Traffic safety assessment method of axle coupling system based on physical data driving
By constructing a graph isomorphic network model, the shortcomings of traditional neural networks in the safety assessment of railway bridges under multiple scenarios are solved, achieving fast and accurate traffic safety assessment, applicable to different bridge structures and seismic conditions, and reducing computational costs.
Patent Information
- Application Number
- CN202510477054.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-16
- Publication Date
- 2025-08-01
AI Technical Summary
Traditional neural network models perform poorly in assessing the operational safety of railway bridges under various scenarios, especially in responding to emergencies where they cannot quickly and accurately assess traffic safety.
A graph isomorphic network model based on physical data is constructed. Through graph structuring and attention mechanisms, parameter changes in the vehicle-bridge coupling system are captured, and safety assessment is performed using velocity spectrum intensity and derailment coefficient.
It improves the accuracy and adaptability of driving safety assessments, reduces computing resource requirements, and can provide fast and reliable safety assessment results in complex environments.
Smart Images
Figure CN120409214A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of vehicle-bridge coupling systems, and specifically to a driving safety assessment method for a vehicle-bridge coupling system based on physical data-driven. Background Art
[0002] Currently, graph neural networks and graph representation methods are used to address the limitations in the running safety assessment of railway bridges in multiple scenarios. Traditional neural network models, such as long short-term memory networks, convolutional neural networks, etc., usually perform poorly on complex systems with dynamic topologies and material properties of engineering structures. This limitation makes traditional neural network models unable to quickly and accurately conduct driving safety assessments when dealing with emergencies such as earthquakes. Summary of the Invention
[0003] The technical problem to be solved by the present invention is to provide a driving safety assessment method for a vehicle-bridge coupling system based on physical data-driven, construct a graph isomorphism network model, effectively capture the impact of changes in parameters such as bridge structure form, train speed, and track irregularity on running safety, not only improve the accuracy of driving safety assessment, but also significantly save computing resources.
[0004] The technical solution of the present invention is as follows:
[0005] A driving safety assessment method for a vehicle-bridge coupling system based on physical data-driven specifically includes the following steps:
[0006] (1) Graph-structurize the vehicle-bridge coupling system to generate a graph structure G=(V, E), where V represents the nodes in the graph structure G, E represents the edges connecting the nodes in the graph structure G, and the nodes in the graph structure G include train nodes, bridge nodes, and pier nodes in the vehicle-bridge coupling system, and each node is assigned a feature vector;
[0007] (2) Construct a graph isomorphism network model and train it;
[0008] (3) Input the graph structure into the trained graph isomorphism network model for driving safety assessment, and output safety assessment indicators, where the safety assessment indicators include speed spectrum intensity VSI indicator and derailment coefficient.
[0009] The feature vector X of the train node v,train is the train speed, that is, X v,train =(v train ,…) T , v train represents the train speed at different times, and T represents the total sampling time; the feature vector X of the bridge node v,bridge is the track irregularity sequence, that is, X v,bridge =(r v , ra , r g , r c , …) T , r v represents the vertical profile irregularity data of the track at different times, r a represents the alignment irregularity data of the track at different times, r g represents the gauge irregularity data of the track at different times, r c represents the lateral irregularity data of the track at different times; the eigenvector X of the pier node v,pier is the seismic acceleration sequence, that is, x v,pier =(a x , a y , a z , …) T , a x represents the longitudinal seismic acceleration, a y represents the lateral seismic acceleration, a z represents the vertical seismic acceleration.
[0010] The described graph isomorphism network model includes an input layer and an attention module. After the input layer encodes the eigenvectors X v,train , X v,bridge , X v,pier of each node in the input graph structure G, the attention module first calculates the attention coefficient of each node, then aggregates the features of its neighbor nodes according to the attention coefficient of each node, and finally updates the node to obtain the updated eigenvector of the node.
[0011] The described attention module includes multi-head attention, point-wise feed-forward network, residual module, aggregation module, and update module. The multi-head attention is used to calculate the attention coefficient of each node, as shown in the following formulas (1) and (2). Then, after the point-wise feed-forward network performs a non-linear transformation, the residual module performs a residual connection operation. The aggregation module aggregates the features of its neighbor nodes according to the attention coefficient of each node, as shown in the following formula (3). Finally, the update module updates the node, as shown in the following formula (4):
[0012]
[0013] In formulas (1)-(4), e vu represents the edge feature between node v and its adjacent node u, e vi represents the edge feature between node v and its adjacent node i; N(v) represents the set of neighbor nodes of node v; W represents the weight matrix; linear represents the linear layer; LeakyReLu and σ are both activation functions used to introduce non-linear characteristics; x v represents the eigenvector of node v, x uThe eigenvector representing node u;
[0014] α vu represents the attention coefficient between node pair (v, u), that is, node u aggregates the eigenvector to node v according to the ratio of α vu ; represents the information aggregated from all neighbor nodes of node v in the l-th layer of the graph isomorphism network, where l = 1, 2, …, N, and N represents the total number of layers of the graph isomorphism network; ∥ represents the multi-head attention mechanism, and K represents the number of heads of the multi-head attention; represents the updated eigenvector of node v in the l-th layer of the graph isomorphism network, and ε (l) is a learnable parameter, represents the eigenvector of node v in the (l - 1)-th layer of the graph isomorphism network, MLP is a multi-layer linear perceptron, and Mean is the summation average operation.
[0015] The calculation processes of the velocity spectrum intensity VSI index and the derailment coefficient are shown in the following formulas (5) and (6):
[0016]
[0017] In formula (5), is the velocity response spectrum of the moving measurement point, h is the damping ratio of the bridge, T is the natural period of the bridge, and v train is the train speed; in formula (6), P and Q are respectively the vertical and lateral components of the wheel-rail force, N and F are respectively the normal and tangential forces at the wheel-rail contact point; θ is the angle formed between the tangential force F and the horizontal plane.
[0018] Advantages of the present invention:
[0019] (1) By introducing the graph isomorphism network model, the present invention significantly enhances the adaptability to unknown engineering scenarios, enabling the model to learn from limited training data and generalize to different bridge structures, train speeds, and seismic conditions, overcoming the problems of the traditional neural network's dependence on training data and poor adaptability to new scenarios.
[0020] (2) Compared with the traditional method, the present invention effectively reduces the demand for computing resources through the graph structure representation method and the self-evolution mechanism of the graph isomorphism network model. The graph isomorphism network model does not need to be retrained when dealing with complex structures and diverse working conditions, and can dynamically adjust the topological structure, thereby reducing the actual calculation cost.
[0021] (3) In the multi-scenario seismic operation safety assessment of the present invention, the velocity spectrum intensity (VSI index) and the derailment coefficient can be used as the core assessment indicators to provide more accurate train operation safety assessment results. Especially in emergencies such as earthquakes, the graph isomorphism network model has the ability to respond quickly and can analyze the operation safety of bridges and trains in real time.
[0022] (4) The present invention is not only applicable to vehicle-bridge coupling systems with different structural forms, but also can adapt to safety assessments in the case of structural damage, such as scenarios of voids in track slabs and damage to sliding layers. Through the self-evolution mechanism, the graph isomorphism network model can provide reliable safety assessment support in complex actual engineering environments and has good adaptability and scalability.
[0023] (5) The present invention greatly simplifies the modeling process of the vehicle-bridge coupling system, enables more convenient safety assessment of the vehicle-bridge coupling system, and provides convenience for practical applications.
[0024] In summary, the present invention shows significant advantages in terms of accuracy, adaptability, and resource utilization, and is especially suitable for high-demand track engineering safety assessments, with high practical application value. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 is a schematic structural diagram of a vehicle-bridge coupling system.
[0026] Figure 2 is a flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0027] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0028] See Figure 1 , in the vehicle-bridge coupling system, a train node is arranged on each carriage of the train 1, a plurality of bridge nodes are arranged on the bridge 3 between adjacent bridge piers 2, and a pier node is arranged on each bridge pier 2.
[0029] See Figure 2 , a train operation safety assessment method for a vehicle-bridge coupling system based on physical data driving specifically includes the following steps:
[0030] (1) Perform graph structuring on the vehicle-bridge coupling system to generate a graph structure G = (V, E), where V represents the nodes in the graph structure G and E represents the edges connecting the nodes. The nodes in the graph structure G include train nodes, bridge nodes, and pier nodes in the vehicle-bridge coupling system. Each node is assigned a feature vector and normalized.
[0031] The feature vector X of the train node v,train is the train speed, i.e., X b,train =(v train ,…) T , v train represents the train speed at different times, and T represents the total sampling time. The feature vector X of the bridge node v,bridge is the track irregularity sequence, i.e., X v,bridge =(r v , r a , r g , r c ,…) T , r v represents the vertical profile irregularity data of the track at different times, r a represents the alignment irregularity data of the track at different times, r g represents the gauge irregularity data of the track at different times, r c represents the lateral irregularity data of the track at different times. The feature vector X of the pier node v,pier is the seismic acceleration sequence, i.e., X v,pier =(a x , a y , a z ,…) T , a x represents the longitudinal seismic acceleration, a y represents the lateral seismic acceleration, a z represents the vertical seismic acceleration;
[0032] (2) Extract the features of the structure of the vehicle-bridge coupling system. The extracted features include spatial topological features and material properties. That is, obtain the graph structure data and its corresponding adjacency matrix through the graph representation method, and fill the stiffness of the corresponding edges into the adjacency matrix to obtain the stiffness adjacency matrix. The stiffness adjacency matrix is shown in the following formula (7):
[0033]
[0034] In formula (7), K TT is the stiffness of the coupler connecting the train cars, K TB is the virtual spring stiffness between the wheel and the rail, K BB is the flexural stiffness of the bridge itself, K BP is the stiffness of the bridge bearing;
[0035] (3) Construct a graph isomorphism network model and train it. The feature vectors of the nodes are dynamically updated based on the graph isomorphism network model, and all relevant data used for training the model are generated by finite element model simulation;
[0036] The graph isomorphism network model includes an input layer and an attention module. The input layer encodes the feature vectors X v,train , X v,bridge , X v,pier of each node in the input graph structure G. The attention module includes multi-head attention, point-wise feed-forward network, residual module, aggregation module, and update module. The multi-head attention is used to calculate the attention coefficient of each node, as shown in the following formulas (1) and (2). Then, after the point-wise feed-forward network performs a non-linear transformation, the residual module performs a residual connection operation. The aggregation module aggregates the features of its neighbor nodes according to the attention coefficient of each node, as shown in the following formula (3). Finally, the update module updates the nodes, as shown in the following formula (4):
[0037]
[0038] In formulas (1)-(4), e vu represents the edge feature between node v and its adjacent node u, and e vi represents the edge feature between node v and its adjacent node i; N(v) represents the set of neighbor nodes of node v; W represents the weight matrix; linear represents the linear layer; LeakyReLu and σ are both activation functions used to introduce non-linear characteristics; x v represents the feature vector of node v, and x u represents the feature vector of node u; α vu represents the attention coefficient between node pair (v, u), that is, node u aggregates the feature vector to node v according to the ratio of a vu ; represents the information aggregated from all neighbor nodes of node v in the l-th layer of the graph isomorphism network, where l = 1, 2, …, N, and N represents the total number of layers of the graph isomorphism network; ∥ represents the multi-head attention mechanism, and K represents the number of heads of the multi-head attention; represents the updated feature vector of node v in the l-th layer of the graph isomorphism network, and ε (l) is a learnable parameter, represents the feature vector of node v in the (l - 1)-th layer of the graph isomorphism network, MLP is a multi-layer linear perceptron, and Mean represents the sum-average operation;
[0039] (4) Input the graph structure into the trained graph isomorphism network model for train operation safety assessment, and output safety assessment indicators. The safety assessment indicators include the velocity spectrum intensity VSI index and the derailment coefficient. The calculation process is shown in the following formulas (5) and (6):
[0040]
[0041] In formula (5), is the velocity response spectrum of the moving measurement point, h is the damping ratio of the bridge, T is the natural period of the bridge, and v train is the train speed. Since the natural period of the bridge structure is generally 0.1 - 2.5 s, the upper and lower limits of the integral are also set to 0.1 - 2.5 s here. In formula (6), P and Q are the vertical and horizontal components of the wheel-rail force respectively, N and F are the normal and tangential forces at the wheel-rail contact point respectively; θ is the angle formed between the tangential force F and the horizontal plane.
[0042] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A driving safety assessment method for a vehicle-bridge coupling system based on physical data-driven, characterized in that: Specifically, it includes the following steps: (1) Structure the vehicle-bridge coupling system into a graph to generate a graph structure G = (V, E), where V represents the nodes in the graph structure G, and E represents the edges connecting the nodes in the graph structure G. The nodes in the graph structure G include train nodes, bridge nodes, and pier nodes in the vehicle-bridge coupling system, and each node is assigned a feature vector; (2) Construct and train a graph isomorphism network model; (3) Input the graph structure into the trained graph isomorphism network model for train operation safety assessment, and output safety assessment indicators. The safety assessment indicators include the velocity spectrum intensity VSI indicator and the derailment coefficient.
2. The driving safety assessment method for the axle coupling system based on physical data-driven according to claim 1, wherein: The feature vector X of the train node v,train is the train speed, i.e., X v,train =(v train ,…) T , v train represents the train speeds at different times, and T represents the total sampling time; The feature vector X of the bridge node v,bridge is the track irregularity sequence, i.e., X v,bridge =(r v , r a , r g , r c , …) T , r v represents the vertical profile irregularity data of the track at different times, r a represents the alignment irregularity data of the track at different times, r g represents the gauge irregularity data of the track at different times, r c represents the lateral irregularity data of the track at different times; the feature vector X v,pier of the pier node is the seismic acceleration sequence, that is, X v,pier =(a x , a y , a z , …) T , a x represents the longitudinal seismic acceleration, a y represents the lateral seismic acceleration, a z represents the vertical seismic acceleration.
3. The driving safety assessment method for the axle coupling system based on physical data-driven according to claim 2, characterized in that: The described graph isomorphism network model includes an input layer and an attention module. The input layer encodes the feature vectors X v,train , X v,bridge , X v,pier of each node in the input graph structure G. After that, the attention module first calculates the attention coefficient of each node, then aggregates the features of its neighbor nodes according to the attention coefficient of each node, and finally updates the nodes to obtain the feature vectors of the updated nodes.
4. The driving safety assessment method for the axle coupling system based on physical data-driven according to claim 3, characterized in that: The attention module includes multi-head attention, a position-wise feed-forward network, a residual module, an aggregation module, and an update module. The multi-head attention is used to calculate the attention coefficient of each node, as shown in the following equations (1) and (2). Then, after the position-wise feed-forward network performs a non-linear transformation, the residual module performs a residual connection operation. Next, the aggregation module aggregates the features of its neighbor nodes according to the attention coefficient of each node, as shown in the following equation (3). Finally, the update module updates the nodes, as shown in the following equation (4): In formulas (1)-(4), e vu represents the edge feature between node v and its adjacent node u, and e vi represents the edge feature between node v and its adjacent node i; N(v) represents the set of neighbor nodes of node v; W represents the weight matrix; linear represents the linear layer; LeakyReLu and σ are both activation functions used to introduce non-linearity; x v represents the feature vector of node v, and x u represents the feature vector of node u; α vu represents the attention coefficient between node pair (v, u), that is, node u aggregates the feature vector to node v according to the ratio of α vu ; represents the information aggregated from all neighbor nodes of node v in the l-th layer graph isomorphism network, where l = 1, 2, …, N, and N represents the total number of layers of the graph isomorphism network; ∥ represents the multi-head attention mechanism, and K represents the number of heads of the multi-head attention; represents the updated feature vector of node v in the l-th layer graph isomorphism network, and ε (l) is a learnable parameter, represents the feature vector of node v in the (l - 1)-th layer graph isomorphism network, MLP is the multi-layer linear perceptron, and Mean is the sum-average operation.
5. The driving safety assessment method of the axle coupling system based on physical data driving according to claim 1, wherein: The calculation processes of the velocity spectrum intensity VSI indicator and the derailment coefficient are shown in the following equations (5) and (6): In Equation (5), is the velocity response spectrum of the moving measurement point, h is the damping ratio of the bridge, T is the natural period of the bridge, and v train is the train speed; in Equation (6), P and Q are the vertical and lateral components of the wheel-rail force respectively, n and F are the normal and tangential forces at the wheel-rail contact point respectively; θ is the angle formed between the tangential force F and the horizontal plane.