Traffic safety analysis method and system for large-span cable-stayed bridge of high-speed rail

Abstract

The invention belongs to the technical field of bridge engineering and rail transit, and particularly relates to a high-speed rail long-span cable-stayed bridge traffic safety analysis method and system. Comprising the following steps: establishing a physical constraint relation model between vehicle body vertical acceleration and secondary suspension vertical force resultant force; obtaining a vehicle dynamic response predicted value by using the German track irregularity spectrum, the track deformation quantity and a PI-LSTM model; obtaining a prediction interval of the vehicle dynamic response predicted value by using a cGANs model, the random noise variable, the real error value and the vehicle dynamic response predicted value; the maximum value between the peak value of the prediction interval and the vehicle dynamic response prediction value is obtained; comparing the maximum value with a standard limit value to obtain a comparison result; the comparison result is used for traffic safety evaluation. According to the method, high-precision and high-robustness full-life-cycle evaluation is provided for the traffic safety of the large-span cable-stayed bridge of the high-speed rail.

Description

Technical Field

[0001] The present invention belongs to the technical field of bridge engineering and rail transportation, and particularly relates to a driving safety analysis method and system for a high-speed railway long-span cable-stayed bridge. Background Art

[0002] Due to their strong spanning capacity and lightweight structure, long-span cable-stayed bridges have become the preferred bridge type for high-speed rail crossings over rivers and canyons. However, their low damping and highly flexible structural characteristics easily couple with external excitations such as track irregularities, wind loads, and temperature fluctuations when trains pass at high speeds, causing significant vehicle-bridge coupled vibrations that directly impact driving safety and ride comfort. Traditional safety analysis of long-span cable-stayed bridges on high-speed rail relies on finite element simulation or measured data regression, which presents the following problems: First, the simplified physical model leads to insufficient accuracy; second, the neural network prediction lacks physical constraints, limiting its generalization ability; and third, it is difficult to quantify the uncertainty of the prediction results, making it impossible to provide a reliable safety assessment range. Summary of the Invention

[0003] In order to solve the above-mentioned technical defects existing in the prior art, the present invention combines a neural network embedded with physical constraints with a generative adversarial network to achieve high-precision prediction and uncertainty quantification of vehicle dynamic response, and ultimately complete driving safety assessment.

[0004] The present invention is achieved through the following technical solutions: Firstly, a driving safety analysis method for a high-speed railway long-span cable-stayed bridge is proposed, comprising the following steps: establishing a physical constraint relationship model between the vertical acceleration of the vehicle body and the resultant force of the secondary suspension vertical force; obtaining a vehicle dynamic response prediction value by using the German track irregularity spectrum, track deformation and PI-LSTM model; the physical constraint relationship model is embedded in the physical information neural network of the PI-LSTM model; obtaining a prediction interval of the vehicle dynamic response prediction value by using a cGANs model, random noise variables, a true error value and the vehicle dynamic response prediction value; the true error value is the error value between the vehicle dynamic response prediction value and the true value; obtaining a maximum value between the peak value of the prediction interval and the vehicle dynamic response prediction value; comparing the maximum value with the specification limit to obtain a comparison result; and the comparison result is used for driving safety assessment.

[0005] Secondly, a driving safety analysis system for a high-speed railway large-span cable-stayed bridge is proposed, including: a model building module, used to establish a physical constraint relationship model between the vertical acceleration of the vehicle body and the resultant force of the secondary suspension vertical force; a response prediction module, used to obtain the vehicle dynamic response prediction value by using the German track irregularity spectrum, track deformation and PI-LSTM model; the physical constraint relationship model is embedded in the physical information neural network of the PI-LSTM model; an interval acquisition module, used to obtain the prediction interval of the vehicle dynamic response prediction value by using the cGANs model, random noise variables, true error values ​​and the vehicle dynamic response prediction value; the true error value is the error value between the vehicle dynamic response prediction value and the true value; a maximum value acquisition module, used to obtain the maximum value between the peak value of the prediction interval and the vehicle dynamic response prediction value; a numerical comparison module, used to compare the maximum value with the specification limit to obtain a comparison result; the comparison result is used for driving safety assessment.

[0006] In a third aspect, a computer device is proposed, comprising a memory, a processor, and a transceiver that are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive data, and the processor is used to read the computer program and execute a method for analyzing driving safety of a high-speed railway large-span cable-stayed bridge as described in the first aspect.

[0007] In a fourth aspect, a computer-readable storage medium is proposed, on which instructions are stored. When the instructions are run on a computer, a method for analyzing driving safety of a high-speed railway large-span cable-stayed bridge as described in the first aspect is executed.

[0008] In the fifth aspect, a computer program product comprising instructions is proposed. When the instructions are executed on a computer, the computer is caused to execute a method for analyzing driving safety of a high-speed railway large-span cable-stayed bridge as described in the first aspect; the computer comprises: a general-purpose computer, a special-purpose computer or a programmable device.

[0009] Compared with the existing technology, the present invention has the following advantages and beneficial effects: through the deep integration of physical constraint models and LSTM, the dynamic mechanism of the vehicle suspension system is embedded in the neural network, which significantly improves the prediction accuracy of high-frequency vibration and reduces data dependence; combined with conditional generative adversarial networks (cGANs), time-varying prediction intervals are dynamically generated to achieve efficient modeling of non-Gaussian errors and real-time uncertainty quantification; further, through the physical correlation of multi-source data and prediction interval peak extraction, the specification limit comparison is upgraded to extreme value risk-oriented assessment, which is compatible with design simulation and operation monitoring data, and solves the problems of "model distortion, interval rigidity, and one-sided assessment" of traditional methods, providing a high-precision and highly robust full life cycle assessment solution for the driving safety of high-speed railway large-span cable-stayed bridges. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings: Figure 1 A schematic diagram of the overall technical logic architecture of a method for analyzing driving safety on a high-speed railway long-span cable-stayed bridge provided in Example 1 of the present invention.

[0011] Figure 2 A schematic flow chart of a method for analyzing driving safety on a high-speed railway long-span cable-stayed bridge provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with the examples. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention. The embodiments described below are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0013] Example 1: A method for analyzing the driving safety of a high-speed railway long-span cable-stayed bridge is proposed. The overall technical logic of this method is as follows: Figure 1As shown, the driving safety analysis of a high-speed railway long-span cable-stayed bridge is divided into three stages. The first stage establishes a physical constraint relationship model between the vehicle body vertical acceleration and the resultant secondary suspension vertical force. The second stage uses the PI-LSTM model to predict the vehicle dynamic response. This involves two levels: In the first level, the physical constraint relationship model established in the first stage is embedded in the PI-LSTM model's physical information neural network to guide network training. The German track irregularity spectrum and track deformation are used as inputs to the PI-LSTM model. After learning and analyzing, the PI-LSTM model outputs predicted values ​​for the vehicle body vertical acceleration and the resultant secondary suspension vertical force. In the second level, the German track irregularity spectrum, track deformation, and predicted vehicle body vertical acceleration are used as inputs to the PI-LSTM model. After learning and analyzing, the PI-LSTM model outputs predicted values ​​for the vehicle dynamic response. Phase 3: Analyzing driving safety using the cGANs model. The predicted vehicle dynamic response obtained in Phase 2 is used as conditional information. This conditional information and random noise are fed into the Generator network in the cGANs model, which outputs a prediction error. The error between the predicted and true vehicle dynamic response, the prediction error generated by the Generator network, and the conditional information are fed into the Discriminator network, which outputs a target probability. The Generator and Discriminator networks compete with each other during training. The trained Generator network is then used to calculate prediction intervals for three evaluation indicators: vehicle acceleration, derailment coefficient, and wheel load reduction rate. The train's safety and comfort are then evaluated based on regulatory limits, enhancing the credibility of the predictions and evaluations.

[0014] The specific steps of this method are as follows: Figure 2 Shown, including: Step 1: Establish a physical constraint relationship model between the vehicle body vertical acceleration and the resultant secondary suspension vertical force.

[0015] Before establishing the physical constraint model described in this step, a physical model of the vehicle traveling on the bridge must be created for dynamic analysis. This physical model should consist of at least a vehicle body, two bogies, and four wheelsets. Using the vehicle body as the research object, it is clear that changes in the secondary suspension vertical force inevitably affect the vehicle's acceleration.

[0016] For ease of understanding, this embodiment introduces the definition and function of the secondary suspension vertical force. The secondary suspension vertical force refers to the vertical force borne and transmitted by the secondary suspension system (located between the bogie frame and the carbody underframe) in a railway vehicle. The secondary suspension system is a key component connecting the bogie frame and the carbody. Its primary function is to transmit forces between the carbody and the bogie, including vertical, longitudinal, and lateral forces. Vertical force refers to the vertical force transmitted from the carbody to the bogie frame via the secondary suspension. The secondary suspension system supports the weight of the carbody through components such as air springs, ensuring vehicle stability during operation. It isolates and attenuates vibrations transmitted from the frame to the carbody, thereby improving vehicle operation smoothness and ride comfort. During vehicle operation, the secondary suspension vertical force varies with track irregularities and vehicle dynamic response, affecting the vehicle's vertical stability. Therefore, in step 1, this method establishes a physical constraint relationship model between the carbody vertical acceleration and the resultant secondary suspension vertical force to facilitate subsequent analysis of vehicle driving safety.

[0017] Furthermore, a vehicle body dynamics model is established based on the D'Alembert principle and the vehicle physical model. The vehicle body dynamics model is expressed as: (1); among them, M v Indicates the vehicle body mass, represents the vertical acceleration of the vehicle body, g represents the acceleration due to gravity, F ti represents the secondary suspension vertical force, i =1,2,3,4. Expanding formula (1) yields formula (2): (2); among them, F total represents the resultant secondary suspension vertical force. Equation (2) shows that there is a physical relationship between the vehicle body vertical acceleration and the secondary suspension vertical force. That is, the vehicle body vertical acceleration can be controlled by adjusting the secondary suspension vertical force. Therefore, this embodiment uses Equation (2) as a model for the physical constraint relationship between the vehicle body vertical acceleration and the resultant secondary suspension vertical force.

[0018] The first stage of the method is completed by step 1. Next, the second stage of the method is performed.

[0019] Step 2: Embed the physical constraint relationship model into the physical information neural network of the PI-LSTM model.

[0020] Step 3: Obtain the predicted value of vehicle dynamic response using the German track irregularity spectrum, track deformation, and the PI-LSTM model containing the physical neural network.

[0021] Before explaining step 3 in detail, it is necessary to introduce the German track irregularity spectrum and the PI-LSTM model.

[0022] The German Track Irregularity Spectrum is a power spectral density (PSD) function used to describe track geometry. It reflects the energy distribution of vertical and lateral track irregularities at different wavelengths. This spectrum is obtained by measuring actual track irregularity data and fitting it using statistical methods. It is primarily used to assess track smoothness and its impact on train operation.

[0023] The PI-LSTM model is a novel deep learning model that combines a physical information neural network (PINN) with a long short-term memory network (LSTM). The PINN embeds physical laws into a neural network, leveraging known physical laws to constrain network training and reduce the risk of overfitting. The LSTM is a specialized recurrent neural network with memory cells and a gating mechanism that captures and retains key information in time series, effectively addressing long-term dependencies. The PI-LSTM model leverages the strengths of both PINN and LSTM networks by combining them. Specifically, the PINN utilizes physical laws to constrain network training, ensuring that the model's predictions conform to physical laws. The LSTM captures long-term dependencies in time series, thereby enhancing the model's predictive capabilities. This combination not only improves the model's prediction accuracy but also enhances its generalization and noise resistance.

[0024] Based on the above introduction to the German track irregularity spectrum and the PI-LSTM model, the purpose of this step is to use the PI-LSTM model to obtain the predicted value of the vehicle dynamic response.

[0025] As we all know, before using a neural network model for prediction, the model needs to be initialized, including training, verification, and testing. The specific steps include: Step 3.1: Collect sensor data from multiple static deflection measurement points on the bridge to obtain a bridge health monitoring dataset.

[0026] Static deflection refers to the deformation of a structure under static forces, typically manifesting as vertical displacement of a specific point on the structure. Static deflection measurement points are used to measure this displacement and are typically located at key locations within the structure, such as the midspan of a beam or near its supports.

[0027] Step 3.2: Use cubic spline interpolation to process the health monitoring data set to obtain the full bridge alignment of the bridge.

[0028] Step 3.3: Use Ansys and Simpack co-simulation to establish the bridge-track mapping model and vehicle-track coupling model.

[0029] Among them, the bridge-track mapping model is a model used to study the relationship between bridge deformation and track geometry. Since the deformation of the bridge will directly affect the safety and comfort of train operation, this model can map the vertical or lateral deformation of the bridge to the geometry of the track, thereby predicting the deformation of the track. Therefore, it is often used to analyze the impact of bridge structure deformation on the track geometry. The method to establish a bridge-track mapping model is: First, in Ansys, use the APDL command flow or Workbench interface to establish a rail model. For example, establish a 21m long 60-rail model with a sleeper spacing of 0.6m, steel material, a density of 7850kg / m³, and a Young's modulus of 2.1×10 11 Pa, with a Poisson's ratio of 0.3. Modal analysis was performed on the rail model to obtain the cutoff frequency. The flexible track was imported into Simpack, and after generating the fbi file, the ftr file was prepared. Next, a finite element model of the bridge was created in Ansys and exported to a format recognizable by Simpack, such as STL or STEP. Finally, the rail model and bridge finite element model were imported into Simpack to create a bridge-rail mapping model.

[0030] Furthermore, the vehicle-track coupling model is a dynamic model used to study the interaction between the vehicle and track to assess the stability and safety of train operation and the dynamic response of the track structure. The vehicle-track coupling model treats the vehicle and track as a holistic system, considering the dynamic interaction between them. To establish the vehicle-track coupling model, first, create a vehicle model in Simpack, including modeling components such as wheelsets and frames, and combine them into a complete vehicle model. Then, couple the vehicle model in Simpack with the track model (a rigid-flexible coupling approach can be used, coupling a rigid vehicle with a flexible track). Finally, set simulation parameters such as simulation time and time step, run the simulation, analyze the vehicle's operating performance on the track, such as vehicle body vibration acceleration, wheel-rail force, and derailment coefficient, set vehicle-track contact parameters such as wheel-rail contact stiffness and damping, and output the vehicle-track coupling model.

[0031] Step 3.4: Obtain the track deformation according to the full bridge alignment and bridge-track mapping model.

[0032] Based on the full bridge alignment and bridge-track mapping model, the deformation of the track under bridge deformation was calculated through joint simulation analysis using Ansys and Simpack. Specifically, a static analysis of the bridge's finite element model was performed in Ansys to calculate the displacement and stress fields under the applied deformation conditions. Displacement data for each node of the bridge after deformation was extracted and used as input for subsequent track deformation calculations. In Simpack, a joint simulation analysis was conducted combining the track model and bridge deformation data. Simpack's multi-body dynamics analysis function was used to consider the interaction between the track and bridge and calculate the deformation of the track under bridge deformation.

[0033] Step 3.5: Obtain the vehicle dynamic response based on the full-bridge alignment, vehicle-track coupling model, and track deformation.

[0034] The vehicle dynamic response includes: vehicle body vertical acceleration, left wheel vertical wheel-rail force, right wheel vertical wheel-rail force and right wheel lateral wheel-rail force.

[0035] The specific implementation method of this step is as follows: import the track deformation data into the track model of Simpack (this can be achieved by modifying the initial geometry of the track model or dynamically applying displacement boundary conditions in the simulation); set the simulation parameters in Simpack, such as vehicle running speed and simulation time step; start the simulation, and Simpack will calculate the dynamic response of the vehicle under track deformation conditions based on the input track deformation and vehicle dynamics model, including vehicle body vibration acceleration, wheel-rail force, etc.

[0036] Step 3.6: Substitute the vehicle body vertical acceleration into the physical constraint relationship model to calculate the resultant vertical force of the secondary suspension.

[0037] Step 3.7: Use the track deformation, vehicle dynamic response, and secondary suspension vertical force to create the first training set, the first validation set, and the first test set.

[0038] Step 3.8: Initialize the PI-LSTM model using the first training set, the first validation set, and the first test set.

[0039] Specifically, the PI-LSTM model is trained using the first training set, and the PI-LSTM model is refined and finalized using the first validation set. Finally, the accuracy of the PI-LSTM model is evaluated using the first test set, thereby completing the initialization process.

[0040] After initializing the PI-LSTM model through steps 3.1 through 3.8, the PI-LSTM model can be used to predict vehicle dynamic response. Based on the above introduction to the PI-LSTM model, it is important to note that the physical constraint relationship model established in step 1 is embedded in the PI-LSTM model's physical information neural network, thereby improving prediction accuracy and physical consistency.

[0041] The PI-LSTM model is used to predict vehicle dynamic response, which includes the following steps: First, the German track irregularity spectrum and track deformation are used as input to the PI-LSTM model. After learning and analyzing, the PI-LSTM model outputs the predicted vertical acceleration of the vehicle body. It should be noted that the German track irregularity spectrum and track deformation are arranged in time series order to form an input matrix. After entering the PI-LSTM model, the input matrix is ​​processed by multiple LSTM units. Each LSTM unit receives the input of the current time step and the hidden state and cell state of the previous time step. Within the LSTM unit, a series of gating mechanisms (input gate, forget gate, output gate) control the flow of information, thereby updating the hidden state and cell state. The output of the LSTM unit passes through a linear layer, which maps the LSTM output to the predicted output space.

[0042] The German track irregularity spectrum, track deformation, and predicted vehicle vertical acceleration are then fed into the PI-LSTM model. After learning and analyzing, the model outputs predicted wheel-rail forces. These wheel-rail force predictions include the left wheel vertical wheel-rail force, the right wheel vertical wheel-rail force, and the right wheel lateral wheel-rail force. Similarly, the German track irregularity spectrum, track deformation, and predicted vehicle vertical acceleration are input to the PI-LSTM model and arranged in time series order to form an input matrix. Once the input matrix enters the PI-LSTM model, it is processed by multiple LSTM units. Each LSTM unit receives the input of the current time step and the hidden state and cell state of the previous time step. Within the LSTM unit, a series of gating mechanisms (input gate, forget gate, and output gate) control the flow of information, thereby updating the hidden state and cell state. The output of the LSTM unit passes through a linear layer, which maps the LSTM output to the predicted output space.

[0043] Finally, the derailment coefficient prediction value is calculated based on the wheel-rail force prediction value and wheel load reduction rate prediction value ,in, Dc pred represents the predicted value of the derailment coefficient, Wlc pred represents the predicted value of wheel load reduction rate, represents the predicted value of the vertical wheel-rail force on the right wheel, represents the predicted value of the right wheel lateral wheel-rail force, Indicates the predicted value of the left wheel vertical wheel-rail force.

[0044] After learning and analysis by the PI-LSTM model, the vehicle dynamic response prediction values ​​obtained include: vehicle body vertical acceleration prediction value, derailment coefficient prediction value and wheel load reduction rate prediction value.

[0045] In addition, it should be noted that in the PI-LSTM model prediction process, its loss function consists of two parts: the data loss caused by the data prediction itself And the physical loss function between the output Among them, the loss function expression corresponding to data loss is: (3); The loss function expression corresponding to the physical loss is: (4). In formula (3) and formula (4), LOSS data Indicates data loss, LOSS physics Indicates physical loss, N Indicates the total amount of data. Indicates the i The predicted value of vehicle vertical acceleration, Indicates the i The true value of the vehicle body vertical acceleration, Indicates the i The predicted value of the resultant vertical force of the secondary suspension, Indicates the i The true value of the resultant vertical force of the second suspension, T Indicates transpose.

[0046] To address the data imbalance problem and improve the model's robustness to outliers, this example utilizes the contribution of hyperparameters to physical loss and establishes the model's overall loss function: (5); In formula (5), Loss total represents the total loss, represents a hyperparameter.

[0047] Step 4: Use the cGANs model, random noise variables, true error values, and vehicle dynamic response prediction values ​​to obtain the prediction interval of the vehicle dynamic response prediction value.

[0048] Before we delve into step 3, we need to introduce the cGANs model. The cGANs model is a variant of generative adversarial networks that incorporates conditional information to control the characteristics of generated data. The core of the cGANs model lies in the fact that both the generator and the discriminator receive additional conditional information. The generator receives a random noise vector and the conditional information to generate synthetic data that meets the conditions. The discriminator receives real data and the conditional information to determine whether the data is real. The generator and discriminator are continuously optimized through adversarial training: the generator attempts to generate more realistic data to deceive the discriminator, while the discriminator strives to distinguish between real data and generated data.

[0049] Based on the above basic introduction to the cCANs model, the purpose of this step is to use the cCANs model to control the feasibility of the vehicle dynamic response prediction value output by the PI-LSTM model, thereby improving the accuracy of the safety analysis results.

[0050] Similarly, before using the cGANs model to obtain the prediction interval of the vehicle dynamic response prediction value, the cGANs model also needs to be initialized. You can refer to the initialization method of the PI-LSTM model, including model training, verification, and testing.

[0051] First, the vehicle dynamic response prediction values ​​are used to establish a second training set, a second validation set, and a second test set.

[0052] Then, the cGANs model is initialized using the second training set, the second validation set, and the second test set.

[0053] Specifically, the cGANs model is trained using the second training set. The generator and discriminator compete with each other during the training process until the generator can generate data that is highly similar to the original data distribution, while the discriminator cannot effectively distinguish between real samples and generated samples, and its output probability is close to 0.5, reaching Nash equilibrium; at the same time, the second validation set is used to refine and finalize the cGANs model; finally, the second test set is used to evaluate the accuracy of the cGANs model, thus completing the initialization process.

[0054] After the cGANs model is initialized, the prediction interval of the vehicle dynamic response prediction value can be obtained using the cGANs model. Specifically, the random noise variable, the true error value, and the vehicle dynamic response prediction value are used as inputs to the cGANs model. After analysis and processing by the cGANs model, the prediction interval of the vehicle dynamic response prediction value is output. The true error value is the error between the predicted value and the true value of the vehicle dynamic response. The output prediction intervals of the vehicle dynamic response prediction value are: the prediction interval of the vehicle body vertical acceleration prediction value, the prediction interval of the derailment coefficient prediction value, and the prediction interval of the wheel load reduction rate prediction value.

[0055] After outputting the above prediction intervals, the following steps are performed for each prediction interval: Step 4.1: Set the confidence level for the prediction interval.

[0056] The confidence level in this embodiment is 95%.

[0057] Step 4.2: Filter out the vehicle dynamic response prediction values ​​whose probability of falling into the prediction interval is less than the confidence value.

[0058] Step 4.3: Get the logarithmic error of the filtered values.

[0059] Step 4.4: Determine whether the logarithmic error falls within the specified numerical range; if so, the screened value is regarded as a credible value; otherwise, the screened value is regarded as an unreliable value.

[0060] For predicted values ​​outside the prediction interval, their logarithmic error is calculated. If the logarithmic error is between 0 and 2, the predicted value is still credible even if it is outside the prediction interval. Credible values ​​are predicted values ​​whose logarithmic error falls within the specified interval. These predicted values ​​within the specified interval have a small relative error from the true value and are therefore retained. Unreliable values, on the other hand, are predicted values ​​whose logarithmic error falls outside the specified interval. These values ​​are likely unreliable due to their large error and are therefore discarded.

[0061] Step 5: Obtain the maximum value between the peak value of the prediction interval and the predicted value of the vehicle dynamic response.

[0062] The purpose of this step is to evaluate the track state and ensure the safety of train operation by obtaining the maximum value between the peak value of the prediction interval and the predicted value of the vehicle dynamic response. Among them, the prediction interval is the prediction interval obtained in step 3. By comparing the peak value of the prediction interval and the predicted value of the vehicle dynamic response, key factors that may affect the safety of train operation can be identified, such as vehicle acceleration, wheel weight reduction rate, derailment coefficient, etc. These indicators are significantly correlated with parameters such as the height, track direction, and superelevation of the track. The prediction model can effectively predict the dynamic response of the freight car, and the correlation coefficient between the predicted value and the simulation value is above 0.8, which is a strong correlation. In this way, the response of different vehicle models to the track state can be more accurately reflected, thereby improving the accuracy and consistency of the track state evaluation and promptly eliminating situations that may cause train derailment or other safety problems.

[0063] Step 6: Compare the maximum value with the specification limit and obtain the comparison result.

[0064] After the above step 3, the prediction intervals of the three evaluation indicators have been obtained. Now it is necessary to evaluate the final driving safety. The evaluation criteria are mainly based on the current "High-speed Railway Design Code" (TB10621-2014), which requires the vertical acceleration of the vehicle body. ≤1.3m / s 2 , derailment coefficient ≤0.8, wheel load reduction rate If all three evaluation indicators meet the limit, the vehicle can be evaluated as safe to drive; otherwise, it is evaluated as exceeding the limit and requires maintenance and repair.

[0065] Example 2: Corresponding to Example 1, this example provides a high-speed railway long-span cable-stayed bridge driving safety analysis system, including: Model building module, used to establish the physical constraint relationship model between the vertical acceleration of the vehicle body and the resultant vertical force of the secondary suspension; a response prediction module for obtaining a predicted value of vehicle dynamic response using the German track irregularity spectrum, track deformation, and a PI-LSTM model; wherein the physical constraint relationship model is embedded in the physical information neural network of the PI-LSTM model; An interval acquisition module is configured to acquire a prediction interval of the vehicle dynamic response prediction value using a cGANs model, a random noise variable, a true error value, and the vehicle dynamic response prediction value; the true error value is an error value between the vehicle dynamic response prediction value and the true value; A maximum value acquisition module, configured to acquire a maximum value between a peak value of the prediction interval and the vehicle dynamic response prediction value; The numerical comparison module is used to compare the maximum value with the specification limit to obtain a comparison result; the comparison result is used for driving safety assessment.

[0066] Furthermore, the model building module includes: A vehicle model creation unit, configured to establish a vehicle physical model; the vehicle physical model comprises a vehicle body, a plurality of bogies, and a plurality of wheel sets; a mechanical model creation unit, configured to establish a vehicle body dynamics model based on the D'Alembert principle and the vehicle physical model; A mechanical model processing unit, configured to process the vehicle dynamics model to obtain the physical constraint relationship model; The vehicle body dynamics model is expressed as: (1); among them, M v Indicates the vehicle body mass, represents the vertical acceleration of the vehicle body, g represents the acceleration due to gravity, F ti represents the secondary suspension vertical force, i =1,2,3,4; The expression of the physical constraint relationship model is: (2); among them, F total It represents the resultant vertical force of the secondary suspension.

[0067] Furthermore, the high-speed railway long-span cable-stayed bridge driving safety analysis system further includes: a data acquisition module, configured to collect sensor data from multiple static deflection measurement points on the bridge to obtain a health monitoring data set for the bridge; a data processing module, configured to process the health monitoring data set using cubic spline interpolation to obtain the full bridge alignment of the bridge; The joint simulation module is used to establish a bridge-track mapping model and a vehicle-track coupling model through joint simulation; A track deformation acquisition module, configured to acquire the track deformation according to the full bridge alignment and the bridge-track mapping model; a dynamic response acquisition module, configured to acquire a vehicle dynamic response based on the full-bridge alignment, the vehicle-rail coupling model, and the track deformation; the vehicle dynamic response including: vehicle body vertical acceleration, left wheel vertical wheel-rail force, right wheel vertical wheel-rail force, and right wheel lateral wheel-rail force; a vertical force acquisition module, configured to acquire a resultant vertical force of the secondary suspension according to the vehicle body vertical acceleration and the physical constraint relationship model; a first data partitioning module, configured to establish a first training set, a first validation set, and a first test set using the track deformation, the vehicle dynamic response, and the secondary suspension vertical force; The first model processing module is used to initialize the PI-LSTM model using the first training set, the first validation set and the first test set.

[0068] Furthermore, the vehicle dynamic response prediction value includes: a vehicle body vertical acceleration prediction value, a derailment coefficient prediction value, and a wheel load reduction rate prediction value; The response prediction module includes: an acceleration prediction unit, configured to obtain a predicted value of the vehicle body vertical acceleration based on a German track irregularity spectrum, the track deformation, and the PI-LSTM model; a wheel-rail force prediction unit, configured to obtain the wheel-rail force prediction value based on the German track irregularity spectrum, the track deformation, the vehicle body vertical acceleration prediction value, and the PI-LSTM model; the wheel-rail force prediction value includes: a left wheel vertical wheel-rail force prediction value, a right wheel vertical wheel-rail force prediction value, and a right wheel lateral wheel-rail force prediction value; The wheel-rail force processing unit is used to obtain the derailment coefficient prediction value and the wheel load reduction rate prediction value according to the wheel-rail force prediction value.

[0069] Furthermore, the high-speed railway long-span cable-stayed bridge driving safety analysis system further includes: a loss function creation module for adjusting the contribution of physical losses using hyperparameters to establish an overall loss function; The expression of the total loss function is: (5); In formula (5), Loss total represents the total loss, LOSS data Indicates data loss, LOSS physics Indicates physical loss, represents a hyperparameter; (3); (4); In formula (3) and formula (4), LOSS data Indicates data loss, LOSS physics Indicates physical loss, N Indicates the total amount of data. Indicates the i The predicted value of vehicle vertical acceleration, Indicates the i The true value of the vehicle body vertical acceleration, Indicates the i The predicted value of the resultant vertical force of the secondary suspension, Indicates the iThe true value of the resultant vertical force of the second suspension, T Indicates transpose.

[0070] a second data partitioning module, configured to establish a second training set, a second validation set, and a second test set using the vehicle dynamic response prediction value; The second model processing module is used to initialize the cGANs model using the second training set, the second validation set and the second test set.

[0071] A confidence setting module, used to set the confidence level of the prediction interval; a numerical screening module, configured to screen out values ​​of the vehicle dynamic response prediction values ​​whose probability of falling within the prediction interval is less than the confidence level; A logarithmic error acquisition module is used to obtain the logarithmic error of the filtered value; The error analysis module is used to determine whether the logarithmic error falls within a specified numerical interval; if so, the screened value is used as a credible value; otherwise, the screened value is used as an unreliable value.

[0072] The working principles and effects of the above functional modules and functional units refer to the above embodiment 1.

[0073] Example 3: Based on the method provided in Example 1 and the system provided in Example 2, this example provides a computer device for executing the method described in Example 1 or any method potentially related to Example 1. The computer device comprises a memory, a processor, and a transceiver, which are communicatively connected in sequence. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the method described in Example 1 or any method potentially related to Example 1. Specifically, the memory may include, but is not limited to, random-access memory (RAM), read-only memory (ROM), flash memory, first-input first-output (FIFO), and / or first-input last-output (FILO) memory. The processor may include, but is not limited to, a microprocessor from the STM32F105 series. Furthermore, the computer device may include, but is not limited to, a power module, a display screen, and other necessary components.

[0074] The working process, working details and technical effects of the aforementioned computer device provided in this embodiment can be referred to the method described in Example 1 or any method that may be involved in Example 1, and will not be repeated here.

[0075] Example 4: This example provides a computer-readable storage medium that stores the method described in Example 1 or any method that may be related to the method described in Example 1, that is, the computer-readable storage medium stores instructions that, when executed on a computer, execute the method described in Example 1 or any method that may be related to the method described in Example 1. The computer-readable storage medium refers to a carrier for storing data, and may include, but is not limited to, computer-readable storage media such as a floppy disk, an optical disk, a hard disk, a flash memory, a USB flash drive, and / or a memory stick. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0076] The working process, working details and technical effects of the aforementioned computer-readable storage medium provided in this embodiment can be referred to the method described in Example 1 or any method that may be related to Example 1, and will not be repeated here.

[0077] Embodiment 5: This embodiment provides a computer program product comprising instructions, which, when executed on a computer, causes the computer to perform the method described in Embodiment 1 or any method that may be related to the method described in Embodiment 1. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0078] It should be understood that the terms "system," "device," "unit," and / or "module" used in this specification are a method for distinguishing different components, elements, parts, portions, or assemblies at different levels. However, if other terms can achieve the same purpose, the terms may be replaced by other expressions.

[0079] As used in this specification and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not refer to the singular but also include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0080] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for analyzing driving safety of a high-speed railway long-span cable-stayed bridge, characterized in that: The following steps are involved: Establish a physical constraint relationship model between the vehicle body vertical acceleration and the resultant vertical force of the secondary suspension; The vehicle dynamic response prediction value is obtained using the German track irregularity spectrum, track deformation and PI-LSTM model; the physical constraint relationship model is embedded in the physical information neural network of the PI-LSTM model; Obtaining a prediction interval of the vehicle dynamic response prediction value using a cGANs model, a random noise variable, a true error value, and the vehicle dynamic response prediction value; the true error value is an error value between the vehicle dynamic response prediction value and the true value; Obtaining a maximum value between a peak value of the prediction interval and the vehicle dynamic response prediction value; The maximum value is compared with the specification limit to obtain a comparison result; the comparison result is used for driving safety assessment.

2. The method for analyzing driving safety of a high-speed railway long-span cable-stayed bridge according to claim 1, characterized in that: Establishing the physical constraint relationship model includes the following steps: Establishing a vehicle physical model; the vehicle physical model includes a vehicle body, multiple bogies, and multiple wheel sets; Establishing a vehicle body dynamics model based on the D'Alembert principle and the vehicle physical model; Processing the vehicle dynamics model to obtain the physical constraint relationship model; The vehicle body dynamics model is expressed as: (1); among them, M v Indicates the vehicle body mass, represents the vertical acceleration of the vehicle body, g represents the acceleration due to gravity, F ti represents the secondary suspension vertical force, i =1,2,3,4; The expression of the physical constraint relationship model is: (2); among them, F total It represents the resultant vertical force of the secondary suspension.

3. A method for analyzing driving safety of a high-speed railway long-span cable-stayed bridge according to claim 1 or 2, characterized in that: Before obtaining the predicted value of vehicle dynamic response, the following steps are also included: Collecting sensor data from multiple static deflection measurement points on the bridge to obtain a health monitoring data set of the bridge; Processing the health monitoring data set using cubic spline interpolation to obtain the full bridge alignment of the bridge; Through joint simulation, a bridge-track mapping model and a vehicle-track coupling model are established; Acquire the track deformation according to the full bridge alignment and the bridge-track mapping model; Obtaining a vehicle dynamic response according to the full-bridge alignment, the vehicle-rail coupling model, and the track deformation; the vehicle dynamic response includes: a vehicle body vertical acceleration, a left wheel vertical wheel-rail force, a right wheel vertical wheel-rail force, and a right wheel lateral wheel-rail force; Obtaining a resultant secondary suspension vertical force according to the vehicle body vertical acceleration and the physical constraint relationship model; Establishing a first training set, a first validation set, and a first test set using the track deformation, the vehicle dynamic response, and the secondary suspension vertical force; Initializing the PI-LSTM model using the first training set, the first validation set, and the first test set; The vehicle dynamic response prediction value includes: a vehicle body vertical acceleration prediction value, a derailment coefficient prediction value, and a wheel load reduction rate prediction value; The method of obtaining the predicted value of the vehicle dynamic response comprises the following steps: Obtaining a predicted value of the vehicle body vertical acceleration according to the German track irregularity spectrum, the track deformation, and the PI-LSTM model; Obtaining the wheel-rail force prediction value according to the German track irregularity spectrum, the track deformation, the vehicle body vertical acceleration prediction value, and the PI-LSTM model; the wheel-rail force prediction value includes: a left wheel vertical wheel-rail force prediction value, a right wheel vertical wheel-rail force prediction value, and a right wheel lateral wheel-rail force prediction value; The derailment coefficient prediction value and the wheel load reduction rate prediction value are obtained according to the wheel-rail force prediction value.

4. The method for analyzing driving safety of a high-speed railway long-span cable-stayed bridge according to claim 2, characterized in that: The loss function of the PI-LSTM model includes: data loss function and physical loss function; The expression of the data loss function is: （3）； The expression of the physical loss function is: （4）； In formula (3) and formula (4), LOSS data Indicates data loss, LOSS physics Indicates physical loss, N Indicates the total amount of data. Indicates the i The predicted value of vehicle vertical acceleration, Indicates the i The true value of the vehicle body vertical acceleration, Indicates the i The predicted value of the resultant vertical force of the secondary suspension, Indicates the i The true value of the resultant vertical force of the second suspension, T represents transpose; The method further comprises the following steps: Using hyperparameters to adjust the contribution of the physical loss and establish an overall loss function; The expression of the total loss function is: (5); In formula (5), Loss total represents the total loss, represents a hyperparameter.

5. A method for analyzing driving safety of a high-speed railway long-span cable-stayed bridge according to claim 1 or 2, characterized in that: Before obtaining the prediction interval of the vehicle dynamic response prediction value, the following steps are also included: Establishing a second training set, a second validation set, and a second test set using the vehicle dynamic response prediction value; Initialize the cGANs model using the second training set, the second validation set, and the second test set.

6. A method for analyzing driving safety of a high-speed railway long-span cable-stayed bridge according to claim 2, characterized in that: After obtaining the prediction interval of the vehicle dynamic response prediction value, the following steps are also included: Setting the confidence level of the prediction interval; screening out the vehicle dynamic response prediction values ​​whose probability of falling into the prediction interval is less than the confidence value; Get the logarithmic error of the filtered values; Determine whether the logarithmic error falls within a specified numerical interval; if so, take the filtered value as a credible value; otherwise, take the filtered value as an unreliable value.

7. A high-speed railway long-span cable-stayed bridge driving safety analysis system, characterized in that: include: Model building module, used to establish the physical constraint relationship model between the vertical acceleration of the vehicle body and the resultant vertical force of the secondary suspension; a response prediction module for obtaining a predicted value of vehicle dynamic response using the German track irregularity spectrum, track deformation, and a PI-LSTM model; wherein the physical constraint relationship model is embedded in the physical information neural network of the PI-LSTM model; An interval acquisition module is configured to acquire a prediction interval of the vehicle dynamic response prediction value using a cGANs model, a random noise variable, a true error value, and the vehicle dynamic response prediction value; the true error value is an error value between the vehicle dynamic response prediction value and the true value; A maximum value acquisition module, configured to acquire a maximum value between a peak value of the prediction interval and the vehicle dynamic response prediction value; The numerical comparison module is used to compare the maximum value with the specification limit to obtain a comparison result; the comparison result is used for driving safety assessment.

8. A high-speed railway long-span cable-stayed bridge driving safety analysis system according to claim 7, characterized in that: The model building module includes: A vehicle model creation unit, configured to establish a vehicle physical model; the vehicle physical model comprises a vehicle body, a plurality of bogies, and a plurality of wheel sets; a mechanical model creation unit, configured to establish a vehicle body dynamics model based on the D'Alembert principle and the vehicle physical model; A mechanical model processing unit, configured to process the vehicle dynamics model to obtain the physical constraint relationship model; The vehicle body dynamics model is expressed as: (1); among them, M v Indicates the vehicle body mass, represents the vertical acceleration of the vehicle body, g represents the acceleration due to gravity, F ti represents the secondary suspension vertical force, i =1,2,3,4; The expression of the physical constraint relationship model is: (2); among them, F total It represents the resultant vertical force of the secondary suspension.

9. A high-speed railway long-span cable-stayed bridge driving safety analysis system according to claim 7 or 8, characterized in that: Also includes: a data acquisition module, configured to collect sensor data from multiple static deflection measurement points on the bridge to obtain a health monitoring data set for the bridge; a data processing module, configured to process the health monitoring data set using cubic spline interpolation to obtain the full bridge alignment of the bridge; The joint simulation module is used to establish a bridge-track mapping model and a vehicle-track coupling model through joint simulation; A track deformation acquisition module, configured to acquire the track deformation according to the full bridge alignment and the bridge-track mapping model; a dynamic response acquisition module, configured to acquire a vehicle dynamic response based on the full-bridge linear shape, the vehicle-track coupling model, and the track deformation; The vehicle dynamic response includes: vehicle body vertical acceleration, left wheel vertical wheel-rail force, right wheel vertical wheel-rail force and right wheel lateral wheel-rail force; a vertical force acquisition module, configured to acquire a resultant vertical force of the secondary suspension according to the vehicle body vertical acceleration and the physical constraint relationship model; a first data partitioning module, configured to establish a first training set, a first validation set, and a first test set using the track deformation, the vehicle dynamic response, and the secondary suspension vertical force; A first model processing module, configured to initialize the PI-LSTM model using the first training set, the first validation set, and the first test set; The vehicle dynamic response prediction value includes: a vehicle body vertical acceleration prediction value, a derailment coefficient prediction value, and a wheel load reduction rate prediction value; The response prediction module includes: an acceleration prediction unit, configured to obtain a predicted value of the vehicle body vertical acceleration based on a German track irregularity spectrum, the track deformation, and the PI-LSTM model; a wheel-rail force prediction unit, configured to obtain the wheel-rail force prediction value based on the German track irregularity spectrum, the track deformation, the vehicle body vertical acceleration prediction value, and the PI-LSTM model; the wheel-rail force prediction value includes: a left wheel vertical wheel-rail force prediction value, a right wheel vertical wheel-rail force prediction value, and a right wheel lateral wheel-rail force prediction value; The wheel-rail force processing unit is used to obtain the derailment coefficient prediction value and the wheel load reduction rate prediction value according to the wheel-rail force prediction value.

10. The high-speed railway long-span cable-stayed bridge driving safety analysis system according to claim 8, characterized in that: Also includes: The loss function creation module is used to adjust the contribution of hyperparameters to physical loss and establish the overall loss function; The expression of the total loss function is: (5); In formula (5), Loss total represents the total loss, LOSS data Indicates data loss, LOSS physics Indicates physical loss, represents a hyperparameter; （3）； （4）； In formula (3) and formula (4), LOSS data Indicates data loss, LOSS physics Indicates physical loss, N Indicates the total amount of data. Indicates the i The predicted value of vehicle vertical acceleration, Indicates the i The true value of the vehicle body vertical acceleration, Indicates the i The predicted value of the resultant vertical force of the secondary suspension, Indicates the i The true value of the resultant vertical force of the second suspension, T Indicates transpose.

11. A high-speed railway long-span cable-stayed bridge driving safety analysis system according to claim 7 or 8, characterized in that: Also includes: a second data partitioning module, configured to establish a second training set, a second validation set, and a second test set using the vehicle dynamic response prediction value; The second model processing module is used to initialize the cGANs model using the second training set, the second validation set and the second test set.

12. The high-speed railway long-span cable-stayed bridge driving safety analysis system according to claim 8, characterized in that: Also includes: A confidence setting module, used to set the confidence level of the prediction interval; a numerical screening module, configured to screen out values ​​of the vehicle dynamic response prediction values ​​whose probability of falling within the prediction interval is less than the confidence level; A logarithmic error acquisition module is used to obtain the logarithmic error of the filtered value; An error analysis module is used to determine whether the logarithmic error falls within a specified numerical interval; If so, the filtered value is used as the credible value; Otherwise, the filtered values ​​are treated as untrusted values.

Images