Driving safety evaluation method, system, terminal and medium for vehicle-track-bridge system

Abstract

The present invention discloses a vehicle-track-bridge system driving safety evaluation method, system, terminal and medium, wherein the method comprises: constructing a train-track-bridge system coupled stochastic dynamics model; constructing different damage condition combinations of bridges, and simulating the driving process under different damage condition combinations to obtain corresponding mechanical response sequences; using TOPSIS evaluation algorithm to evaluate and score the mechanical response sequences under different damage condition combinations, using the scoring results as model input, and using the damage condition combination as a label to construct a sample data set; using the sample data set to train a prediction model based on machine learning to obtain a bridge damage prediction model; obtaining a dynamic response sequence when the train passes through a track-bridge under actual working conditions, and evaluating and scoring it; inputting the scoring results of the actual working conditions into the bridge damage prediction model to obtain the actual damage working condition combination of the bridge. The present invention can realize the driving safety evaluation of high-speed railways under long-term service.

Description

Technical Field

[0001] The present invention relates to the field of transportation technology, and in particular to a method, system, terminal and medium for evaluating driving safety of a vehicle-track-bridge system. Background Art

[0002] At present, the development of high-speed railways is mostly carried out by constructing and paving with bridges leading the way. As the speed of high-speed railways continues to increase, the safety during driving also requires more protection and verification.

[0003] At present, there are few studies on the safety evaluation methods of the driving process, and the evaluation still relies solely on the specifications and standards. However, the existing specifications and standards only provide regulations under the limit damage value, and cannot monitor and warn the structure itself. The mainstream evaluation method is based on the probability density evolution theory and applied to the coupled random vibration model of train-bridge and train-track-bridge, and improves the calculation accuracy compared with other methods; however, this method focuses on the calculation method and formula derivation under the influence of external excitation. The overall evaluation process is not intelligent, the process calculation time is long, and the computing resources are seriously occupied. There is still a blank area in the safety evaluation of high-speed railways under long-term service. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method, system, terminal and medium for evaluating the driving safety of a vehicle-track-bridge system, which improves the efficiency of the driving safety evaluation and realizes the driving safety evaluation of high-speed railways under long-term service.

[0005] In a first aspect, a method for evaluating driving safety of a vehicle-track-bridge system is provided, comprising the following steps:

[0006] S1: Construct a coupled stochastic dynamic model of the train-track-bridge system based on the actual train-track-bridge system structure;

[0007] S2: Construct different damage condition combinations of the bridge, and simulate the driving process under different damage condition combinations based on the train-track-bridge system coupled stochastic dynamics model to obtain the mechanical response sequence under different damage condition combinations;

[0008] S3: Based on the mechanical response sequence under different damage condition combinations, the TOPSIS evaluation algorithm is used to evaluate and score each damage condition combination, and the scoring results are used as model input, and the corresponding damage condition combination is used as a label to construct a sample data set;

[0009] S4: Use the sample data set to train the prediction model based on machine learning to obtain a bridge damage prediction model;

[0010] S5: Obtain the dynamic response sequence of the train when it passes through the track-bridge under actual working conditions, and use the TOPSIS evaluation algorithm to evaluate and score the actual working conditions;

[0011] S6: Input the scoring results of the actual working conditions into the bridge damage prediction model to obtain the actual damage working condition combination of the bridge.

[0012] Furthermore, the mechanical response sequence is an acceleration sequence, velocity sequence or displacement sequence of a train; or, the mechanical response sequence is an acceleration sequence, velocity sequence or displacement sequence of a track; or, the mechanical response sequence is an acceleration sequence, velocity sequence or displacement sequence of a bridge.

[0013] Furthermore, the mechanical response sequence under different damage condition combinations is evaluated and scored using the TOPSIS evaluation algorithm, specifically including:

[0014] A decision matrix is ​​constructed based on the mechanical response sequences under all damage condition combinations, and each row in the decision matrix corresponds to a mechanical response sequence under a damage condition combination;

[0015] The decision matrix is ​​normalized; if there are negative numbers in the normalized decision matrix, the normalized decision matrix is ​​normalized so that there are no negative numbers in the normalized decision matrix;

[0016] Based on the normalized or standardized decision matrix, an ideal solution based on the maximum value and an anti-ideal solution based on the minimum value are constructed;

[0017] Based on the normalized or standardized decision matrix, as well as the ideal solution and anti-ideal solution, the scores corresponding to each damage condition combination are calculated and normalized to obtain the final scoring results under each damage condition combination.

[0018] Furthermore, step S5 specifically includes:

[0019] The dynamic response sequence of the train passing through the track-bridge under the actual working condition is used together with the mechanical response sequence under different damage working condition combinations obtained in step S2 to construct a decision matrix, and then the actual working condition is evaluated and scored using the TOPSIS evaluation algorithm.

[0020] Furthermore, the prediction model based on machine learning is an XGBoost model.

[0021] Furthermore, it also includes: classifying and warning according to the damage level based on the actual damage working condition combination of the bridge.

[0022] In a second aspect, a vehicle-track-bridge system driving safety evaluation system is provided, comprising:

[0023] The sensor module is used to collect the dynamic response sequence of the train when it passes through the track-bridge under actual working conditions and transmit it to the processing module;

[0024] The processing module is configured to execute the vehicle-track-bridge system driving safety evaluation method as described above.

[0025] Furthermore, the processing module includes:

[0026] A model building unit, used to build a train-track-bridge system coupled stochastic dynamics model according to the actual train-track-bridge system structure;

[0027] The simulation unit is used to construct different damage condition combinations of the bridge, and simulate the driving process under different damage condition combinations based on the train-track-bridge system coupling stochastic dynamics model to obtain the mechanical response sequence under different damage condition combinations;

[0028] An evaluation unit is used to evaluate and score each damage condition combination based on the mechanical response sequence under different damage condition combinations using the TOPSIS evaluation algorithm, and to construct a sample data set using the scoring results as model input and the corresponding damage condition combination as a label;

[0029] A model training unit, used to train a prediction model based on machine learning using a sample data set to obtain a bridge damage prediction model;

[0030] The actual working condition evaluation unit is used to obtain the dynamic response sequence of the train when it passes through the track-bridge under the actual working condition, and evaluate and score the actual working condition using the TOPSIS evaluation algorithm;

[0031] The actual damage prediction unit is used to input the scoring results of the actual working conditions into the bridge damage prediction model to obtain the actual damage working condition combination of the bridge.

[0032] In a third aspect, an electronic terminal is provided, comprising:

[0033] Memory on which computer programs or instructions are stored;

[0034] The processor is used to load and execute the computer program or instruction to implement the vehicle-track-bridge system driving safety evaluation method as described above.

[0035] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program or instruction is stored, and when the stored computer program or instruction is executed by a processor, the vehicle-track-bridge system driving safety evaluation method as described above is implemented.

[0036] The present invention proposes a vehicle-track-bridge system driving safety evaluation method, system, terminal and medium, which have the following beneficial effects: The present invention combines the TOPSIS evaluation algorithm and the prediction model algorithm based on machine learning, and provides a more flexible, efficient and accurate driving safety evaluation method based on the existing standard single limit evaluation method. The TOPSIS evaluation algorithm avoids the problem of relying on a single limit in traditional specifications through multi-dimensional comprehensive evaluation and scoring, and can evaluate the safety of the system more comprehensively and accurately; considering that the evaluation scores of bridges under different damage condition combinations are different, therefore, the present application introduces a prediction model algorithm based on machine learning, and uses the evaluation scores of different damage condition combinations as the input of the prediction model, and uses the corresponding damage condition combination as the model output, so that it can not only achieve evaluation and scoring through the mechanical response sequence under actual working conditions, but also predict the actual damage of the bridge based on the scoring results. The TOPSIS evaluation algorithm is used to reduce the dimension of the mechanical response sequence feedback generated by the structure during the train passing time, and the reduced dimension data is used as the input of the bridge damage prediction model, which improves the processing efficiency and stability of the bridge damage prediction model. Compared with the traditional method, the present invention provides a more intelligent and systematic evaluation scheme, which provides new ideas and technical support for the safety assessment of high-speed railway driving under long-term service. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 is a flow chart of a driving safety evaluation method for a vehicle-track-bridge system provided by an embodiment of the present invention;

[0039] Figure 2 is a gradient descent graph of the combined score of damage conditions provided by an embodiment of the present invention, wherein (a) is a gradient descent graph of the combined score of damage conditions at a vehicle speed of 300 km / h, and (b) is a gradient descent graph of the combined score of damage conditions at a vehicle speed of 350 km / h;

[0040] Figure 3 It is a probability density evolution diagram of damage condition combinations provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0042] like Figure 1 As shown, an embodiment of the present invention provides a method for evaluating driving safety of a vehicle-track-bridge system, comprising the following steps:

[0043] S1: Construct a coupled stochastic dynamic model of the train-track-bridge system based on the actual train-track-bridge system structure.

[0044] There are many existing methods for constructing a coupled stochastic dynamic model of a train-track-bridge system, and the method for constructing a coupled stochastic dynamic model of a train-track-bridge system is not the focus of the present invention. Therefore, for ease of understanding, only one method for constructing a coupled stochastic dynamic model of a train-track-bridge system is used as an example for explanation below.

[0045] Consider the probability distribution characteristics of the system structural parameters under the joint coupling effect of multiple random parameters on the train-track-bridge system. Suppose the system random parameter vector point set of the train-track-bridge system coupling stochastic dynamics model is , and expressed as follows:

[0046]

[0047] In the formula, is the random parameter vector of the vehicle, such as the random stiffness of the vehicle structure spring, the vehicle live load, the moment of inertia of the vehicle body, and the primary and secondary spring damping stiffness of the vehicle; is the random parameter vector of the bridge system, such as mass density, elastic modulus, structural damping characteristics, cross-sectional size error and other random factors. It should be noted that the bridge and track are a whole system. Therefore, the bridge system here includes the track structure and the bridge structure. Therefore, this parameter is a unified representation of the random parameters of the bridge system including the bridge structure and the track structure. It is a random load parameter, mainly including external random load parameters such as track irregularity.

[0048] The train-track-bridge random vibration equations are established by combining the vehicle space random dynamic equations, track space random vibration equations, and bridge space random vibration equations. The train-track-bridge system random vibration equations at time t are established and expressed as follows:

[0049]

[0050] In the formula, , and Respectively represent the acceleration, velocity and displacement of the vehicle; , and They respectively represent the consideration of vehicle structure random parameters The random mass matrix, random damping matrix and random stiffness matrix of the affected vehicle; In order to consider the external random load parameters such as track irregularity The vehicle random load matrix; , and denote the acceleration, velocity and displacement of the track structure respectively; , and They represent the consideration of the random parameters of the bridge system. The random mass matrix, random damping matrix and random stiffness matrix of the affected track structure; In order to consider the external random load parameters such as track irregularity The random load matrix of the track structure; , and They represent the acceleration, velocity and displacement of each degree of freedom of the bridge structure respectively; , and They represent the consideration of the random parameters of the bridge system. The random mass matrix, random damping matrix and random stiffness matrix of the affected bridge structure; In order to consider the external random load parameters such as track irregularity Random load matrix of bridge structure.

[0051] Then the vehicle space random dynamic equation and track space random vibration equation are coupled to each other through the wheel-rail relationship, and the track space random vibration equation and bridge space random vibration equation are coupled to each other through the bridge-rail relationship. Then the above train-track-bridge random vibration equations are coupled to obtain the following equation, which is the train-track-bridge system coupled random dynamic model, which is expressed as follows:

[0052]

[0053] In the formula, The random mass matrix representing the orbital structure, and They represent the random stiffness matrix and random damping matrix of the track structure after considering the coupling of wheel-rail relationship and bridge-rail relationship; represents the random mass matrix of the bridge structure, and They represent the random stiffness matrix and random damping matrix of the bridge structure after considering the bridge-track coupling relationship; and They represent the interaction stiffness matrix and damping matrix of the vehicle to the track structure, and They represent the interaction stiffness matrix and damping matrix of the track structure to the vehicle respectively; and They represent the interaction stiffness matrix and damping matrix of the bridge structure on the track structure, and denote the interaction stiffness matrix and damping matrix of the track structure to the bridge structure respectively; , and They respectively represent the vehicle random load matrix, track structure random load matrix and bridge structure random load matrix after considering the coupling of wheel-rail relationship and bridge-rail relationship.

[0054] S2: Construct different damage condition combinations of the bridge, and simulate the driving process under different damage condition combinations based on the train-track-bridge system coupled stochastic dynamics model to obtain the mechanical response sequence under different damage condition combinations.

[0055] The actual damage conditions of bridges are diverse. For example, for a three-span bridge, the damage conditions of the middle span and the two side spans may be different. Therefore, it is necessary to first construct possible damage condition combinations, and then construct different damage constitutive models according to different damage condition combinations, and combine them with the train-track-bridge system coupled stochastic dynamics model to simulate the driving process under different damage condition combinations, and then obtain the mechanical response sequence under different damage condition combinations. In specific implementation, the mechanical response sequence can select the acceleration sequence, velocity sequence or displacement sequence of the train, the acceleration sequence, velocity sequence or displacement sequence of the track, or the acceleration sequence, velocity sequence or displacement sequence of the bridge.

[0056] In real engineering, staged random damage is related to many parameters and belongs to a high-dimensional nonlinear dynamic system. However, the focus of the present invention is not to build a damage constitutive model, but to propose a damage-based driving safety evaluation method. Therefore, in order to quickly experiment and facilitate explanation, the damage matrix is ​​transformed into The random mass matrix, random damping matrix and random stiffness matrix of the bridge structure are introduced to realize the simulation of different damage conditions. At the same time, in this embodiment, the damage random mass matrix is ​​considered to be is almost unaffected by the damage matrix. Therefore, the damage random mass matrix of the bridge structure is , damage random damping matrix and the damage random stiffness matrix It is expressed as follows:

[0057]

[0058]

[0059]

[0060] In the formula, and They represent the damping damage matrix and stiffness damage matrix of the bridge structure respectively.

[0061] S3: Based on the mechanical response sequence under different damage condition combinations, the TOPSIS evaluation algorithm is used to evaluate and score each damage condition combination. The scoring results are used as model input, and the corresponding damage condition combination is used as a label to construct a sample data set.

[0062] The TOPSIS (Technique for Order Preference by Similarity to an Ideal Solution) evaluation algorithm is a ranking technique that is close to the ideal solution, also known as the superior and inferior solution distance method. It is based on comparing the similarity between the evaluation scheme and the ideal solution and the anti-ideal solution, and achieves evaluation and scoring by calculating the distance between each evaluation scheme and the ideal solution and the anti-ideal solution.

[0063] The advantage of the TOPSIS evaluation algorithm is that it can more conveniently reflect the details of the mechanical response. In theory, the closer the train-track-bridge response is to 0, the smoother it is, which also means that the vehicle runs smoothly and safely. In this embodiment, a train-track-bridge system coupled stochastic dynamics model is used to simulate the driving process under a large number of damage conditions, and the optimal mechanical response sequence of each condition combination is selected as a benchmark, laying the foundation for later machine learning training. The TOPSIS evaluation algorithm makes full use of the information of the original data, and the results can accurately reflect the gap between the evaluation schemes, and can obtain good comparable evaluation ranking results.

[0064] In this embodiment, the specific process of evaluating and scoring each damage condition combination using the TOPSIS evaluation algorithm based on the mechanical response sequence under different damage condition combinations includes:

[0065] S31: First, the mechanical response sequence is forward processed. The forward processing process is expressed by the following formula:

[0066]

[0067] In the formula, represents the jth mechanical response value in the mechanical response sequence after the forward processing, represents the jth mechanical response value in the mechanical response sequence, It represents the optimal value of the mechanical response in the mechanical response sequence (it is generally believed that the most ideal condition is that the mechanical response value when a train passes is the optimal value. Taking the bridge acceleration as an example, its optimal value is 0). Indicates taking all The sequence of The maximum value in .

[0068] It should be noted that whether to perform forward processing can be selected according to actual needs, and it is not a necessary processing process.

[0069] S32: A decision matrix is ​​constructed based on the mechanical response sequences under all damage condition combinations, where each row in the decision matrix corresponds to a mechanical response sequence under one damage condition combination.

[0070] Assuming that the number of damage condition combinations is n and the length of the mechanical response sequence is m, the decision matrix X is expressed as follows:

[0071]

[0072] In the formula, It represents the mth mechanical response value in the mechanical response sequence under the nth combination of damage conditions.

[0073] The above decision matrix can be simplified as follows:

[0074]

[0075] In the formula, It represents the jth mechanical response value in the mechanical response sequence under the i-th combination of damage conditions.

[0076] S33: normalizing the decision matrix; if there are negative numbers in the normalized decision matrix, normalizing the normalized decision matrix so that there are no negative numbers in the normalized decision matrix.

[0077] The normalized decision matrix Z is expressed as follows:

[0078]

[0079] In the formula, .

[0080] In the process of processing the decision matrix, it is necessary to ensure that all the data in the final decision matrix are greater than or equal to 0. Therefore, if there are negative numbers in the decision matrix after normalization, it is re-normalized by the following formula to obtain the standardized decision matrix:

[0081]

[0082]

[0083] In the formula, represents the standardized decision matrix.

[0084] S34: Based on the normalized or standardized decision matrix, construct an ideal solution based on the maximum value and an anti-ideal solution based on the minimum value. Taking the decision matrix Z after planning as an example, the ideal solution and the anti-ideal solution are expressed as follows:

[0085]

[0086]

[0087] In the formula, and represent ideal solution and anti-ideal solution respectively.

[0088] S35: Based on the normalized or standardized decision matrix, as well as the ideal solution and the anti-ideal solution, the scores corresponding to each damage condition combination are calculated and normalized to obtain the final scoring results under each damage condition combination.

[0089] Still taking the decision matrix Z after planning as an example, we first calculate the distance between each damage condition combination and the ideal solution and anti-ideal solution. The calculation formula is as follows:

[0090]

[0091]

[0092] In the formula, and Respectively represent the distance between the i-th group of damage condition combination and the ideal solution and anti-ideal solution; and denote the jth value in the ideal solution and anti-ideal solution respectively;

[0093] Then the score of the i-th group of damage condition combinations is calculated by the following formula:

[0094]

[0095] In the formula, represents the score of the i-th group of damage condition combinations, , and at this time The bigger, The smaller it is, the closer it is to the maximum value;

[0096] Finally, the scores of each damage condition combination are normalized using the following formula:

[0097]

[0098] in, It represents the final scoring result of the i-th group of damage condition combinations. At this time, the sum of the scoring results of each group of damage condition combinations is equal to 1.

[0099] S4: Use the sample data set to train the prediction model based on machine learning to obtain a bridge damage prediction model.

[0100] In specific implementation, the prediction model based on machine learning can select BP neural network, random forest or XGBoost (eXtreme Gradient Boosting Tree) model. Considering that the prediction efficiency and accuracy of the XGBoost model are higher, in this embodiment, the prediction model based on machine learning is taken as an example of the XGBoost model, that is, the scoring result of the damage condition combination is used as the input of the XGBoost model, and the corresponding damage condition combination is used as the output of the XGBoost model. The core of XGBoost is the use of gradient boosting technology. This embodiment builds an XGBoost model, effectively combines the score obtained by TOPSIS dimensionality reduction evaluation with the damage condition combination to achieve a training cycle, and uses a large amount of data for training. The L1 regularization term and the L2 regularization term in the model can effectively prevent the model from overfitting. At the same time, adjusting the learning rate and the number of learners of machine learning can also ensure the good prediction accuracy of the algorithm and avoid misjudgment of the structural state in actual engineering. The training process should also avoid overfitting results. It is necessary to ensure a certain degree of robustness of the model. Therefore, when building the model, the sample data set is divided into a training set and a test set for verification through data segmentation, and the accuracy of model training is repeatedly determined through data shuffling and cross-validation.

[0101] The core idea of ​​the mathematical logic of the XGBoost model is to continuously add new trees to optimize the model according to the prediction error (residual) of the current model in each iteration, so that the new tree can minimize the prediction error, and the maximum depth and minimum node weight pruning strategy can be used in the process of building the tree through pruning methods to effectively control the complexity of the tree. Therefore, the XGBoost model is an efficient, flexible and scalable implementation of the gradient boosting tree, and its integrated model is shown in the following formula:

[0102]

[0103] In the formula, is the predicted value corresponding to the qth sample; is the qth sample input; is an independent function in space, i.e., the kth decision tree; is the number of decision trees; is the function space composed of all decision trees.

[0104] The objective function L is shown as follows:

[0105]

[0106]

[0107] In the formula, Q represents the number of samples; Represents the loss function, which is the predicted value and actual value The error between is a regularization term, which represents the model complexity of the kth decision tree. The regularization term is to penalize more complex models and reduce the overfitting effect by reducing the depth of the decision tree and the weight value of a single leaf node. is the number of leaf nodes of the current decision tree, is the weight of the jth leaf node; and are pre-given parameters.

[0108] Using incremental training to minimize the objective function, Add during round iteration , the updated objective function is as follows:

[0109]

[0110] In the formula, For the The predicted value of the round iteration; The new function added for each iteration. The formula is derived through the second-order Taylor expansion and is approximately The optimal solution is obtained in the form of a quadratic function of And the corresponding optimal solution objective function It is expressed as follows:

[0111]

[0112]

[0113] In the formula, ; ; Represents the sample set under the jth leaf node.

[0114] During the training process, the splitting criterion of the XGBoost model adopts a greedy algorithm to improve the parallelization of the algorithm. At the same time, XGBoost introduces parallel computing in the process of tree node splitting to improve the training speed. The feature node that maximizes the gain of the objective function is selected as the split point. The importance of the feature is extracted using the gain, that is, the more times a feature is selected as a split node, the higher the importance of the feature. The gain is calculated starting from the root node of the tree using the following formula:

[0115]

[0116] In the formula, are the first-order derivative and second-order derivative of the left subtree of the current node when splitting, They are the first-order derivative and second-order derivative of the right subtree of the current node when splitting. In actual partitioning, XGBoost will partition based on the node with the largest gain. is the score of the new left cotyledon (i.e., the objective function of the left cotyledon after the feature splitting of the node); is the score of the new right cotyledon (i.e., the objective function of the right cotyledon after feature splitting of the node); is the score of the original leaf node (i.e. the objective function of the node before feature splitting). Its meaning is: to judge the information gain after splitting the node, and Whether the sum is greater than the unsplit case, and considering whether the model is too complicated, if the added score is less than the regularization term, the node will no longer split.

[0117] S5: Obtain the dynamic response sequence of the train when passing through the track-bridge under actual working conditions, and use the TOPSIS evaluation algorithm to evaluate and score the actual working conditions.

[0118] In specific implementation, the dynamic response sequence of the train when passing through the track-bridge under actual working conditions is used together with the mechanical response sequence under different damage condition combinations obtained in step S2 to construct a decision matrix, that is, the dynamic response sequence under the actual working conditions is added to the decision matrix of step S2 as new row data, and then the actual working conditions are evaluated and scored using the TOPSIS evaluation algorithm.

[0119] S6: Input the scoring results of the actual working conditions into the bridge damage prediction model to obtain the actual damage working condition combination of the bridge.

[0120] In some embodiments, the vehicle-track-bridge system driving safety evaluation method further includes: classifying and warning according to the damage level according to the actual damage working condition combination of the bridge. In the actual damage working condition combination of the bridge, each damage corresponds to its damage level, and classifying and warning according to the damage level is carried out so as to facilitate timely processing according to the warning level. Classifying different driving response warning levels during driving under long-term service is particularly important for the operation and maintenance of high-speed railways under long-term service performance.

[0121] The vehicle-track-bridge system driving safety evaluation method provided in the above embodiment combines the TOPSIS evaluation algorithm and the prediction model algorithm based on machine learning, and provides a more flexible, efficient and accurate driving safety evaluation method on the basis of the existing single limit evaluation method of the specification. The TOPSIS evaluation algorithm avoids the problem of relying on a single limit in the traditional specification through multi-dimensional comprehensive evaluation and scoring, and can evaluate the safety of the system more comprehensively and accurately; considering that the evaluation scores of bridges under different damage working condition combinations are different, therefore, the present application introduces a prediction model algorithm based on machine learning, and uses the evaluation scores of different damage working condition combinations as the input of the prediction model, and the corresponding damage working condition combination as the model output, so that it can not only achieve the evaluation and scoring through the mechanical response sequence under the actual working condition, but also predict the actual damage of the bridge based on the scoring result. The TOPSIS evaluation algorithm is used to perform dimensionality reduction analysis on the feedback of the mechanical response sequence generated by the structure during the train passing time, and the dimensionality reduction data is used as the input of the bridge damage prediction model, which improves the processing efficiency and stability of the bridge damage prediction model. Compared with the traditional method, the present invention provides a more intelligent and systematic evaluation scheme, which provides new ideas and technical support for the safety evaluation of high-speed railway driving under long-term service. It should be particularly noted that the vehicle-track-bridge system driving safety evaluation method provided by the present invention is applied to high-speed train application scenarios, and is generally best applied to scenarios where high-speed trains pass through the train-track-bridge system at a speed of more than 250 km / h. This is because when the vehicle speed is low, the differences between the responses are also reduced, and there will be slight differences and changes in the score changes between the damage groups, resulting in relatively poor model prediction accuracy.

[0122] The embodiment of the present invention further provides a vehicle-track-bridge system driving safety evaluation system, comprising:

[0123] The sensor module is used to collect the dynamic response sequence of the train when it passes through the track-bridge under actual working conditions and transmit it to the processing module;

[0124] The processing module is configured to execute the vehicle-track-bridge system driving safety evaluation method as described in the above-mentioned embodiment.

[0125] Specifically, the processing module includes:

[0126] A model building unit, used to build a train-track-bridge system coupled stochastic dynamics model according to the actual train-track-bridge system structure;

[0127] The simulation unit is used to construct different damage condition combinations of the bridge, and simulate the driving process under different damage condition combinations based on the train-track-bridge system coupling stochastic dynamics model to obtain the mechanical response sequence under different damage condition combinations;

[0128] An evaluation unit is used to evaluate and score each damage condition combination based on the mechanical response sequence under different damage condition combinations using the TOPSIS evaluation algorithm, and to construct a sample data set using the scoring results as model input and the corresponding damage condition combination as a label;

[0129] A model training unit, used to train a prediction model based on machine learning using a sample data set to obtain a bridge damage prediction model;

[0130] The actual working condition evaluation unit is used to obtain the dynamic response sequence of the train when it passes through the track-bridge under the actual working condition, and evaluate and score the actual working condition using the TOPSIS evaluation algorithm;

[0131] The actual damage prediction unit is used to input the scoring results of the actual working conditions into the bridge damage prediction model to obtain the actual damage working condition combination of the bridge.

[0132] It should be understood that the functional units in the above embodiments may be concentrated in one processing unit, or each unit may exist physically separately, or two or more units may be integrated in one unit, and may be implemented in the form of hardware or software.

[0133] An embodiment of the present invention further provides an electronic terminal, including:

[0134] Memory on which computer programs or instructions are stored;

[0135] The processor is used to load and execute the computer program or instruction to implement the vehicle-track-bridge system driving safety evaluation method as described in the above embodiment.

[0136] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program or instruction is stored. When the stored computer program or instruction is executed by a processor, the vehicle-track-bridge system driving safety evaluation method as described in the above embodiment is implemented.

[0137] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0138] The technical solution of the present invention is further illustrated below in combination with an experimental case.

[0139] In this experimental case, the damage simulation in the train-track-bridge system coupled stochastic dynamics model was designed using Matlab program. The bridge parameters set in the program were based on a 32m standard simply supported beam bridge on a high-speed railway in China (the cross-sectional area of ​​the bridge is A=8.89 m 2 , polar moment of inertia Iy = 10.95 m 4 ), this case uses three spans to establish a coupled stochastic dynamic model of the train-track-bridge system, and handles the damage of each span of the three-span simply supported beam separately. The simulated vehicle uses CRH380A high-speed rail vehicle simulation, which uses 4 motors and 4 trailers, with a motor of 4800kg and a trailer of 4400kg. Its first-order and second-order natural frequencies are 5.12Hz and 17.04 Hz respectively; the bridge irregularity modulation coefficient c=0.65, the modulation transition section length L0=20 m, and the train starts running from −150 m away from the bridge head to ensure that the train vibration is stable before passing the bridge. For the simulation of damage, this case defines different damages of the bridge by changing the concrete stiffness state, and simulates different damage conditions by gradually reducing the failure percentage.

[0140] In the damage definition part, the damage categories are divided into four attributes. By referring to the relevant design indicators such as the "Basic Code for Design of Railway Bridges" (TB / T10002.1-99) and the "Design Code for High-speed Railways" (TB10621-2014), the first type of damage EI (Ⅰ) is defined as its safety assumption is equivalent to the stiffness of C60 concrete in a healthy state without shrinkage creep; the second type of damage EI (Ⅱ) is defined as its stiffness is equivalent to EI (Ⅰ) in the 95% effective stiffness state; the third type of damage EI (Ⅲ) is defined as its stiffness is equivalent to EI (Ⅰ) in the 90% effective stiffness state; the fourth type of damage EI (Ⅳ) is defined as its stiffness is equivalent to EI (Ⅰ) in the 85% effective stiffness state, and its overall structural stiffness decreases gradually from 100% to 85%. When using four types of damage to simulate the damage condition combination, since the mid-span damage has the most obvious effect on the overall response when the train-track-bridge is coupled, the overall property decline and gradient decline of the beam are considered. A total of 10 damage condition combinations are created on the 3-span simply supported bridge through permutations and combinations, as shown in Table 1 below. The damage condition combination design order is as follows. This case performs the gradient simulation of the program according to the subjective order in Table 1.

[0141]

[0142] For the simulation of speed, an important dependent variable, this case chooses to simulate speeds of 300km / h and 350km / h. For each damage condition combination, two speed conditions are simulated in the train-track-bridge program. This design is to meet the diversity of samples and the probability of achieving results, avoid the contingency in single samples and small samples, and provide a good foundation for the training set and test set of later machine learning.

[0143] In this case, the acceleration of the bridge was used as the mechanical response for the experiment. In this paper, two groups of simulations were conducted on the same simply supported bridge with different damage condition combinations at speeds of 300km / h and 350km / h. 100 groups of high-speed CRH380A models were simulated at each speed, for a total of 200 train simulations for each damage combination condition, and a total of 1000 train simulations for each speed condition. A total of 4300 key quantities of acceleration were extracted for each train, and the key quantities of acceleration response were used as evaluation indicators to construct a decision matrix. The positive ideal solution and negative ideal solution were then determined through the TOPSIS evaluation algorithm to determine the score of each damage condition combination, and the train plans were ranked according to the score, and the plot was drawn based on the score, damage condition, and speed condition. Figure 2 It should be noted that the decision matrix is ​​constructed by classification based on speed, that is, the acceleration response of each damage condition combination at each speed is used to construct the decision matrix, and then a bridge damage prediction model based on XGBoost will be trained for each speed in the future. The reason is that in the actual high-speed train operation process, the speed of each train passing through the bridge section has been pre-planned and is a fixed value.

[0144] like Figure 2 As shown in (a) and (b), the horizontal axis represents the damage condition combination, and each 100 value represents a damage condition combination. The damage condition combinations are progressive from left to right, and the vertical axis is the vehicle dynamic response score. It can be seen from the figure that at speeds of 300km / h and 350km / h, the driving interval score completely decreases according to the set gradient, and the boundary between the best condition group and the worst condition group is clear.

[0145] Random phenomena are ubiquitous in train-track-bridge systems. Yu Zhiwu's team has done a lot of work in the vibration and probability density analysis of train-track-bridge systems, assuming that the vehicle always runs at a uniform speed V; the vehicle body, wheelset and rails are all regarded as rigid bodies; the wheelset does not leave the rails, and the wheels do not slide when running on the rails, and there is no track climbing, rail jumping or derailment. According to the assumptions, considering the randomness of vertical track unevenness excitation, the wheel-rail geometric relationship is expressed as assuming that the vehicle-bridge system has only a random excitation source of track height unevenness, and no new random sources are generated in the middle, then this system satisfies the probability conservation condition. According to this probability density evolution theory, combined with the coupled random dynamics model of the train-track-bridge system, the probability density evolution changes of the above ten working conditions in this case are fitted, such as Figure 3 As shown in the figure, the X-axis represents the mechanical response, the Y-axis represents the damage condition combination, from the near screen segment to the far screen segment, it is the damage condition combination 1-10, and the Z-axis represents the probability density value. Figure 3 It can be directly observed that each damage condition combination gradually decreases from the response domain (1-10), and its probability density value gradually increases (1-10), which also shows that the design gradient of this damage condition combination is reasonable and provides a good data foundation and scientific basis for subsequent model training.

[0146] Taking the data at a speed of 350km / h as an example, the TOPSIS evaluation score is used as the input, and the damage condition combination of the bridge is used as the output to construct two sets of data sets. In this case, the mid-span prediction is listed as data set A; the damage on both sides of the side span is constructed consistently, and the prediction effect and principle are exactly the same, so it is set as data set B. Data sets A and B are divided according to the ratio of training set (70%) and test set (30%), and the data set sequence is split during training, and the order is reshuffled to achieve the effect of "shuffling" the data. The bridge damage prediction model based on XGBoost is established through the training set data, and then tested and verified in the test set.

[0147] In the training of the bridge damage prediction model based on XGBoost, the overfitting problem on the training set caused by unreasonable data set division is avoided. The contribution of the penalty factor is increased by setting the L1 and L2 regularization items. In this case, L1 is taken as 0.2, L2 is taken as 0.2, and the sample sampling rate, tree sampling rate, and node sampling rate are set to 0.3, 0.4, and 0.3. At the same time, the model is cross-validated by k-folds, and the data set is divided into k mutually exclusive subsets of similar size. Each time, the union of k-1 subsets is used as the training set, and the remaining subset is used as the test set. K training and testing are performed, and finally the mean of the k evaluation results is returned.

[0148] In the XGBoost model, the accuracy (CA), recall, precision and F1 score are all proportional values. The larger the better. After adjusting the parameters, as shown in Table 2, the prediction accuracy of the cross-validation model can reach 85% in the test set and 84.1% in the test set. The F1 values ​​are 83.2% and 81.8% respectively. The cross-validation set accuracy reaches 82.7%, and the F1 value is 82%. The key values ​​are all above 80%, and a good prediction effect can be achieved in the cross-validation prediction. As shown in Table 3, the prediction accuracy of the cross-center model can reach 87.5% in the test set and 86.7% in the test set, with F1 values ​​of 86.2% and 85.1% respectively. The cross-validation set accuracy reaches 84.7%, and the F1 value is 84.4%. Most of the key values ​​are within 85%, and only the value in the cross-validation set is 0.03 away from 85%. The reason why the prediction effect of the side span is better than that of the cross-center prediction model is that the left and right side span models are symmetrical, and the amount of data and accuracy are more than the cross-center training model. The prediction accuracy of all models is also slightly higher than that of the cross-center model. However, both cases achieve a prediction effect of 80%+, and both models have good prediction effects, achieving the expected effect.

[0149]

[0150]

[0151] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0152] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0153] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0154] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0155] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for evaluating driving safety of a vehicle-track-bridge system, characterized in that: The steps include: S1: Construct a coupled stochastic dynamic model of the train-track-bridge system based on the actual train-track-bridge system structure; S2: Construct different damage condition combinations of the bridge, and simulate the driving process under different damage condition combinations based on the train-track-bridge system coupled stochastic dynamics model to obtain the mechanical response sequence under different damage condition combinations; S3: Based on the mechanical response sequence under different damage condition combinations, the TOPSIS evaluation algorithm is used to evaluate and score each damage condition combination, and the scoring results are used as model input, and the corresponding damage condition combination is used as a label to construct a sample data set; S4: Use the sample data set to train the prediction model based on machine learning to obtain a bridge damage prediction model; S5: Obtain the dynamic response sequence of the train when it passes through the track-bridge under actual working conditions, and use the TOPSIS evaluation algorithm to evaluate and score the actual working conditions; S6: Input the scoring results of the actual working conditions into the bridge damage prediction model to obtain the actual damage working condition combination of the bridge.

2. The vehicle-track-bridge system driving safety evaluation method according to claim 1, characterized in that: The mechanical response sequence is an acceleration sequence, a velocity sequence or a displacement sequence of a train; or, the mechanical response sequence is an acceleration sequence, a velocity sequence or a displacement sequence of a track; or, the mechanical response sequence is an acceleration sequence, a velocity sequence or a displacement sequence of a bridge.

3. The vehicle-track-bridge system driving safety evaluation method according to claim 1, characterized in that: The mechanical response sequence under different damage condition combinations is evaluated and scored using the TOPSIS evaluation algorithm, specifically including: A decision matrix is ​​constructed based on the mechanical response sequences under all damage condition combinations, and each row in the decision matrix corresponds to a mechanical response sequence under a damage condition combination; The decision matrix is ​​normalized; if there are negative numbers in the normalized decision matrix, the normalized decision matrix is ​​normalized so that there are no negative numbers in the normalized decision matrix; Based on the normalized or standardized decision matrix, an ideal solution based on the maximum value and an anti-ideal solution based on the minimum value are constructed; Based on the normalized or standardized decision matrix, as well as the ideal solution and anti-ideal solution, the scores corresponding to each damage condition combination are calculated and normalized to obtain the final scoring results under each damage condition combination.

4. The vehicle-track-bridge system driving safety evaluation method according to claim 3, characterized in that: Step S5 specifically includes: The dynamic response sequence of the train passing through the track-bridge under the actual working condition is used together with the mechanical response sequence under different damage working condition combinations obtained in step S2 to construct a decision matrix, and then the actual working condition is evaluated and scored using the TOPSIS evaluation algorithm.

5. The vehicle-track-bridge system driving safety evaluation method according to claim 1, characterized in that: The prediction model based on machine learning is an XGBoost model.

6. The vehicle-track-bridge system driving safety evaluation method according to claim 1, characterized in that: Also includes: Classified warnings are carried out according to the damage level based on the actual damage condition combination of the bridge.

7. A vehicle-track-bridge system driving safety evaluation system, characterized in that: include: The sensor module is used to collect the dynamic response sequence of the train when it passes through the track-bridge under actual working conditions and transmit it to the processing module; The processing module is configured to execute the driving safety evaluation method of the vehicle-track-bridge system according to any one of claims 1 to 6.

8. The vehicle-track-bridge system driving safety evaluation system according to claim 7, characterized in that: The processing module comprises: A model building unit, used to build a train-track-bridge system coupled stochastic dynamics model according to the actual train-track-bridge system structure; The simulation unit is used to construct different damage condition combinations of the bridge, and simulate the driving process under different damage condition combinations based on the train-track-bridge system coupling stochastic dynamics model to obtain the mechanical response sequence under different damage condition combinations; An evaluation unit is used to evaluate and score each damage condition combination based on the mechanical response sequence under different damage condition combinations using the TOPSIS evaluation algorithm, and to construct a sample data set using the scoring results as model input and the corresponding damage condition combination as a label; A model training unit, used to train a prediction model based on machine learning using a sample data set to obtain a bridge damage prediction model; The actual working condition evaluation unit is used to obtain the dynamic response sequence of the train when it passes through the track-bridge under the actual working condition, and evaluate and score the actual working condition using the TOPSIS evaluation algorithm; The actual damage prediction unit is used to input the scoring results of the actual working conditions into the bridge damage prediction model to obtain the actual damage working condition combination of the bridge.

9. An electronic terminal, characterized in that: include: Memory on which computer programs or instructions are stored; A processor is used to load and execute the computer program or instruction to implement the vehicle-track-bridge system driving safety evaluation method as described in any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: The computer program or instruction stored therein implements the driving safety evaluation method of the vehicle-track-bridge system as claimed in any one of claims 1 to 6 when executed by the processor.

Images