Method and device for predicting vehicle dynamic response

Abstract

The application discloses a kind of prediction method and device of vehicle dynamic response, the method includes: obtaining target track irregularity data;The target track irregularity data is input to vehicle response prediction model, and vehicle dynamic response prediction data is obtained;The vehicle response prediction model is the deep learning model of the convolutional neural network structure of introducing coordinate attention mechanism and multi-scale attention mechanism structure, the vehicle response prediction model is obtained by training the deep learning model to history data;The coordinate attention mechanism is used to determine the channel relationship and spatial mileage point position of track irregularity data, and the coordinate attention weight of track irregularity data is generated;The multi-scale attention mechanism is used to capture long short-term trend of track sequence;The vehicle response prediction model includes: CA-CNN structure and MUSE structure.The application is used to improve the prediction accuracy and prediction speed of vehicle dynamic response data.

Description

Technical Field

[0001] This invention relates to the field of railway track technology, and in particular to a method and apparatus for predicting vehicle dynamic response. Background Technology

[0002] Track quality assessment, as part of track maintenance, is a crucial technical step in ensuring the safe operation of high-speed railway trains and the comfort of passengers. Current track quality assessment methods are based on the amplitude of track geometric irregularities. However, using only a single track irregularity index to evaluate track quality without considering the vehicle's dynamic response is insufficient. For example, some track sections may exhibit problems where the amplitude of each track irregularity does not exceed the limit, yet the vehicle experiences a significant vibration response. Conversely, some track sections may have certain track irregularities exceeding the limit without causing a deterioration in vehicle response. These issues indicate that the vehicle's vibration response is the result of nonlinear coupling of multiple track irregularities.

[0003] To improve track quality assessment standards and guide track maintenance, it is necessary to study the relationship between track geometric irregularities and vehicle response. The key to this work is to find a model that accurately predicts vehicle response under track irregularity conditions, and then evaluate track geometry by combining actual track irregularity indicators with the predicted vehicle response.

[0004] The following methods exist for predicting vehicle response caused by irregular tracks.

[0005] Firstly, the focus can be on establishing mechanistic models to simulate the nonlinear dynamic behavior of vehicles and predict their responses. For example, commercial software such as SIMPACK can be used to build three-dimensional vehicle-track dynamic models to study the relationship between track irregularities and the dynamic performance of rail vehicles. However, the performance of these models depends on the reliability of their parameters, which are easily affected by various real-world variables. Furthermore, because the actual parameters of the vehicle-track system are difficult to obtain and change dynamically over time, applying theoretical models to track maintenance practice is challenging. In addition, numerical iterative methods for solving mechanistic models are extremely time-consuming.

[0006] Secondly, the vehicle-track system can be described using a linear transfer function, and parameters can be estimated based on system identification theory. However, system identification theory can only be applied to linear systems and under constant speed conditions, which limits its ability to predict vehicle response. Summary of the Invention

[0007] This invention provides a method for predicting vehicle dynamic response, thereby improving the prediction accuracy and speed of vehicle dynamic response data. The method includes:

[0008] Acquire data on the target orbital irregularities;

[0009] The target track irregularity data is input into the vehicle response prediction model to obtain the vehicle dynamic response prediction data corresponding to the target track irregularity data. The vehicle response prediction model is a deep learning model with a convolutional neural network structure that incorporates a coordinate attention mechanism and a multi-scale attention mechanism structure. The vehicle response prediction model is trained on the deep learning model using historical data. The historical data includes historical track irregularity data and corresponding historical vehicle dynamic response prediction data.

[0010] The coordinate attention mechanism is used to determine the channel relationships and spatial mileage point positions of the track irregularity data, and to generate the coordinate attention weights of the track irregularity data; the multi-scale attention mechanism is used to capture the long-term and short-term trends of the track sequence.

[0011] The vehicle response prediction model includes a CA-CNN structure and a MUSE structure. The CA-CNN structure is used to extract a first track irregularity feature from the input track irregularity data through convolution. A second track irregularity feature is obtained based on the coordinate attention weights of the track irregularity data and the first track irregularity feature. The MUSE structure is used to perform a depthwise convolution on the second track irregularity feature by combining the long-term and short-term trends of the track sequence to obtain the vehicle response prediction data corresponding to the input track irregularity data.

[0012] This invention also provides a vehicle dynamic response prediction device to improve the prediction accuracy and speed of vehicle dynamic response data. The device includes:

[0013] The track irregularity data acquisition module is used to acquire target track irregularity data;

[0014] The vehicle dynamic response data prediction module is used to input the target track irregularity data into the vehicle response prediction model to obtain the vehicle dynamic response prediction data corresponding to the target track irregularity data. The vehicle response prediction model is a deep learning model with a convolutional neural network structure incorporating a coordinate attention mechanism and a multi-scale attention mechanism structure. The vehicle response prediction model is obtained by training the deep learning model with historical data. The historical data includes historical track irregularity data and corresponding historical vehicle dynamic response prediction data.

[0015] The coordinate attention mechanism is used to determine the channel relationships and spatial mileage point positions of the track irregularity data, and to generate the coordinate attention weights of the track irregularity data; the multi-scale attention mechanism is used to capture the long-term and short-term trends of the track sequence.

[0016] The vehicle response prediction model includes a CA-CNN structure and a MUSE structure. The CA-CNN structure is used to extract a first track irregularity feature from the input track irregularity data through convolution. A second track irregularity feature is obtained based on the coordinate attention weights of the track irregularity data and the first track irregularity feature. The MUSE structure is used to perform a depthwise convolution on the second track irregularity feature by combining the long-term and short-term trends of the track sequence to obtain the vehicle response prediction data corresponding to the input track irregularity data.

[0017] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described method for predicting vehicle dynamic response.

[0018] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting vehicle dynamic response.

[0019] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described method for predicting vehicle dynamic response.

[0020] In this embodiment of the invention, target track irregularity data is acquired; the target track irregularity data is input into a vehicle response prediction model to obtain vehicle dynamic response prediction data corresponding to the target track irregularity data; the vehicle response prediction model is a deep learning model with a convolutional neural network structure incorporating a coordinate attention mechanism and a multi-scale attention mechanism structure, and the vehicle response prediction model is obtained by training the deep learning model with historical data; the historical data includes historical track irregularity data and corresponding historical vehicle dynamic response prediction data; wherein, the coordinate attention mechanism is used to determine the channel relationship and spatial mileage point position of the track irregularity data, and generate coordinate attention weights for the track irregularity data; the multi-scale attention mechanism is used to capture the long-term and short-term trends of the track sequence; the vehicle response prediction model includes a CA-CNN structure and a MUSE structure; the CA-CNN structure is used to obtain the target track irregularity data from the input... In the track irregularity data, a first track irregularity feature is extracted by convolution; based on the coordinate attention weights of the track irregularity data and the first track irregularity feature, a second track irregularity feature is obtained; the MUSE structure is used to combine the long-term and short-term trends of the track sequence and perform deep convolution on the second track irregularity feature to obtain the vehicle response prediction data corresponding to the input track irregularity data. By introducing coordinate attention mechanism and multi-scale attention mechanism into the vehicle response prediction model, the prediction of vehicle dynamic response based on track irregularity data can be accurately realized. This not only improves the prediction accuracy of the vehicle response prediction model for vehicle dynamic response data, but also improves the prediction speed of vehicle dynamic response data. It solves the problem that the prediction process is time-consuming and labor-intensive due to the susceptibility of the constructed three-dimensional vehicle-track dynamics model to the influence of the real environment under the existing technology, and also solves the limitation of vehicle response prediction due to the linear transfer function under the existing technology. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0022] Figure 1 This is a flowchart illustrating a method for predicting vehicle dynamic response in an embodiment of the present invention.

[0023] Figure 2 This is a specific example diagram of a vehicle response prediction model in an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of the structure of a vehicle dynamic response prediction device according to an embodiment of the present invention;

[0025] Figure 4 This is a specific example diagram of a vehicle dynamic response prediction device according to an embodiment of the present invention.

[0026] Figure 5 This is a schematic diagram of a computer device used for predicting vehicle dynamic response in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0028] In this document, the term "and / or" merely describes a relationship, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0029] In the description of this specification, the terms "comprising," "including," "having," and "containing" are open-ended terms, meaning that they include but are not limited to. The terms "an embodiment," "a specific embodiment," "some embodiments," and "for example," etc., refer to specific features, structures, or characteristics described in connection with that embodiment or example that are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. The order of steps involved in the various embodiments is used to illustrate the implementation of this application, and the order of steps is not limited and can be adjusted appropriately as needed.

[0030] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0031] The following methods exist for predicting vehicle response caused by irregular tracks.

[0032] Firstly, the focus can be on establishing mechanistic models to simulate the nonlinear dynamic behavior of vehicles and predict their responses. For example, commercial software such as SIMPACK can be used to build three-dimensional vehicle-track dynamic models to study the relationship between track irregularities and the dynamic performance of rail vehicles. However, the performance of these models depends on the reliability of their parameters, which are easily affected by various real-world variables. Furthermore, because the actual parameters of the vehicle-track system are difficult to obtain and change dynamically over time, applying theoretical models to track maintenance practice is challenging. In addition, numerical iterative methods for solving mechanistic models are extremely time-consuming.

[0033] Secondly, the vehicle-track system can be described using a linear transfer function, and parameters can be estimated based on system identification theory. However, system identification theory can only be applied to linear systems and under constant speed conditions, which limits its ability to predict vehicle response.

[0034] In summary, current actual detection results for vehicle dynamic response are limited, and most vehicle dynamic response data is currently obtained through simulation. However, simulation models have high parameter requirements and are easily affected by real-world conditions. To address these issues, this invention provides a method for predicting vehicle dynamic response, thereby improving the prediction accuracy and speed of vehicle dynamic response data. See [link to relevant documentation]. Figure 1 The method may include:

[0035] Step 101: Obtain target track irregularity data;

[0036] Step 102: Input the above target track irregularity data into the vehicle response prediction model to obtain the vehicle dynamic response prediction data corresponding to the target track irregularity data; the above vehicle response prediction model is a deep learning model with a convolutional neural network structure that introduces a coordinate attention mechanism and a multi-scale attention mechanism structure. The above vehicle response prediction model is obtained by training the above deep learning model with historical data; the above historical data includes historical track irregularity data and corresponding historical vehicle dynamic response prediction data.

[0037] The aforementioned coordinate attention mechanism is used to determine the channel relationships and spatial mileage point locations of the track irregularity data, and to generate the coordinate attention weights of the track irregularity data; the aforementioned multi-scale attention mechanism is used to capture the long-term and short-term trends of the track sequence.

[0038] The aforementioned vehicle response prediction model includes a CA-CNN structure and a MUSE structure. The CA-CNN structure is used to extract the first track irregularity feature from the input track irregularity data through convolution. Based on the coordinate attention weights of the track irregularity data and the first track irregularity feature, the second track irregularity feature is obtained. The MUSE structure is used to combine the long-term and short-term trends of the track sequence and perform deep convolution on the second track irregularity feature to obtain the vehicle response prediction data corresponding to the input track irregularity data.

[0039] In practice, the first step is to obtain data on the irregularity of the target trajectory.

[0040] In this embodiment, the target track irregularity data includes measured track geometry data and vehicle simulation data.

[0041] For example, track measurement geometric data may include data from track inspection trains detecting track irregularities on high-speed railway lines; vehicle simulation data may include data obtained by simulating vehicle response using a multibody model of the vehicle system and SIMPACK software.

[0042] The measured data of vehicle dynamic response in subsequent steps may include data obtained by inspecting vehicle response using a track inspection train.

[0043] Specifically, the measured-simulation dataset contains four track irregularity data points from three high-speed railways: left elevation / reduction, right elevation / reduction, left track alignment, and right track alignment. The vehicle response data includes 14 items: wheel-rail forces (left and right vertical forces on axles 1, 2, 3, and 4), load reduction rate (axles 1, 2, 3, and 4), and vehicle vertical acceleration (front and rear vertical acceleration of the vehicle body).

[0044] In the above embodiment, the CA-CNN-MUSE model is trained using measured track geometry data and simulated vehicle data to obtain a vehicle response prediction model. The measured vehicle response data is then used to test the model's performance after training.

[0045] In practice, after acquiring the target track irregularity data, the target track irregularity data is input into the vehicle response prediction model to obtain the vehicle dynamic response prediction data corresponding to the target track irregularity data.

[0046] In this embodiment, the vehicle response prediction model is a deep learning model with a convolutional neural network structure that incorporates a coordinate attention mechanism and a multi-scale attention mechanism structure. The vehicle response prediction model is obtained by training the deep learning model with historical data. The historical data includes historical data on track irregularities and corresponding historical prediction data of vehicle dynamic response.

[0047] In the above embodiments, in order to link track geometric irregularities with vehicle response and improve track quality assessment standards and track maintenance, a large amount of dynamic track irregularity detection data of high-speed railways can be used to train a vehicle response prediction model, which is a prediction model of vehicle dynamic response based on deep learning.

[0048] In the embodiment, the coordinate attention mechanism described above is used to determine the channel relationship and spatial mileage point position of the track irregularity data, and to generate the coordinate attention weight of the track irregularity data; the multi-scale attention mechanism described above is used to capture the long-term and short-term trends of the track sequence.

[0049] In the above embodiment, the CNN structure consists of alternating convolutional and pooling layers to extract different features of track irregularities. The results of the CNN structure are input into two stacked multi-scale attention layers, each consisting of a multi-head self-attention mechanism and depthwise separable convolutions, encoding global and local relationships in parallel. Coordinate attention is added to the CNN to focus on important feature channels and important mileage locations. Fully connected layers before and after the multi-scale self-attention layers perform non-linear mapping and change the data dimensionality. Finally, the vehicle response prediction is output through the fully connected layers.

[0050] In this embodiment, the vehicle response prediction model includes a CA-CNN structure and a MUSE structure. The CA-CNN structure is used to extract a first track irregularity feature from the input track irregularity data through convolution. Based on the coordinate attention weights of the track irregularity data and the first track irregularity feature, a second track irregularity feature is obtained. The MUSE structure is used to combine the long-term and short-term trends of the track sequence and perform a deep convolution on the second track irregularity feature to obtain the vehicle response prediction data corresponding to the input track irregularity data.

[0051] Among them, the coordinate attention mechanism is called Coordinate Attention (CA); the multi-scale attention mechanism is called Multi-scale Attention (MUSE).

[0052] In this embodiment, a convolutional neural network model with coordinate attention mechanism and multi-scale attention mechanism is introduced, and the resulting vehicle response prediction model can be called the CA-CNN-MUSE model, which is used to predict vehicle response.

[0053] It is worth noting that: Coordinate Attention (CA) is introduced into CNN to focus on important channel relationships and spatial odometry locations, and then the convolutional neural network is used to learn the characteristics of track irregularities; Multi-scale Attention (MUSE) is used to capture the long-term and short-term trends of the track sequence.

[0054] In one embodiment, the CA-CNN structure includes a CA structure and a CNN structure; the CNN structure includes two convolutional layers, two max pooling layers, and a stretching layer; the two convolutional layers have 4 and 8 kernels respectively, a kernel size of 1×5, and a stride of 1; the two max pooling layers have a kernel size of 1×2 and a stride of 2; the stretching layer is used to compress the multidimensional vector of the first track irregularity feature into a one-dimensional feature vector to obtain the first track irregularity feature representing the global feature.

[0055] In one embodiment, the MUSE structure includes three convolutional substructures and a gating structure; each convolutional substructure contains multiple convolutional kernels with kernel sizes of 1, 3, and 5; the convolutional substructures are used to capture features of different ranges; the gating structure is used to adaptively adjust the weights of different convolutional substructures based on a gating mechanism to aggregate information from different convolutional substructures.

[0056] For example, Figure 2 The vehicle response prediction model shown here processes the input data to obtain the output data as follows:

[0057] ① The input data is the measured track geometry data, and the simulated vehicle response is used as the standard output;

[0058] ② Pass the input data into the first layer of the CNN, and then merge the output of the first layer and the result of the input data output from CA into the second layer of the CNN as the output of the entire CA-CNN;

[0059] ③ The output of CA-CNN is passed through a fully connected layer and then used as the input to MUSE. The following formula (3) is because MUSE itself consists of two structures (see Figure 2 (Right side);

[0060] ④ The output of the MUSE structure is passed through a fully connected layer and compared with the vehicle response obtained from the simulation. The result is then passed to the loss function to automatically optimize and adjust the model parameters.

[0061] The following combination Figure 2 The CNN structure, CA structure, and MUSE structure described above will be explained in detail below:

[0062] 1. CNN module (i.e., CNN structure)

[0063] Since the maximum manageable wavelength for track irregularities is 120m, to fully acquire long-range wavelength information, we use all track irregularity data within a 120m range as input. We utilize a CNN to learn the features of the input vector and then input the CNN output features into a multi-scale attention layer to predict the vehicle response at the current mileage point. Assuming the current mileage point is t, we predict the vehicle response at mileage point T as follows:

[0064]

[0065] The input should now be:

[0066] X = {x (t-L+1) ,...,x (t-1) ,x (t) ],[x (t-L+2) ,...,x (t) ,x (t+1) ],...,[x (t-L+T) ,...,x (t+T) ,x (t+T-1) ]} (2)

[0067] in, x (t) Let C be the orbital irregularity vector. Let L be the K-dimensional vehicle response vector, and L be the number of mileage points in the 120m segment. The input sequence size is T×L×C, and the output sequence size is T×1×K.

[0068] The CNN consists of two convolutional layers (Conv1D) with 4 and 8 kernels respectively, a kernel size of 1 × 5, and a stride of 1. It also includes two max-pooling layers with a kernel size of 1 × 2 and a stride of 2. By combining convolution and max-pooling operations, multidimensional features related to the track irregularities are extracted. A stretching layer compresses the multidimensional feature vector into a one-dimensional feature vector, yielding the global features.

[0069] 2. CA module (i.e., CA structure)

[0070] The aforementioned CA structure is inserted between two layers of a CNN model, introducing coordinate attention to focus on important channel relationships and spatial odometry locations. This modified CNN could actually be named CA-CNN because it incorporates a multi-scale attention mechanism (MUSE) to capture the long-term and short-term trends of the orbital sequence; therefore, the complete model is named CA-CNN-MUSE.

[0071] Introducing coordinate attention into a CNN allows us to focus on the odometry locations in the spatial dimension L' and the channels in the channel dimension C', which have a significant impact on the output. The input size for coordinate attention is T×L'×C'. After 1D average pooling in dimension T, the size becomes 1×L'×C'. 1×1 convolutions are then used to reduce and increase the dimensionality of the channels, where r is the reduction factor. Finally, the coordinate attention weights are generated using the sigmoid function.

[0072] 3. MUSE module (i.e., MUSE structure)

[0073] Multi-scale attention diagram as follows Figure 2 As shown.

[0074] This module comprises two parts: a multi-head self-attention mechanism for capturing global features and a depthwise convolution for capturing local features. For an input sequence X, the output Y through the multi-scale attention layer can be represented as:

[0075] Y = Attention(X) + Conv(X) (3)

[0076] The multi-head self-attention mechanism can handle long-term dependencies. In this module, the input sequence X is mapped to three different representations: the query matrix Q, the key matrix K, and the value matrix V.

[0077] Q,K,V=Linear1(X),Linear2(X),Linear3(X) (4)

[0078] The output expression is:

[0079]

[0080] Attention(X)=Attention(Q,K,V)=σ(QW Q ,KW K VW V W O (6)

[0081] Among them, W Q W K W V W O These are projection parameters, V = XW V .

[0082] The convolutional module is used to capture local contextual sequence information within the same mapping space. This module is based on depthwise separable convolution and uses three convolutional sub-modules, each containing multiple convolutional kernels of sizes 1, 3, and 5, to capture features of different ranges. Simultaneously, a gating mechanism is introduced to adaptively adjust the weights of different convolutional units to aggregate information from different convolutional sub-modules. Depthwise convolution first performs independent convolution on each channel, followed by ordinary convolution. The computation process of the convolutional module can be represented as:

[0083]

[0084] Where α is the weighting coefficient.

[0085] The output of a single convolutional submodule can be represented as:

[0086] Conv k (X) = Depth_conv k (V)W out (8)

[0087] In one embodiment, it also includes:

[0088] Acquire measured data of vehicle dynamic response corresponding to target track irregularities;

[0089] The mean square error between the measured vehicle dynamic response data and the predicted vehicle dynamic response data is used as the loss function of the vehicle response prediction model.

[0090] For example, for each dataset, the ratio of training data to test data is 7:3. During training, the loss function is the mean squared error of the actual and predicted vehicle responses, with L1-norm and L2-norm regularization terms added to the model parameters.

[0091]

[0092] Where T is the sequence length, W represents all trainable model parameters, and λ1 and λ2 are regularization coefficients. The learning rate is set to 0.001, and the optimizer is the Adam algorithm.

[0093] In one embodiment, it also includes:

[0094] Based on mean absolute error, root mean square error, Hill's inequality coefficients and / or correlation coefficients, the performance of the vehicle response prediction model is evaluated using measured vehicle dynamic response data and predicted vehicle dynamic response data.

[0095] In the above embodiments, four indicators can be used to evaluate the model performance: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Hill's Inequality Coefficient (TIC), and Correlation Coefficient (ρ). The calculation formulas for each indicator are as follows:

[0096]

[0097]

[0098]

[0099]

[0100] In the formula, M is the length of the test data; y k and These are the k-th sample values ​​of the actual value and the predicted value, respectively; and , respectively, represent the expected value of the actual value and the model's predicted value. MAE and RMSE reflect the absolute accuracy of the prediction; the smaller their values, the better the model's performance. TIC and ρ are relative accuracy indicators; a smaller TIC (from 0 to 1) means higher accuracy. ρ ranges from -1 to 1; the closer its absolute value is to 1, the higher the accuracy.

[0101] The following, in conjunction with Tables 1 to 4, explains the beneficial effects of the vehicle response prediction model provided in the embodiments of the present invention:

[0102] To evaluate the CA-CNN-MUSE model proposed in this embodiment of the invention, this invention constructs multiple comparative models and tests them on a real-world simulation dataset. The experimental results are shown in Table 1.

[0103] The LSTM model consists of two stacked LSTM layers, each with 64 hidden nodes. The CNN-LSTM network shares the same CNN modules as the proposed CA-CNN-MUSE network, and also has two stacked LSTM layers. CA-CNN-LSTM adds the CA modules to the CNN-LSTM. The CNN-MUSE network replaces the LSTM modules in the CNN-LSTM network with MUSE modules.

[0104] To compare the performance of different models, the accuracy metrics of each vehicle's response were averaged to evaluate the overall model accuracy. In addition, more metrics were used, including the number of parameters, the number of connections, and inference time on the test set. From Table 1, which lists the performance metrics of different models, it can be seen that:

[0105] Table 1

[0106]

[0107] (1) Adding CA to the CNN-LSTM model improves the accuracy metrics of RMSE, MAE, and TIC. Replacing LSTM with MUSE in CNN-LSTM also improves the accuracy metrics. Using both CA and MUSE, CA-CNN-MUSE achieves the best RMSE, MAE, and TIC.

[0108] (2) Replacing LSTM with MUSE in CNN-LSTM increases the number of parameters and connections, but reduces inference time because MUSE's multi-head attention can be computed in parallel. After adding the CA module to CNN-MUSE, the inference time only increases by 0.04s.

[0109] In summary, CA-CNN-MUSE outperforms other models in terms of evaluation metrics.

[0110] 2. In order to reflect the experimental results on different railway lines, this embodiment of the invention studies the performance of the model on different lines based on the measured-simulation dataset, as shown in Table 2, which is a table of performance indicators for different lines.

[0111] It can be seen that CA-CNN-MUSE performs better on lines 2 and 3 compared to CNN-LSTM.

[0112] Table 2

[0113]

[0114] 3. The accuracy metrics for the responses of the 14 vehicles in the measured-simulation dataset are shown in Table 3. Table 3 illustrates the accuracy metrics for each vehicle response parameter in the measured-simulation data.

[0115] It can be seen that the CA-CNN-MUSE model can effectively estimate wheel-rail vertical force, load reduction rate, and vehicle body vertical acceleration.

[0116] Table 3

[0117]

[0118] 4. To reflect the experimental results on the actual test dataset.

[0119] This invention also trained and tested the proposed CA-CNN-MUSE model on a real-world track measurement dataset. The real-world dataset consisted of 17km of detection data extracted from a comprehensive track inspection train database for a high-speed railway line. Track irregularities included 11 parameters: left elevation / reduction, right elevation / reduction, left track alignment, right track alignment, long-wave left elevation / reduction, long-wave right elevation / reduction, long-wave left track alignment, long-wave right track alignment, levelness, triangular irregularities, and track gauge. Vehicle response included 6 dynamic responses: left vertical force, left lateral force, right vertical force, right lateral force, vehicle lateral acceleration, and vehicle vertical acceleration.

[0120] Using 11 track irregularities and vehicle speed as inputs to the network, the wavelength components below 2m in the signal are first removed by wavelet decomposition and reconstruction, and then input into the network to predict the vehicle response.

[0121] Table 4 summarizes the accuracy metrics of the CA-CNN-MUSE model, which are the performance metrics of each vehicle response parameter in the measured data. It can be seen that the CA-CNN-MUSE model can effectively estimate wheel-rail vertical force and acceleration, with good prediction accuracy for vehicle body vertical acceleration.

[0122] Table 4

[0123] Output RMSE MAE ρ TIC left vertical force 1259.4704 936.4605 0.7313 0.0096 Right vertical force 1214.3663 908.8109 0.7258 0.0123 Vertical acceleration of vehicle body 0.0076 0.0061 0.8993 0.2433

[0124] In summary, the main advantages of deep learning are its powerful nonlinear modeling capabilities and end-to-end training, ultimately leading to a significant improvement in model accuracy. This invention establishes a CA-CNN-MUSE model, which combines CNN and MUSE to estimate vehicle responses. Experiments show that compared to LSTM and CNN-LSTM models, this model improves prediction accuracy while maintaining good computational speed.

[0125] Specifically, the prediction accuracy of the CA-CNN-MUSE model proposed in this embodiment of the invention is higher than that of the traditional LSTM model and CNN-LSTM model, and its inference speed is higher than that of the CNN-LSTM model. The wheel-rail vertical force and vehicle acceleration waveforms and PSD predicted by the CA-CNN-MUSE model are in good agreement with the actual data, making it suitable for multi-body simulation models of vehicle systems and measured data from actual high-speed lines.

[0126] Of course, it is understood that there may be other variations of the above detailed process, and all such variations should fall within the protection scope of this invention.

[0127] In this embodiment of the invention, target track irregularity data is acquired; the target track irregularity data is input into a vehicle response prediction model to obtain vehicle dynamic response prediction data corresponding to the target track irregularity data; the vehicle response prediction model is a deep learning model with a convolutional neural network structure incorporating a coordinate attention mechanism and a multi-scale attention mechanism structure, and is obtained by training the deep learning model with historical data; the historical data includes historical track irregularity data and corresponding historical vehicle dynamic response prediction data; wherein, the coordinate attention mechanism is used to determine the channel relationship and spatial mileage point position of the track irregularity data, and to generate coordinate attention weights for the track irregularity data; the multi-scale attention mechanism is used to capture the long-term and short-term trends of the track sequence; the vehicle response prediction model includes a CA-CNN structure and a MUSE structure; the CA-CNN structure is used to obtain the target track irregularity data from the input... In the track irregularity data, convolution extracts the first track irregularity feature; based on the coordinate attention weights of the track irregularity data and the first track irregularity feature, the second track irregularity feature is obtained; the MUSE structure is used to combine the long-term and short-term trends of the track sequence and perform deep convolution on the second track irregularity feature to obtain the vehicle response prediction data corresponding to the input track irregularity data. By introducing coordinate attention mechanism and multi-scale attention mechanism into the vehicle response prediction model, the prediction of vehicle dynamic response based on track irregularity data can be accurately realized. This not only improves the prediction accuracy of the vehicle response prediction model for vehicle dynamic response data, but also improves the prediction speed of vehicle dynamic response data. It solves the problem that the prediction process is time-consuming and labor-intensive due to the susceptibility of the constructed three-dimensional vehicle-track dynamics model to the influence of the real environment under the existing technology, and also solves the limitation of vehicle response prediction due to the linear transfer function under the existing technology.

[0128] This invention also provides a vehicle dynamic response prediction device, as described in the following embodiments. Since the principle behind this device is similar to that of the vehicle dynamic response prediction method, its implementation can be found in the implementation of the vehicle dynamic response prediction method; repeated details will not be elaborated further.

[0129] This invention also provides a vehicle dynamic response prediction device to improve the prediction accuracy and speed of vehicle dynamic response data, such as... Figure 3 As shown, the device includes:

[0130] Track irregularity data acquisition module 301 is used to acquire target track irregularity data;

[0131] The vehicle dynamic response data prediction module 302 is used to input the aforementioned target track irregularity data into the vehicle response prediction model to obtain the vehicle dynamic response prediction data corresponding to the target track irregularity data. The aforementioned vehicle response prediction model is a deep learning model with a convolutional neural network structure incorporating a coordinate attention mechanism and a multi-scale attention mechanism structure. The aforementioned vehicle response prediction model is obtained by training the aforementioned deep learning model with historical data. The aforementioned historical data includes historical track irregularity data and corresponding historical vehicle dynamic response prediction data.

[0132] The aforementioned coordinate attention mechanism is used to determine the channel relationships and spatial mileage point locations of the track irregularity data, and to generate the coordinate attention weights of the track irregularity data; the aforementioned multi-scale attention mechanism is used to capture the long-term and short-term trends of the track sequence.

[0133] The aforementioned vehicle response prediction model includes a CA-CNN structure and a MUSE structure. The CA-CNN structure is used to extract the first track irregularity feature from the input track irregularity data through convolution. Based on the coordinate attention weights of the track irregularity data and the first track irregularity feature, the second track irregularity feature is obtained. The MUSE structure is used to combine the long-term and short-term trends of the track sequence and perform deep convolution on the second track irregularity feature to obtain the vehicle response prediction data corresponding to the input track irregularity data.

[0134] In one embodiment, the target track irregularity data includes measured track geometry data and vehicle simulation data.

[0135] In one embodiment, the CA-CNN structure includes a CA structure and a CNN structure; the CNN structure includes two convolutional layers, two max pooling layers, and a stretching layer; the two convolutional layers have 4 and 8 kernels respectively, a kernel size of 1×5, and a stride of 1; the two max pooling layers have a kernel size of 1×2 and a stride of 2; the stretching layer is used to compress the multidimensional vector of the first track irregularity feature into a one-dimensional feature vector to obtain the first track irregularity feature representing the global feature.

[0136] In one embodiment, the MUSE structure includes three convolutional substructures and a gating structure; each convolutional substructure contains multiple convolutional kernels with kernel sizes of 1, 3, and 5; the convolutional substructures are used to capture features of different ranges; the gating structure is used to adaptively adjust the weights of different convolutional substructures based on a gating mechanism to aggregate information from different convolutional substructures.

[0137] In one embodiment, such as Figure 4 As shown, it also includes:

[0138] Loss function determination module 401 is used for:

[0139] Acquire measured data of vehicle dynamic response corresponding to target track irregularities;

[0140] The mean square error between the measured vehicle dynamic response data and the predicted vehicle dynamic response data is used as the loss function of the vehicle response prediction model.

[0141] In one embodiment, it also includes:

[0142] The model performance evaluation module is used for:

[0143] Based on mean absolute error, root mean square error, Hill's inequality coefficients and / or correlation coefficients, the performance of the vehicle response prediction model is evaluated using measured vehicle dynamic response data and predicted vehicle dynamic response data.

[0144] This invention provides an embodiment of a computer device for implementing all or part of the above-described vehicle dynamic response prediction method. Specifically, the computer device includes the following components:

[0145] The computer device comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between related devices; the computer device can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the computer device can be implemented with reference to the embodiments of the method for predicting vehicle dynamic response and the device for predicting vehicle dynamic response, the contents of which are incorporated herein by reference, and repeated details will not be described again.

[0146] Figure 5 This is a schematic block diagram illustrating the system configuration of the computer device 1000 according to an embodiment of this application. Figure 5 As shown, the computer device 1000 may include a central processing unit 1001 and a memory 1002; the memory 1002 is coupled to the central processing unit 1001. It is worth noting that... Figure 5 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0147] In one embodiment, the vehicle dynamic response prediction function can be integrated into the central processing unit 1001. The central processing unit 1001 can be configured to perform the following control:

[0148] Acquire data on the target orbital irregularities;

[0149] The target track irregularity data is input into the vehicle response prediction model to obtain the vehicle dynamic response prediction data corresponding to the target track irregularity data. The vehicle response prediction model is a deep learning model with a convolutional neural network structure that incorporates a coordinate attention mechanism and a multi-scale attention mechanism structure. The vehicle response prediction model is trained on the deep learning model using historical data. The historical data includes historical track irregularity data and corresponding historical vehicle dynamic response prediction data.

[0150] The coordinate attention mechanism is used to determine the channel relationships and spatial mileage point positions of the track irregularity data, and to generate the coordinate attention weights of the track irregularity data; the multi-scale attention mechanism is used to capture the long-term and short-term trends of the track sequence.

[0151] The vehicle response prediction model includes a CA-CNN structure and a MUSE structure. The CA-CNN structure is used to extract a first track irregularity feature from the input track irregularity data through convolution. A second track irregularity feature is obtained based on the coordinate attention weights of the track irregularity data and the first track irregularity feature. The MUSE structure is used to perform a depthwise convolution on the second track irregularity feature by combining the long-term and short-term trends of the track sequence to obtain the vehicle response prediction data corresponding to the input track irregularity data.

[0152] In another embodiment, the vehicle dynamic response prediction device can be configured separately from the central processing unit 1001. For example, the vehicle dynamic response prediction device can be configured as a chip connected to the central processing unit 1001, and the vehicle dynamic response prediction function can be realized through the control of the central processing unit.

[0153] like Figure 5 As shown, the computer device 1000 may further include: a communication module 1003, an input unit 1004, an audio processor 1005, a display 1006, and a power supply 1007. It is worth noting that the computer device 1000 does not necessarily need to include... Figure 5 All components shown; in addition, the computer device 1000 may also include Figure 5 For components not shown, please refer to existing technologies.

[0154] like Figure 5 As shown, the central processing unit 1001, sometimes also referred to as a controller or operation control, may include a microprocessor or other processor device and / or logic device. The central processing unit 1001 receives input and controls the operation of various components of the computer device 1000.

[0155] The memory 1002 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 1001 may execute the program stored in the memory 1002 to perform information storage or processing, etc.

[0156] Input unit 1004 provides input to central processing unit 1001. This input unit 1004 may be, for example, a keypad or touch input device. Power supply 1007 provides power to computer device 1000. Display 1006 displays images, text, and other display objects. This display may be, for example, an LCD display, but is not limited to this.

[0157] The memory 1002 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs, etc. The memory 1002 can also be some other type of device. The memory 1002 includes a buffer memory 1021 (sometimes referred to as a buffer). The memory 1002 may include an application / function storage unit 1022 for storing application programs and function programs or processes for executing operations of the computer device 1000 via the central processing unit 1001.

[0158] The memory 1002 may also include a data storage unit 1023 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the computer device. The driver storage unit 1024 of the memory 1002 may include various drivers for the computer device for communication functions and / or for performing other functions of the computer device (such as messaging applications, address book applications, etc.).

[0159] The communication module 1003 is a transmitter / receiver 1003 that transmits and receives signals via the antenna 1008. The communication module (transmitter / receiver) 1003 is coupled to the central processing unit 1001 to provide input signals and receive output signals, which can be the same as in a conventional mobile communication terminal.

[0160] Based on different communication technologies, multiple communication modules 1003 can be configured in the same computer device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module (transmitter / receiver) 1003 is also coupled to a speaker 1009 and a microphone 1010 via an audio processor 1005 to provide audio output via the speaker 1009 and receive audio input from the microphone 1010, thereby realizing typical telecommunications functions. The audio processor 1005 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 1005 is also coupled to a central processing unit 1001, enabling on-device recording via the microphone 1010 and on-device playback of stored sound via the speaker 1009.

[0161] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting vehicle dynamic response.

[0162] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described method for predicting vehicle dynamic response.

[0163] In this embodiment of the invention, target track irregularity data is acquired; the target track irregularity data is input into a vehicle response prediction model to obtain vehicle dynamic response prediction data corresponding to the target track irregularity data; the vehicle response prediction model is a deep learning model with a convolutional neural network structure incorporating a coordinate attention mechanism and a multi-scale attention mechanism structure, and the vehicle response prediction model is obtained by training the deep learning model with historical data; the historical data includes historical track irregularity data and corresponding historical vehicle dynamic response prediction data; wherein, the coordinate attention mechanism is used to determine the channel relationship and spatial mileage point position of the track irregularity data, and generate coordinate attention weights for the track irregularity data; the multi-scale attention mechanism is used to capture the long-term and short-term trends of the track sequence; the vehicle response prediction model includes a CA-CNN structure and a MUSE structure; the CA-CNN structure is used to obtain the target track irregularity data from the input... In the track irregularity data, a first track irregularity feature is extracted by convolution; based on the coordinate attention weights of the track irregularity data and the first track irregularity feature, a second track irregularity feature is obtained; the MUSE structure is used to combine the long-term and short-term trends of the track sequence and perform deep convolution on the second track irregularity feature to obtain the vehicle response prediction data corresponding to the input track irregularity data. By introducing coordinate attention mechanism and multi-scale attention mechanism into the vehicle response prediction model, the prediction of vehicle dynamic response based on track irregularity data can be accurately realized. This not only improves the prediction accuracy of the vehicle response prediction model for vehicle dynamic response data, but also improves the prediction speed of vehicle dynamic response data. It solves the problem that the prediction process is time-consuming and labor-intensive due to the susceptibility of the constructed three-dimensional vehicle-track dynamics model to the influence of the real environment under the existing technology, and also solves the limitation of vehicle response prediction due to the linear transfer function under the existing technology.

[0164] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0165] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0166] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0167] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0168] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are 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 within the scope of protection of the present invention.

Claims

1. A method of predicting the dynamic response of a vehicle, characterized in that, The method comprises the following steps: acquiring target track irregularity data; inputting the target track irregularity data into a vehicle response prediction model to obtain vehicle dynamic response prediction data corresponding to the target track irregularity data; the vehicle response prediction model is a deep learning model with a coordinate attention mechanism and a multi-scale attention mechanism structure, and the vehicle response prediction model is obtained by training the deep learning model based on historical data; the historical data comprises track irregularity historical data and corresponding vehicle dynamic response historical prediction data; wherein the coordinate attention mechanism is used to determine the channel relationship and spatial mileage point position of the track irregularity data, and generate a coordinate attention weight of the track irregularity data; the multi-scale attention mechanism is used to capture long and short term trends of track sequences; the vehicle response prediction model comprises a CA-CNN structure and a MUSE structure; the CA-CNN structure is used to extract first track irregularity features from the input track irregularity data; second track irregularity features are obtained based on the coordinate attention weight of the track irregularity data and the first track irregularity features; the MUSE structure is used to combine the long and short term trends of the track sequences to perform deep convolution on the second track irregularity features to obtain vehicle response prediction data corresponding to the input track irregularity data.

2. The method of claim 1, wherein, The target track irregularity data comprises track measured geometry data and vehicle simulation data.

3. The method of claim 1, wherein, The CA-CNN structure comprises a CA structure and a CNN structure; the CNN structure comprises two convolution layers, two max-pooling layers and a stretching layer; the number of convolution kernels of the two convolution layers is 4 and 8 respectively, the size of the convolution kernel is 1x5, and the step is 1; the size of the pooling kernel of the two max-pooling layers is 1x2, and the step is 2; the stretching layer is used to compress the multi-dimensional first track irregularity feature vector into a one-dimensional feature vector to obtain the first track irregularity features representing global features.

4. The method of claim 1, wherein, The MUSE structure comprises three convolution substructures and a gating structure; each convolution substructure comprises a plurality of convolution kernels with sizes of 1, 3 and 5 respectively; the convolution substructure is used to capture features of different ranges; the gating structure is used to adaptively adjust the weights of different convolution substructures based on a gating mechanism to converge information of different convolution substructures.

5. The method of claim 1, wherein, The method further comprises the following steps: acquiring vehicle dynamic response measured data corresponding to the target track irregularity data; using the mean square error of the vehicle dynamic response measured data and the vehicle dynamic response prediction data as a loss function of the vehicle response prediction model.

6. The method of claim 1, wherein, The method further comprises the following steps: performing performance evaluation on the vehicle response prediction model based on the vehicle dynamic response measured data and the vehicle dynamic response prediction data according to the mean absolute error, the root mean square error, the Hill inequality coefficient and / or the correlation coefficient.

7. A device for predicting the dynamic response of a vehicle, characterized in that The method comprises the following steps: an track irregularity data acquisition module is configured to acquire target track irregularity data; The vehicle dynamic response data prediction module is configured to input the target track irregularity data into a vehicle response prediction model to obtain vehicle dynamic response prediction data corresponding to the target track irregularity data; the vehicle response prediction model is a deep learning model with a coordinate attention mechanism and a multi-scale attention mechanism; the vehicle response prediction model is obtained by training the deep learning model based on historical data; the historical data includes track irregularity historical data and corresponding vehicle dynamic response historical prediction data; The coordinate attention mechanism is configured to determine channel relationships and spatial mile point positions of the track irregularity data and generate coordinate attention weights of the track irregularity data; and the multi-scale attention mechanism is configured to capture long-term and short-term trends of track sequences. The vehicle response prediction model includes a CA-CNN structure and a MUSE structure; the CA-CNN structure is configured to extract first track irregularity features from the input track irregularity data by convolution; and the MUSE structure is configured to perform deep convolution on the second track irregularity features based on the coordinate attention weights of the track irregularity data and the first track irregularity features to obtain vehicle response prediction data corresponding to the input track irregularity data.

8. The apparatus of claim 7, wherein, The target track irregularity data includes track measured geometry data and vehicle simulation data.

9. The apparatus of claim 7, wherein, The CA-CNN structure includes a CA structure and a CNN structure; the CNN structure includes two convolution layers, two max-pooling layers and a stretching layer; the number of convolution kernels of the two convolution layers is 4 and 8 respectively, the size of the convolution kernel is 1*5, and the step is 1; the size of the pooling kernel of the two max-pooling layers is 1*2, and the step is 2; and the stretching layer is configured to compress a multi-dimensional first track irregularity feature vector into a one-dimensional feature vector to obtain the first track irregularity features representing global features.

10. The apparatus of claim 7, wherein, The MUSE structure includes three convolution sub-structures and a gating structure; each convolution sub-structure includes multiple convolution kernels with sizes of 1, 3 and 5; the convolution sub-structures are configured to capture features of different ranges; and the gating structure is configured to adaptively adjust weights of different convolution sub-structures based on a gating mechanism to converge information of different convolution sub-structures.

11. The apparatus of claim 7, wherein, Further comprising: The loss function determination module is configured to: obtain vehicle dynamic response measured data corresponding to the target track irregularity data; use mean square errors between the vehicle dynamic response measured data and the vehicle dynamic response prediction data as a loss function of the vehicle response prediction model.

12. The apparatus of claim 7, wherein, Further comprising: The model performance evaluation module is configured to: perform performance evaluation on the vehicle response prediction model based on mean absolute errors, root mean square errors, Hill inequality coefficients and / or correlation coefficients according to the vehicle dynamic response measured data and the vehicle dynamic response prediction data.

13. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the method of any one of claims 1 to 6.

14. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the method in any one of claims 1 to 6.

15. A computer program product, characterised in that, The computer program product comprises a computer program, and the computer program is executed by the processor to implement the method in any one of claims 1 to 6.

Images