information features, vehicle operation features, and environmental information features; the hidden layer is composed of three types of hybrid neural networks, and the three types of hybrid neural networks include a feature extraction network, a physical constraint network, and a health index mapping network; the output layer outputs the traffic safety and health index, which reflects the overall state of road and vehicle operation, and the value range is [0, 100]. The higher the value, the better the system state.

[0011] In a preferred embodiment, the ST-GCN network is used to learn the correlation features between the road state and the vehicle operation state to identify vehicle-road faults and obtain the fault types. The specific steps are as follows: D1. Construct a spatio-temporal graph G=(V, ), where V is the spatio-temporal graph node, which are respectively road nodes and vehicle nodes. The road nodes include data such as position, stress, and crack, and the vehicle nodes include data such as speed, acceleration, and lane change frequency; is the spatial edge adjacency matrix, including road connection edges and lane interaction edges. Among them, the road adjacent edges connect adjacent road nodes, and the weights are calculated according to the spatial distance. The lane interaction edges connect vehicle nodes and road nodes at corresponding positions, and the weights are calculated based on the influence degree; is the temporal edge adjacency matrix, and the temporal edges are used to connect the states of the same node at adjacent times to form a time series; is the Kronecker product, which combines the spatial and temporal dimensions; D2. By performing graph convolution operations on the spatio-temporal graph to extract the correlation features between nodes, the graph convolution includes a spatial graph convolution layer and a temporal graph convolution layer, and residual connections are performed through spatio-temporal residual blocks to enhance feature propagation. Among them, the spatial graph convolution layer is used to capture the spatial dependence relationships between nodes at different positions in the vehicle-road system and model the structural correlation. Its formula is , where is the node feature matrix at time t, N is the number of nodes, C is the feature dimension, is the k-th order polynomial expansion of the spatial adjacency matrix, is 's degree matrix, is the learnable weight matrix, is the activation function, K is the order of the polynomial, which controls the propagation range of spatial dependence; the temporal graph convolution layer is used to capture the evolution law of the vehicle-road system state over time and model the dynamic characteristics. Its calculation formula is , where is the node feature matrix at time t + m, is the learnable weight matrix in the time dimension, M is the time window radius, which controls the utilization range of historical and future information, is the activation function; the spatio-temporal feature H is obtained by fusing the features of the spatial graph convolution layer and the temporal graph convolution layer, where ; D3. Fault type identification: Simultaneously use the global average pooling method and the global maximum pooling method to aggregate the fault features, retain the statistical characteristics and extreme value information of the features, and obtain the fault aggregation feature where , GAP() is the global average pooling method, and GMP() is the global maximum pooling method; map the fault features to obtain the fault feature vector , where , is the weight matrix, , is the bias term; classify the fault feature vector z through a fully connected layer and the Softmax function, and output the probabilities of each fault type , where K is the number of fault types, and P(y = k|z) is the probability that the sample belongs to the k-th type of fault; determine the fault type according to the fault diagnosis rule, where the fault diagnosis rule is , where is the determined fault type.

[0012] In a preferred embodiment, a graded warning is performed according to the fault type and the traffic safety health index, including: setting a different alarm threshold for each fault type, and dividing the three-color warning into yellow warning, orange warning, and red warning in combination with the traffic safety health index, wherein the yellow warning means that the fault reaches the set alarm threshold, and the traffic safety health index is between 50 and 60; the orange warning means that the fault reaches the set alarm threshold, and the traffic safety health index is between 30 and 50; the red warning means that the fault reaches the set alarm threshold, and the traffic safety health index is less than 30; Therefore, this application includes the following beneficial effects: The embodiment of the present application uses a spatiotemporal alignment algorithm to unify the timestamps of road information, vehicle information and environment, time align the data obtained by high- and low-frequency sensors, and map the vehicle position to a unified road network coordinate system, thereby achieving accurate spatiotemporal synchronization of multi-source data, avoiding analysis errors caused by data asynchrony, and providing a reliable data basis for fault diagnosis; constructing a Hybrid-PINN model, introducing traffic flow dynamics equations, combining data drive with physical laws, using deep learning to mine data features, and using physical equations to ensure that the results conform to the actual physical mechanism, making the calculation of the traffic safety and health index more scientific and explainable; using the ST-GCN network to learn the spatiotemporal correlation characteristics of road status and vehicle operation status, capturing the spatial dependency of nodes at different positions through the spatial graph convolution layer, and capturing the temporal evolution of the system state through the temporal convolution layer, which can accurately identify the vehicle-road fault type; the system divides the three-color warning according to the identified fault type and the traffic safety and health index, can quickly locate the problem and take targeted measures, improve operation and maintenance efficiency, reduce the probability of safety accidents, and ensure the safety and smoothness of traffic. As a result, technical problems in the prior art such as single monitoring dimension, lack of physical constraints, and insufficient utilization of spatiotemporal features in fault diagnosis are solved.

[0013] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 A schematic diagram of the composition of a smart traffic information collection and warning system provided according to an embodiment of the present application; Figure 2 A schematic diagram of a traffic data collection module provided according to an embodiment of the present application; Figure 3Schematic diagram of a feature extraction and fusion module provided according to an embodiment of the present application; Figure 4 Schematic diagram of a health index calculation module provided according to an embodiment of the present application; Figure 5 Flowchart for generating a traffic safety health index provided according to an embodiment of the present application; Figure 6 Schematic diagram of a traffic fault diagnosis and warning module provided according to an embodiment of the present application; Figure 7 Flowchart of an intelligent transportation information collection and warning system provided according to an embodiment of the present application. Detailed implementation manner

[0015] The embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present application, but should not be construed as limiting the present application.

[0016] An intelligent transportation information collection and warning system according to an embodiment of the present application will be described below with reference to the accompanying drawings. In view of the problems mentioned in the above background art, such as the lack of an effective cross-system data fusion mechanism and the neglect of dynamic laws, the present application provides an intelligent transportation information collection and warning system. In this system, through a spatio-temporal alignment algorithm, the timestamps of road information, vehicle information, and the environment are unified, the data obtained by high- and low-frequency sensors are time-aligned, and the vehicle positions are mapped to a unified road network coordinate system, achieving accurate spatio-temporal synchronization of multi-source data, avoiding analysis errors caused by data asynchronization, and providing a reliable data basis for fault diagnosis; a Hybrid-PINN model is constructed, introducing traffic flow dynamic equations, combining data-driven and physical laws, using deep learning to mine data features, and ensuring that the results conform to the actual physical mechanism through physical equations, making the calculation of the traffic safety health index more scientific and interpretable; using the ST-GCN network to learn the spatio-temporal correlation features of road states and vehicle operating states, capturing the spatial dependence relationships of nodes at different positions through the spatial graph convolutional layer, and capturing the time evolution laws of system states through the time convolutional layer, capable of accurately identifying vehicle-road fault types; the system divides three-color warnings according to the identified fault types, combines with the traffic safety health index, can quickly locate problems and take targeted measures, improve the operation and maintenance efficiency, reduce the probability of safety accidents, and ensure the safety and smoothness of traffic. Thus, the technical problems in the prior art, such as single monitoring dimension, lack of physical constraints, and insufficient utilization of spatio-temporal features in fault diagnosis, are solved.

[0017] Specifically, Figure 1A schematic diagram of the composition of a smart traffic information collection and warning system provided in an embodiment of the present application.

[0018] like Figure 1 As shown, the intelligent traffic information collection and warning system 10 includes: Traffic data collection module 100, feature extraction and fusion module 200, health index calculation module 300, traffic fault diagnosis and early warning module 400.

[0019] Among them, the traffic data collection module 100 is used to collect traffic safety information and synchronize the traffic safety data using a spatiotemporal alignment algorithm, wherein the traffic safety information includes road health information, vehicle operation information and environmental information; the feature extraction and fusion module 200 extracts the road health information features, vehicle operation information features and environmental information features respectively, and splices the road information features, vehicle operation information features and environmental information features to obtain a comprehensive feature vector; the health index calculation module 300 calculates the traffic safety health index by constructing a Hybrid-PINN model and introducing traffic flow dynamics equations. If the traffic safety health index is lower than the target threshold, fault diagnosis is performed; the traffic fault diagnosis and warning module 400 identifies vehicle-road faults by learning the correlation features between road status and vehicle operation status through the ST-GCN network, obtains the fault type, and warns the corresponding operation and maintenance personnel according to the fault type.

[0020] It can be understood that the embodiment of the present application unifies the timestamps of road information, vehicle information and environment through the spatiotemporal alignment algorithm, realizes time alignment of data obtained by high and low frequency sensors, and maps the vehicle position to a unified road network coordinate system, thereby realizing accurate spatiotemporal synchronization of multi-source data, avoiding analysis errors caused by data asynchrony, and providing a reliable data basis for fault diagnosis; constructing a Hybrid-PINN model, introducing traffic flow dynamics equations, combining data drive with physical laws, using deep learning to mine data features, and using physical equations to ensure that the results conform to the actual physical mechanism, making the calculation of the traffic safety and health index more scientific and explainable; using the ST-GCN network to learn the spatiotemporal correlation characteristics of road status and vehicle operation status, capturing the spatial dependency of nodes at different positions through the spatial graph convolution layer, and capturing the time evolution law of the system state through the time convolution layer, which can accurately identify the vehicle-road fault type; the system divides the three-color warning according to the identified fault type and the traffic safety and health index, and can quickly locate the problem and take targeted measures to improve operation and maintenance efficiency, reduce the probability of safety accidents, and ensure the safety and smoothness of traffic. As a result, technical problems in the prior art such as single monitoring dimension, lack of physical constraints, and insufficient utilization of spatiotemporal features in fault diagnosis are solved.

[0021] In the embodiment of the present application, the traffic data collection module 100 includes:Figure 2 As shown, traffic information is collected and data is synchronized in time and space.

[0022] Among them, traffic information collection is used to obtain road, vehicle, and environmental information. Among them, the health information of the road, including road stress, cracks, etc., is obtained through fiber optic sensors and high-definition cameras. The real-time position, speed, etc. of the vehicle are obtained through in-vehicle terminal data. Traffic flow, lane occupancy, etc. are obtained through geomagnetic sensors. The environmental information nearby is obtained through weather stations; data spatio-temporal synchronization is used to correlate and align road information, vehicle operation information, and environmental information through spatio-temporal alignment methods to obtain synchronized road, vehicle, and environmental information.

[0023] It can be understood that the embodiments of the present application utilize fiber optic sensors and high-definition cameras to obtain road health information, which has the advantages of high sensitivity, real-time performance, and non-contact detection. Fiber optic sensors can accurately capture changes in road stress, while high-definition cameras can clearly capture images of the road surface and accurately detect surface defects such as cracks and potholes through image recognition technology. Through in-vehicle terminals and geomagnetic sensors, real-time and dynamic tracking of vehicle operating states can be achieved. The spatio-temporal alignment method is used to correlate and align the collected data, effectively solving the problem of asynchronous multi-source data and greatly improving the correlation and value between data.

[0024] In the embodiments of the present application, the traffic data collection module is used to collect traffic safety information, and the specific steps are as follows: A1. Traffic information collection: Distributed fiber optic sensors are deployed every 300m along the road, a unique sensor identifier is set, and the fixed sampling frequency of the sensors is set to obtain the road stress distribution obtained by the sensors ; A high-definition camera is used in cooperation with a laser light source to obtain road surface images, and road crack information I(t) and pothole information L(t) are obtained through image processing; traffic flow Q and lane occupancy O are obtained through geomagnetic sensors; the real-time speed, acceleration, braking state, and GPS position of the vehicle are obtained through in-vehicle terminal data; the collected temperature, visibility, wind speed, and rainfall data are obtained through the nearest weather station.

[0025] It can be understood that in the embodiments of the present application, distributed fiber optic sensors are deployed every 300m along the road, which can achieve long-distance and continuous monitoring of the road. With the unique sensor identifier and fixed sampling frequency, the data source can be accurately located and stable and continuous monitoring sequences can be obtained. Its high-sensitivity detection of road stress can timely detect abnormalities caused by vehicle loads. Using a high-definition camera in cooperation with a laser light source to collect road surface images can obtain road surface details at extremely high speed and accuracy, providing a high-definition data basis for road surface defect detection. Through image processing technology, crack and pothole information can be effectively identified.

[0026] For example, in the scenario of high - temperature warning on an urban expressway, the distributed optical fiber sensor detected that the road surface stress at a certain section suddenly increased by 25% at 14:00. The images collected by the high - definition camera showed that there were reticular micro - cracks on the road surface in this area. The data from in - vehicle terminals indicated that the average vehicle speed decreased by 18% and the number of sudden braking events increased by 22% on this section of the road. The meteorological station simultaneously uploaded that the real - time temperature reached 38°C. Through the spatio - temporal alignment algorithm, the system accurately associated the thermal expansion stress concentration on the road surface, the expansion of micro - cracks, and the vehicle deceleration behavior in the high - temperature environment, determined that there was a risk of road diseases induced by high temperature on this section of the road, and immediately triggered an orange warning.

[0027] A2. Data spatio - temporal synchronization: Attach a unified timestamp to all data, encapsulate the data obtained by high - frequency sensors into a sliding window, generate synchronized data points for the data obtained by low - frequency sensors through linear interpolation to achieve time alignment, map the vehicle position P(t) to a unified road network coordinate system, determine the spatial distance d(t) between the vehicle and the nearest road monitoring point. When d(t) is less than the target threshold, establish data association; Align the road data O and vehicle data T through the spatio - temporal alignment function, and the alignment function is , where is the spatial association threshold, is the time association threshold, is the result of mapping the vehicle position to the road network coordinate system, represents the position of the K - th road sensor; Calculate the correlation coefficient R of the aligned data to verify the alignment effect, and require R≥0.9, where , where Cov is the covariance, is the standard deviation of the road data, is the standard deviation of the vehicle data.

[0028] It can be understood that in the embodiments of the present application, through technical means such as unified timestamps, coordinate systems, distance determination, and spatio - temporal alignment functions, the spatio - temporal asynchrony problem between road, vehicle, and environmental data is effectively solved, and accurate matching of multi - source heterogeneous data is achieved. This enables the correlation analysis of different types and different acquisition frequencies of data under the same spatio - temporal benchmark, avoids misjudgment and missed judgment caused by data misalignment, and significantly improves the utilization value of data and the reliability of analysis results. At the same time, the alignment effect is quantitatively verified through the correlation coefficient R to ensure the quality of data alignment, providing a solid data foundation for subsequent railway safety status assessment and fault diagnosis.

[0029] In the embodiments of the present application, the feature extraction and fusion module 200 includes: As Figure 3 shown, road information feature extraction, vehicle operation information feature extraction, environmental information feature extraction, and feature fusion.

[0030] Among them, road information feature extraction is used to extract the fiber optic stress feature, crack feature, and pothole feature of the road from road information data, and obtain road information features. ; Vehicle operation feature extraction is used to extract vehicle operation features from in-vehicle terminal data, calculate traffic flow features from geomagnetic data, and obtain vehicle operation features. ; Environmental information feature extraction is used to extract environmental information features from the data obtained from weather stations. ; Feature fusion is used to splice the road information features and vehicle operation features and environmental information features to obtain a comprehensive feature vector. .

[0031] It can be understood that in the embodiments of the present application, by separately extracting road information features, vehicle operation features, and environmental features and performing fusion, it is possible to comprehensively and multi-dimensionally mine traffic operation state information. Road information features such as fiber optic stress, cracks, and potholes can directly reflect the physical state of infrastructure; vehicle operation features and traffic flow features reveal the interaction between microscopic driving behaviors and macroscopic traffic laws; environmental features such as temperature and visibility quantify the impact of external factors on the vehicle-road system. Splicing the three into a comprehensive feature vector transforms the originally scattered and single data into an overall representation containing traffic system interaction information, providing a richer and more comprehensive data basis for subsequent health index calculation and fault diagnosis, and can effectively improve the ability to identify traffic safety hazards and analysis accuracy.

[0032] In the embodiments of the present application, road health information features, vehicle operation information features, and environmental information features are separately extracted, and the road information features, vehicle operation information features, and environmental information features are spliced to obtain a comprehensive feature vector. The specific steps are as follows: B1. Road information feature extraction: Extract fiber optic stress features for stress , including time domain features and frequency domain features. Among them, time domain features include mean, standard deviation, and peak factor CF, and frequency domain features include spectrum S(f). Extract the main frequency , frequency band energy ratio ; Extract crack features for road crack information I(t). The crack features include crack length , width and depth ; Extract features for pothole information L(t). The pothole features include pothole length , width and depth ; Splice the extracted fiber optic stress features, crack features, and pothole features to obtain road information features. ; It can be understood that in the embodiments of the present application, by extracting time-domain and frequency-domain features from stress data, the dynamic change law of road stress can be comprehensively understood. The time-domain features reflect the overall trend and fluctuation degree of stress changes, while the frequency-domain features reveal the frequency components of stress changes, which helps to discover potential periodic stress anomalies; by extracting geometric parameters from crack and pothole information, the development of cracks and potholes can be intuitively grasped, and their impact on vehicle running smoothness can be evaluated.

[0033] B2. Vehicle operation information feature extraction: Vehicle operation features within 15 seconds of the vehicle are extracted through in-vehicle terminal data, including average speed, number of emergency brakes, and lane-changing frequency. Traffic flow features are calculated through geomagnetic data, and the traffic flow features include traffic flow Q, traffic density M, and lane occupancy rate O; the extracted vehicle operation features and traffic flow features are spliced to obtain vehicle operation features ; It can be understood that in the embodiments of the present application, by extracting vehicle operation features and traffic flow features and splicing them, a multi-level feature system covering microscopic individual behaviors and macroscopic traffic situations can be constructed. The vehicle operation features formed after the two are spliced not only retain the dynamic details of individual vehicles but also incorporate the global trends of traffic flow, and can effectively capture the correlation between individuals and groups, providing a more comprehensive analysis dimension for traffic anomaly warning and driving risk assessment.

[0034] B3. Environmental information feature extraction: Environmental information features are extracted according to the data obtained from the weather station to obtain environmental information features , and the environmental information features are temperature, visibility, wind speed, and rainfall data; It can be understood that in the embodiments of the present application, extracting environmental information features can construct a dynamic environmental perception dimension of the traffic system. Data such as temperature, visibility, wind speed, and rainfall directly affect road performance and vehicle operation safety. Integrating these environmental features with road and vehicle data can quantitatively measure the impact degree of environmental factors on the traffic system in real time, providing a scientific basis for differential warning.

[0035] B4. Feature fusion: The road information features and vehicle operation features and environmental information features are standardized, and the standardized feature vectors are directly spliced according to the dimensions to construct a comprehensive feature vector , where .

[0036] It can be understood that in the embodiments of the present application, the road information features, vehicle operation features, and environmental information features are integrated into a comprehensive feature vector, breaking the information barriers between data, organically combining the originally independent three types of features, and forming an overall representation containing multi-faceted state information of the vehicle-road system. It provides more comprehensive and rich data input for subsequent health index calculation and fault diagnosis models, and enhances the model's ability to identify complex working conditions and potential faults.

[0037] In the embodiments of the present application, the health index calculation module 300 includes: as Figure 4 shown, feature learning, physical constraint embedding, and health index calculation.

[0038] Among them, feature learning is used to deeply mine the comprehensive feature vector and automatically extract the key features in the data through the feature extraction network of the Hybrid-PINN model; physical constraint embedding is used to introduce the traffic flow dynamics equation into the Hybrid-PINN model to describe the interaction relationship between vehicles and roads from the physical mechanism level; health index calculation is used to map the information processed by feature learning and physical constraint embedding into a traffic safety health index, and this index intuitively reflects the overall state of road and vehicle operation in the numerical interval of [0, 100].

[0039] In the embodiments of the present application, to construct a Hybrid-PINN model and introduce the traffic flow dynamics equation to calculate the traffic safety health index, the specific steps are as follows: C1. Feature learning: Input the comprehensive feature vector into the model, perform non-linear transformation through the multi-layer LeakyReLU activation function, mine the relationship between features, and enhance the feature propagation ability through residual connection. Among them, for the l-th layer, the calculation formula is: , , where L is the number of network layers, is the output feature vector of the l-th layer, is the weight matrix of the l-th layer, is the bias vector of the l-th layer, is the output of the previous layer.

[0040] It can be understood that in the embodiments of the present application, the multi-layer LeakyReLU activation function endows the model with powerful non-linear transformation ability. In the actual scenario, the feature relationship between the road and vehicle operation is often non-linear, and LeakyReLU can effectively capture this complex relationship, enabling the model to learn more accurate feature representations. The residual connection allows feature information to be directly propagated across layers, reducing the network training difficulty and helping to train deeper networks.

[0041] C2. Physical constraint embedding: Introduce the traffic flow dynamics equation to describe the movement law of vehicles in traffic flow and their interaction relationship with the road environment. Among them, the traffic flow dynamics equation is the LWR equation, and the formula is , where is the traffic density, which refers to the number of vehicles per unit length of the road at time t and position x, is the traffic flow, which refers to the number of vehicles passing through a certain section per unit time at time t and position x; Take the traffic flow dynamics equation as a constraint condition and integrate it into the model by minimizing the residual term; It can be understood that in the embodiments of the present application, integrating the traffic flow dynamics equation as a constraint condition into the model can ensure that the model output conforms to the real physical law and improve the accuracy of the model. In addition, the physical constraint is based on basic physical principles and is not limited to specific training data. When the model encounters a new scenario, due to following the physical law, it can better adapt to changes and enhance the generalization ability of the model.

[0042] C3. Health index calculation: Map the health index in the health index mapping network, and generate a standardized health index through the Sigmoid activation function in the output layer network. Among them, the health index mapping formula is , where and are the weight matrix and bias vector of the health index mapping network, is the output of the previous layer network; The formula for generating the standardized health index is , where is the Sigmoid activation function, which maps the output value to the range of [0, 100].

[0043] For example, as Figure 5 shows, in the daily monitoring of a main road in a certain city, the input layer of the Hybrid-PINN model receives a comprehensive feature vector, which includes features such as road stress (15 MPa), crack length (3 cm), average vehicle speed (50 km / h), traffic flow (1200 vehicles / h), air temperature (32 °C), etc. Through a 3-layer LeakyReLU network for feature learning, the non-linear correlation between road stress and vehicle deceleration in a high-temperature environment is mined, and then the LWR equation is embedded as a physical constraint to calculate the theoretical relationship between traffic density (25 vehicles / km) and flow. It is found that there is a residual between the measured flow (1200 vehicles / h) and the theoretical value (1350 vehicles / h), indicating that there may be a potential risk of local congestion. Optimize the model by minimizing the residual term, and finally output a health index of 68, which is the Attention level.

[0044] It can be understood that through the collaborative design of the health index mapping network and the output layer in the embodiments of the present application, the accurate quantification and intuitive presentation of the health status of the traffic system are achieved. Among them, the health index mapping network performs a non-linear transformation on the input features using the weight matrix and the bias vector, deeply mines the complex relationship between road and vehicle operation features, and converts the original data into potential representations related to the health status. The output layer maps the mapped features to the [0, 100] interval according to the Sigmoid activation function, giving the health index a standardized and intuitive expression.

[0045] In the embodiments of the present application, the Hybrid-PINN model consists of an input layer, a hidden layer, and an output layer. Among them, the input layer receives the comprehensive feature vector output by the feature extraction and fusion module , which includes road information features, vehicle operation features, and environmental information features; the hidden layer consists of three types of hybrid neural networks, and the three types of hybrid neural networks include a feature extraction network, a physical constraint network, and a health index mapping network; the output layer outputs the traffic safety health index, which reflects the overall state of road and vehicle operation, and the value range is [0, 100]. The higher the value, the better the system state.

[0046] It can be understood that in the embodiments of the present application, the data-driven and physical laws are deeply integrated through three types of hybrid neural networks. Among them, the input layer receives the comprehensive feature vector, providing a rich data basis for the model. In the hidden layer, the feature extraction network can automatically mine data features, the physical constraint network introduces traffic flow dynamics equations to ensure that the results conform to physical mechanisms, and the health index mapping network converts complex features into intuitive health indexes. The collaboration of the three avoids the limitations of relying solely on data or physical models, significantly improving the scientificity and reliability of traffic safety assessment. The traffic safety health index of the output layer reflects the overall state of the system in a quantitative form, facilitating the operation and maintenance personnel to quickly grasp the traffic safety situation.

[0047] In the embodiments of the present application, if the traffic safety health index is lower than the target threshold, fault diagnosis is performed, including: grading the health status according to the traffic safety health index. Among them, if the traffic safety health index is greater than or equal to 90, the health status is marked as Excellent; otherwise, if the traffic safety health index is greater than or equal to 80, the health status is marked as Good; otherwise, if the traffic safety health index is greater than or equal to 60, the health status is marked as Attention; otherwise, if the traffic safety health index is greater than or equal to 40, the health status is marked as Warning; otherwise, the health status is marked as Danger; when the traffic safety health index is less than 60, fault diagnosis is performed.

[0048] It can be understood that by grading the traffic safety and health index in the embodiments of the present application, the health assessment results can be transformed into intuitive and clear safety status identifiers, providing clear guidance for operation and maintenance decisions. When the health index is lower than 60 and enters the "Warning" and lower levels, the fault diagnosis process is started in a timely manner, which helps to discover potential hidden dangers in advance and achieve efficient management from preventive maintenance to precise emergency repair.

[0049] For example, in the daily monitoring of a certain section of road, the Hybrid-PINN model calculates that the traffic safety and health index of a certain monitoring area is 55. According to the grading rules, the health status of this area is marked as "Warning". The system immediately triggers the fault diagnosis process. By identifying the lane correlation features through the ST-GCN network, there is a risk of road surface structure damage due to high temperature and overloaded operation on this section of the road.

[0050] In the embodiments of the present application, the traffic fault diagnosis and early warning module 400 includes: as Figure 6 shown, spatio-temporal graph construction, spatio-temporal fault feature extraction, fault type identification, and traffic safety early warning.

[0051] Among them, spatio-temporal graph construction is used to structurally integrate the monitoring data of road nodes and vehicle nodes. By defining the spatial edge adjacency matrix and the temporal edge adjacency matrix, the road connection relationship, lane interaction influence, and state time series are quantified, and a network model that can comprehensively reflect the spatio-temporal correlation of the lane system is constructed; spatio-temporal fault features are used to extract through the collaborative operation of the graph convolutional layer, deeply mine the spatial dependence relationship between nodes and the dynamic evolution law of the system, and combine spatio-temporal residual blocks to enhance feature propagation to achieve precise capture of potential fault features; fault type identification is used to aggregate fault features, retain the feature statistical characteristics and extreme value information, classify through the fully connected layer and the Softmax function, output the probabilities of each fault type, and determine the specific fault type according to the diagnosis rules; traffic safety early warning sets different alarm thresholds according to the fault type, divides into three-color early warnings in combination with the traffic safety and health index, and notifies the operation and maintenance personnel to handle.

[0052] In the embodiments of the present application, the ST-GCN network is used to learn the correlation features of the road state and the vehicle operation state to identify vehicle-road faults and obtain the fault type. The specific steps are as follows: D1. Construct a spatio-temporal graph G=(V, ), where V is the spatio-temporal graph node, which are respectively the road node and the vehicle node. The road node includes data such as position, stress, and crack, and the vehicle node includes data such as speed, acceleration, and lane change frequency; is the spatial edge adjacency matrix, including the road connection edge and the lane interaction edge. Among them, the road adjacency edge connects adjacent road nodes, and the weight is calculated according to the spatial distance. The lane interaction edge connects the vehicle node and the road node at the corresponding position, and the weight is calculated based on the influence degree; is the time edge adjacency matrix. Time edges are used to connect the states of the same node at adjacent moments to form a time series; is the Kronecker product, which combines the spatial and temporal dimensions; It can be understood that by constructing a spatio-temporal graph in the embodiments of the present application, multi-source information of road states and vehicle operation states can be effectively integrated, the state characteristics of each node at different positions and moments can be captured, and the complex spatio-temporal relationships in the railway system can be comprehensively and intuitively described. By combining the spatial and temporal dimensions through the Kronecker product, the model can learn data features from both spatial and temporal dimensions simultaneously, more accurately identify lane failures, improve the accuracy and reliability of fault diagnosis, and provide strong support for traffic safety operation and maintenance.

[0053] D2. By performing graph convolution operations on the spatio-temporal graph to extract the correlation features between nodes, the graph convolution includes a spatial graph convolution layer and a temporal graph convolution layer, and residual connections are made through spatio-temporal residual blocks to enhance feature propagation. Among them, the spatial graph convolution layer is used to capture the spatial dependence relationships between nodes at different positions in the vehicle-road system, model the structural correlations, and its formula is , where is the node feature matrix at time t, N is the number of nodes, C is the feature dimension, is the k-th order polynomial expansion of the spatial adjacency matrix, is 's degree matrix, is the learnable weight matrix, is the activation function, K is the order of the polynomial, which controls the propagation range of spatial dependence; the temporal graph convolution layer is used to capture the evolution law of the vehicle-road system state over time, model the dynamic characteristics, and its calculation formula is , where is the node feature matrix at time t + m, is the learnable weight matrix in the temporal dimension, M is the radius of the time window, which controls the utilization range of historical and future information, is the activation function; the spatio-temporal features H are obtained by fusing the features of the spatial graph convolution layer and the temporal graph convolution layer, where ; It can be understood that through the synergistic effect of the spatial graph convolution layer and the temporal graph convolution layer of the ST-GCN network in the embodiments of the present application, the spatial structural correlations between nodes and the dynamic evolution law of the state over time in the vehicle-road system can be captured respectively, and the introduction of spatio-temporal residual blocks enhances the information propagation efficiency through cross-layer feature fusion, avoiding the problem of gradient disappearance in the training of deep neural networks.

[0054] D3. Fault type identification: The global average pooling method and the global max pooling method are simultaneously used to aggregate the fault features, retaining the statistical characteristics and extreme value information of the features to obtain the fault aggregation features Among them, , GAP() is the global average pooling method, and GMP() is the global max pooling method; the fault features are mapped to obtain the fault feature vector , where , is the weight matrix, , is the bias term; the fault feature vector z is classified through a fully connected layer and the Softmax function, and the probabilities of each fault type are output , where K is the number of fault types, and P(y = k|z) is the probability that the sample belongs to the k-th type of fault; the fault type is determined according to the fault diagnosis rule, where the fault diagnosis rule is , where is the determined fault type.

[0055] Specifically, the fault types include road damage, section congestion, and section accidents. The fault type with the highest probability is obtained from the probabilities of each fault output by the model as the determined fault type.

[0056] It can be understood that in the embodiment of the present application, by simultaneously using global average pooling (GAP) and global max pooling (GMP) to aggregate the fault features, the overall statistical characteristics (such as mean, trend) and local extreme value information (such as mutation points, abnormal peaks) of the features can be taken into account, avoiding the loss of feature information caused by a single pooling method. Specifically, GAP can capture the global distribution trend of the features and reflect the stable state or gradual change abnormality of the system; GMP focuses on the extreme values in the features and is sensitive to sudden faults or transient impacts. The combination of the two can form a complementary feature representation: both a benchmark description of the normal operating state and an accurate capture of abnormal transients, thereby improving the model's ability to identify complex faults. In addition, through the classification mapping of the fully connected layer and the Softmax function, the aggregated features can be transformed into a probability distribution, and combined with the fault diagnosis rule to achieve multi-dimensional evidence fusion, improving the reliability of classification decisions.

[0057] For example, in the real-time monitoring scenario of an urban expressway, spatio-temporal graph G is constructed with data of road nodes (such as stress value of 18 MPa and crack length of 4 cm at K5+200) and vehicle nodes (average speed of 60 km / h and lane-changing frequency of 3 times per minute within a certain period). In the spatial edge adjacency matrix, adjacent road nodes are assigned weights according to the actual distance (the closer the distance, the higher the weight). For the lane interaction edges between vehicle nodes and corresponding road nodes, weights are assigned according to the influence degree of vehicle driving on the road. The temporal edge adjacency matrix connects the states of the same node within 10 minutes. Through the spatial graph convolutional layer, the stress correlation between the road anomaly at K5+200 and the road section within 500 meters before and after is mined. The temporal graph convolutional layer captures the trend that the vehicle speed in this area has been continuously decreasing in the past 5 minutes. After fusion, spatio-temporal feature H is obtained. Global average pooling and global max pooling are performed on H to retain the feature mean and extreme value information. Through calculation by the fully connected layer and the Softmax function, the probability of "road damage" in this area is 85%, exceeding the set threshold. The system determines it as a pavement structure fault according to the fault diagnosis rules and immediately sends a warning to the maintenance department, prompting an emergency inspection and maintenance of this road section.

[0058] In the embodiment of the present application, warnings are issued according to the fault type and the traffic safety and health index, including: Different alarm thresholds are set for each fault type, and three-color warnings are divided in combination with the traffic safety and health index. The three-color warnings include yellow warning, orange warning, and red warning. Among them, the yellow warning means that the fault reaches the set alarm threshold and the traffic safety and health index is between 50 and 60; the orange warning means that the fault reaches the set alarm threshold and the traffic safety and health index is between 30 and 50; the red warning means that the fault reaches the set alarm threshold and the traffic safety and health index is less than 30. Specifically, for the road damage fault, if a yellow warning is triggered, notify the road maintenance personnel to conduct on-site investigation within 1 week, repair the damaged road section, and at the same time mark the road section as "slightly damaged pavement" in the navigation APP, and suggest that drivers slow down. If an orange warning is triggered, notify the road maintenance personnel to complete the repair within 48 hours, close the damaged lane, and turn on the arrow lights 1 km upstream to guide lane changes. If a red warning is triggered, contact the traffic police department, immediately block the two-way traffic of this road section, and start the emergency plan to divert the traffic flow to the alternate route.

[0059] For the road congestion fault, if a yellow warning is triggered, intelligently adjust the traffic light timing at intersections to give priority to diverting the traffic flow in the congested direction. If an orange warning is triggered, notify the traffic police to conduct on-site command and diversion, and set up a temporary traffic control point at the starting point of the congestion. If a red warning is triggered, guide the vehicles to leave the road section upstream of the congestion point and conduct area control of this road section.

[0060] For road section accidents and faults, if a yellow warning is triggered, the accident location and preliminary type will be pushed to the traffic police command center, the emergency lane will be opened for rescue vehicles to pass through, and at the same time, the section will be marked as "Accident Ahead" in the navigation APP, and drivers are advised to slow down. If an orange warning is triggered, the accident location and preliminary type will be pushed to the traffic police command center, the emergency lane will be opened for rescue vehicles to pass through, the accident lane will be closed, and the "zipper-style" alternating traffic will be enabled for the remaining lanes. If a red warning is triggered, the accident section will be immediately closed, the oncoming lane will be used for diversion, and 120 / 119 will be linked to provide the optimal rescue route.

[0061] It can be understood that by setting different alarm thresholds and combining the traffic safety and health index to divide the three-color warning in the embodiments of the present application, hierarchical response and precise control of traffic faults can be achieved. On the one hand, maintenance resources can be precisely allocated according to the severity of the fault and the overall health status of the system, avoiding over-response or insufficient response; on the other hand, the clear hierarchical response process provides clear guidance, reduces decision-making time, improves the timeliness and effectiveness of fault handling, and reduces the probability of traffic safety accidents.

[0062] A specific embodiment will be used below to elaborate on an intelligent transportation information collection and warning system, as Figure 7 shown, including: In a certain intersection section, traffic data is monitored and collected all day long. Among them, in terms of road health information collection, distributed fiber optic sensors capture road stress changes, high-definition cameras and lidar continuously scan the road surface to record crack and pothole characteristics. In terms of vehicle operation information collection, vehicle operation characteristics extracted from in-vehicle terminal data, including average speed, number of hard brakes, and lane change frequency, and traffic flow characteristics calculated from geomagnetic data. In terms of environmental information collection, environmental information near the meteorological station is collected. All data is aligned according to the spatio-temporal alignment algorithm to ensure that all traffic safety data is consistent in the time and space dimensions.

[0063] Feature extraction and fusion are performed on the collected data. Features such as road surface stress, cracks, and potholes are extracted from road information; for vehicle operation information, features such as average vehicle speed, hard brake frequency, number of lane changes, traffic flow, traffic density, and lane occupancy rate are calculated; temperature, visibility, rainfall, etc. are extracted from environmental information. The above features are spliced into a comprehensive feature vector to form a comprehensive feature vector containing multi-dimensional information of roads, vehicles, and the environment, intuitively presenting the operation status of the traffic system.

[0064] Using the Hybrid-PINN model, the comprehensive feature vector is deeply analyzed. The LeakyReLU network in the hidden layer mines the non-linear relationships between features, and the physical constraint embedding link introduces the LWR equation to calculate the theoretical relationship between traffic density and flow. Finally, the output layer of the model maps the processed information into the traffic safety and health index. The calculated traffic safety and health index for this section is 58, falling below the yellow warning threshold of 60, triggering the fault diagnosis process.

[0065] Learn and analyze the correlation features between the road state and vehicle operation state of this section. By constructing a spatial graph containing multiple surrounding intersections and road section nodes and combining time series data over a certain period (such as 30 minutes) in the past, the ST-GCN network identifies an abnormal traffic pattern in the current section. After calculation and judgment, the fault type is determined to be a road section accident. According to the pre-set corresponding rules between fault types and warning levels, since the health index is 58, in the range of 50 - 60, this fault is determined to be an orange warning. Immediately, the accident location and preliminary type are pushed to the traffic police command center, the emergency lane is opened for rescue vehicles to pass, and at the same time, "Accident Ahead" is marked in the navigation APP, advising drivers to slow down.

[0066] In summary, the embodiment of this application adopts the spatio-temporal alignment algorithm to ensure the accurate synchronization of road, vehicle, and environmental information, providing a reliable data basis for analysis. Feature fusion and Hybrid-PINN model calculation deeply combine physical laws with data features, facilitating the intuitive evaluation of the system state. Using the ST-GCN network, the fault type is accurately identified from the spatio-temporal dimension, and according to the three-color warning classification, timely processing is carried out for orange warnings. It realizes the full-process automation and intelligence from data collection, feature mining to fault diagnosis, not only improving the detection efficiency and accuracy of road section accidents, but also providing a scientific decision-making basis for traffic operation and maintenance, effectively reducing the probability and impact of traffic accidents.

[0067] An embodiment of the present application proposes an intelligent transportation information collection and early warning system. In this system, through a spatio-temporal alignment algorithm, the timestamps of road information, vehicle information, and the environment are unified, the data obtained by high- and low-frequency sensors are time-aligned, and the vehicle positions are mapped to a unified road network coordinate system, achieving precise spatio-temporal synchronization of multi-source data, avoiding analysis errors caused by data asynchronization, and providing a reliable data basis for fault diagnosis; a Hybrid-PINN model is constructed, introducing traffic flow dynamics equations, combining data-driven and physical laws, using deep learning to mine data features, and ensuring that the results conform to actual physical mechanisms through physical equations, making the calculation of the traffic safety and health index more scientific and interpretable; using an ST-GCN network to learn the spatio-temporal correlation features of road states and vehicle operating states, capturing the spatial dependence relationships of nodes at different positions through the spatial graph convolutional layer, and capturing the time evolution laws of system states through the temporal convolutional layer, capable of accurately identifying vehicle-road fault types; the system divides three-color early warnings according to the identified fault types, combines with the traffic safety and health index, can quickly locate problems and take targeted measures, improve operation and maintenance efficiency, reduce the probability of safety accidents, and ensure traffic safety and smoothness. Thus, the technical problems in the prior art such as single monitoring dimension, lack of physical constraints, and insufficient utilization of spatio-temporal features in fault diagnosis are solved.

[0068] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or N embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0069] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "N" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0070] Any process or method description, whether in a flowchart or otherwise described herein, can be understood to represent a module, segment, or portion of code that includes one or more executable instructions for implementing a customized logical function or process. The scope of the preferred embodiments of the present application includes additional implementations, where the functions may be executed in a substantially simultaneous manner or in a reverse order according to the functions involved, rather than in the order shown or discussed. This should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0071] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), and the like.

[0072] Those of ordinary skill in the art can understand that all or part of the steps carried out in the methods of the above embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium, and when executed, includes one or a combination of the steps of the method embodiments.

Claims

1. An intelligent transportation information collection and early warning system, characterized in that, Including: A traffic data collection module, a feature extraction and fusion module, a health index calculation module, and a traffic fault diagnosis and warning module. Among them, the traffic data collection module is used to collect traffic safety information and synchronize traffic safety data using a spatio-temporal alignment algorithm. Among them, the traffic safety information includes road health information, vehicle operation information, and environmental information; the feature extraction and fusion module extracts road health information features, vehicle operation information features, and environmental information features respectively, and splices the road information features, vehicle operation information features, and environmental information features to obtain a comprehensive feature vector. Among them, the road information features include road surface health, traffic flow, and lane occupancy rate. The vehicle operation features include vehicle speed, acceleration, sudden braking / lane-changing behavior, and GPS trajectory. The environmental information features include weather and visibility; the health index calculation module calculates the traffic safety health index by constructing a Hybrid-PINN model and introducing a traffic flow dynamics equation. If the traffic safety health index is lower than the target threshold, fault diagnosis is performed; the traffic fault diagnosis and warning module identifies vehicle-road faults by learning the correlation features of road conditions and vehicle operation states through an ST-GCN network, obtains the fault type, and performs hierarchical warning according to the fault type and the traffic safety health index.

2. The intelligent transportation information collection and early warning system according to claim 1, characterized in that The traffic data collection module is used to collect traffic safety information, and the specific steps are as follows: A1. Traffic information collection: Distributed fiber optic sensors are deployed every 300 m along the road, a unique sensor identifier is set, a fixed sampling frequency of the sensors is set, and the road stress distribution obtained by the sensors is acquired. ; A high-definition camera is used in conjunction with a laser light source to obtain road surface images, and road crack information I(t) and pothole information L(t) are obtained through image processing; traffic flow Q and lane occupancy rate O are obtained through geomagnetic sensors; real-time vehicle speed, acceleration, braking state, and GPS position are obtained through in-vehicle terminal data; temperature, visibility, wind speed, and rainfall data are acquired through the nearest weather station. A2. Data spatio-temporal synchronization: Assign a unified timestamp to all data, encapsulate the data obtained by high-frequency sensors into a sliding window, generate synchronized data points for the data obtained by low-frequency sensors through linear interpolation to achieve time alignment, map the vehicle position P(t) to a unified road network coordinate system, determine the spatial distance d(t) between the vehicle and the nearest road monitoring point, and establish data association when d(t) is less than the target threshold; Align the road data O and vehicle data T through a spatio-temporal alignment function, and the alignment function is , where is the spatial correlation threshold is the temporal correlation threshold is the result of mapping the vehicle position to the road network coordinate system represents the position of the K-th road sensor; calculate the correlation coefficient R of the aligned data to verify the alignment effect, and require R≥0.9, where , where Cov is the covariance is the standard deviation of the road data is the standard deviation of the vehicle data 3. A smart transportation information collection and warning system according to claim 1, characterized in that, The feature extraction and fusion module extracts road health information features, vehicle operation information features, and environmental information features respectively, and splices the road information features, vehicle operation information features, and environmental information features to obtain a comprehensive feature vector. The specific steps are as follows: B1. Road information feature extraction: for stress Extract the optical fiber stress features, including time-domain features and frequency-domain features, where the time-domain features include mean, standard deviation, and peak factor CF, and the frequency-domain features include spectrum S(f), and extract the main frequency , frequency band energy ratio ; Extract crack features from the road crack information I(t), where the crack features include crack length , width and depth ; Extract features from the pothole information L(t), where the pothole features include the pothole length , width and depth ; splice the extracted optical fiber stress features, crack features and pothole features to obtain road information features ; B2. Vehicle operation information feature extraction: Extract the vehicle operation features within 15 seconds from in-vehicle terminal data, including average speed, number of hard brakes, and lane-changing frequency. Calculate traffic flow features from geomagnetic data, where the traffic flow features include traffic flow Q, traffic density M, and lane occupancy rate O. Concatenate the extracted vehicle operation features and traffic flow features to obtain vehicle operation features ; B3. Environmental information feature extraction: Extract environmental information features based on the data obtained from the meteorological station to obtain environmental information features , where the environmental information features include temperature, visibility, wind speed, and rainfall data; B4. Feature Fusion: Standardize the road information features and vehicle operation features and environmental information features Perform standardization processing, directly splice the standardized feature vectors by dimension to construct a comprehensive feature vector , where .

4. A smart transportation information collection and warning system according to claim 1, characterized in that, The health index calculation module calculates the traffic safety health index by constructing a Hybrid-PINN model and introducing a traffic flow dynamics equation. The specific steps are as follows: C1. Feature learning: The comprehensive feature vector is input into the model, and undergoes non-linear transformation through the multi-layer LeakyReLU activation function to explore the relationships between features, and the feature propagation ability is enhanced through residual connections. For the l-th layer, the calculation formula is: , , where L is the number of network layers, is the output feature vector of the l-th layer, is the weight matrix of the l-th layer, is the bias vector of the l-th layer, is the output of the previous layer; C2. Physical constraint embedding: Introduce the traffic flow dynamics equation to describe the movement law of vehicles in the traffic flow and their interaction with the road environment. Among them, the traffic flow dynamics equation is the LWR equation, and the formula is , where is the traffic density, which refers to the number of vehicles on the road per unit length at time t and position x, is the traffic flow, which refers to the number of vehicles passing through a certain section per unit time at time t and position x; Take the traffic flow dynamics equation as a constraint condition and incorporate it into the model by minimizing the residual term; C3. Health Index Calculation: Map the health index in the health index mapping network, and generate a normalized health index through the Sigmoid activation function in the output layer network. The health index mapping formula is , where and are the weight matrix and bias vector of the health index mapping network, is the output of the previous layer network; the formula for generating the normalized health index is , where is the Sigmoid activation function, which maps the output value to the range of [0, 100].

5. A smart traffic information collection and warning system according to claim 1, characterized in that, If the traffic safety and health index is lower than the target threshold, fault diagnosis is performed, including: grading the health status according to the traffic safety and health index. Specifically, if the traffic safety and health index is greater than or equal to 90, the health status is marked as Excellent; otherwise, if the traffic safety and health index is greater than or equal to 80, the health status is marked as Good; otherwise, if the traffic safety and health index is greater than or equal to 60, the health status is marked as Attention; otherwise, if the traffic safety and health index is greater than or equal to 40, the health status is marked as Warning; otherwise, the health status is marked as Danger. When the traffic safety and health index is less than 60, fault diagnosis is performed.

6. A smart transportation information collection and warning system according to claim 1, characterized in that, The Hybrid-PINN model consists of an input layer, a hidden layer, and an output layer. Among them, the input layer receives the comprehensive feature vector output by the feature extraction and fusion module , which includes road information features, vehicle operation features, and environmental information features; the hidden layer is composed of three types of hybrid neural networks, and the three types of hybrid neural networks include a feature extraction network, a physical constraint network, and a health index mapping network; the output layer outputs a traffic safety health index, which reflects the overall state of road and vehicle operation, and the value range is [0, 100]. The higher the value, the better the system state.

7. A smart transportation information collection and warning system according to claim 1, characterized in that, The ST-GCN network is used to learn the correlation features between the road state and the vehicle operation state to identify vehicle-road faults and obtain the fault types. The specific steps are as follows: D1. Construct a spatio-temporal graph G=(V, ), where V is the spatio-temporal graph node, which are respectively the road node and the vehicle node. The road node includes data such as position, stress, crack, etc., and the vehicle node includes data such as speed, acceleration, lane-changing frequency, etc.; is the spatial edge adjacency matrix, which includes road connection edges and lane interaction edges. Among them, the road adjacent edges connect adjacent road nodes, and the weights are calculated according to the spatial distance. The lane interaction edges connect the vehicle nodes and the road nodes at the corresponding positions, and the weights are calculated based on the influence degree; is the temporal edge adjacency matrix. The temporal edges are used to connect the states of the same node at adjacent times to form a time series; is the Kronecker product, which combines the spatial and temporal dimensions; D2. By performing graph convolution operations on the spatio-temporal graph to extract the correlation features between nodes, the graph convolution includes a spatial graph convolution layer and a temporal graph convolution layer, and the residual connection is enhanced through a spatio-temporal residual block for feature propagation. Among them, the spatial graph convolution layer is used to capture the spatial dependence between nodes at different positions in the vehicle-road system and model the structural correlation. Its formula is , where is the node feature matrix at time t, N is the number of nodes, C is the feature dimension, is the k-order polynomial expansion of the spatial adjacency matrix, is 's degree matrix, is the learnable weight matrix, is the activation function, K is the polynomial order, which controls the propagation range of spatial dependence; the temporal graph convolution layer is used to capture the evolution law of the vehicle-road system state over time and model the dynamic characteristics. Its calculation formula is , where is the node feature matrix at time t + m, is the learnable weight matrix in the time dimension, M is the time window radius, which controls the utilization range of historical and future information, is the activation function; the spatio-temporal feature H is obtained by fusing the features of the spatial graph convolution layer and the temporal graph convolution layer. Among them, ; D3. Fault type identification: The global average pooling method and the global maximum pooling method are simultaneously used to aggregate fault features, retaining the statistical characteristics and extreme value information of the features to obtain fault aggregation features Among them, , GAP() is the global average pooling method, and GMP() is the global maximum pooling method; the fault features are mapped to obtain a fault feature vector , where , is the weight matrix, , is the bias term; the fault feature vector z is classified through a fully connected layer and a Softmax function to output the probabilities of each fault type , where K is the number of fault types, and P(y=k|z) is the probability that the sample belongs to the kth type of fault; the fault type is determined according to the fault diagnosis rule, where the fault diagnosis rule is , where is the determined fault type 8. A smart transportation information collection and warning system according to claim 1, characterized in that, Grade warnings are issued according to the fault types and the traffic safety and health index, including: Different alarm thresholds are set for each fault type, and three-color warnings are divided in combination with the traffic safety and health index. The three-color warnings include yellow warnings, orange warnings, and red warnings. Among them, the yellow warning means that the fault reaches the set alarm threshold and the traffic safety and health index is between 50 and 60; the orange warning means that the fault reaches the set alarm threshold and the traffic safety and health index is between 30 and 50; the red warning means that the fault reaches the set alarm threshold and the traffic safety and health index is less than 30.

Images