A four-body coupled vehicle base vibration noise prediction optimization method and system

Abstract

The present application relates to a kind of four-body coupling vehicle base vibration noise prediction optimization method and system, it is related to rail transit operation optimization technical field, including obtaining vehicle base multi-physical field correlation monitoring data set;System level assembly is carried out in multi-physical field collaborative simulation environment, generates coupling system state parameter matrix;Sound ray bending trajectory data is generated;And vehicle operation working condition time series data is loaded to wave equation solver, and vibration noise propagation characteristic field distribution data is output;Build network topology structure to form prediction model, execute back propagation algorithm to train network parameter to convergence state;Design vehicle scheduling scheme chromosome coding structure to generate candidate solution population;To initial solution set implements selection, crossover and variation genetic operation, and the output vehicle base operation scheduling real-time optimization scheme.The beneficial effects of the present application are to greatly improve the accuracy and generalization ability of future period vibration noise prediction, realize the fundamental change from passive vibration isolation management to active prediction optimization.

Description

Technical Field

[0001] This invention relates to the field of rail transit operation and maintenance optimization technology, and more specifically, to a method and system for predicting and optimizing vibration and noise at a four-body coupled vehicle depot. Background Technology

[0002] Currently, my country's rail transit system continues to expand, and vehicle depots, as core facilities for daily train maintenance and parking, are facing increasingly prominent vibration and noise problems caused by frequent train entry and exit, low-speed travel, and train formation operations. Traditional vibration and noise control methods mainly rely on offline assessment using single-point vibration monitoring combined with empirical formulas, or passive noise reduction using fixed vibration isolation devices.

[0003] While existing technologies can locally improve wheel-rail excitation characteristics through methods such as track grinding and elastic fastener adjustment, they lack systematic modeling of the multi-body coupled dynamics mechanism of vehicle-track-foundation-building, making it impossible to accurately predict the propagation law and energy distribution of vibration and noise under time-varying operating conditions. Furthermore, existing prediction models are mostly limited to single-physics field analysis, such as considering only track irregularities or soil fluctuations, making it difficult to integrate multi-source heterogeneous data such as suspension parameter degradation, structural modal evolution, and environmental temperature and humidity gradients. This results in insufficient prediction accuracy and weak generalization ability. At the scheduling optimization level, traditional methods are mostly static scheduling for fixed time periods, failing to dynamically link vibration and noise prediction results with operation and maintenance decisions. This makes it difficult to achieve a balance between proactively avoiding periods of excessive vibration and noise and optimizing operational efficiency, resulting in a persistent contradiction between environmental complaint risks and maintenance efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for predicting and optimizing vibration and noise at the base of a four-body coupled vehicle, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0005] Firstly, this application provides a method for predicting and optimizing vibration and noise at a four-body coupled vehicle base, including:

[0006] Acquire a multi-physics field correlation monitoring dataset for the vehicle base, which includes dynamic irregularity detection data of the vehicle base track, vibration response data of the vehicle suspension system, dynamic characteristic parameters of the foundation soil, modal frequency data of the building structure, environmental temperature and humidity monitoring data, and time series data of vehicle operating conditions.

[0007] Based on the multiphysics field correlation monitoring dataset of the vehicle base, a vehicle-track coupled dynamics model is constructed and the vehicle track subsystem matrix is ​​derived. Simultaneously, a foundation-structure coupled dynamics model is established and the soil structure subsystem matrix is ​​derived. The wheel-rail contact nonlinear force function and soil structure boundary impedance parameters are defined. Using the Lagrange multiplier method, the vehicle track subsystem matrix, soil structure subsystem matrix, wheel-rail contact nonlinear force function, and soil structure boundary impedance parameters are assembled hierarchically in a multiphysics field co-simulation environment to form an overall coupled dynamics equation set. A variable-step implicit integration algorithm is then used to solve the overall coupled dynamics equation set, generating the coupled system state parameter matrix.

[0008] The coupled system state parameter matrix is ​​input into a preset energy field analysis module to extract the system vibration energy distribution characteristics and output vibration energy density field data. Simultaneously, environmental temperature and humidity monitoring data is imported into the acoustic propagation calculation engine to construct a refraction and propagation model of sound waves in a temperature and humidity gradient field, generating sound ray bending trajectory data. Vehicle operating condition time series data is loaded into the wave equation solver to establish a dispersion characteristic model of vibration waves in the foundation soil, obtaining frequency-wavenumber domain response data. Then, a finite element-boundary element coupled algorithm is used to perform spatiotemporal dual discretization and integration operations on the vibration energy density field data, sound ray bending trajectory data, and frequency-wavenumber domain response data to output vibration and noise propagation characteristic field distribution data.

[0009] The vibration and noise propagation feature field distribution data is imported into a spatiotemporal attention mechanism deep learning prediction framework to build a network topology to form a prediction model. Historical field distribution data is input into the input layer of the prediction model for data standardization mapping to generate standardized feature vectors. The standardized feature vectors are encoded with temporal dependencies using a long short-term memory network layer to extract the hidden state feature sequence. The hidden state feature sequence is then mapped to the vibration and noise prediction values ​​for future periods through a fully connected layer. The backpropagation algorithm is executed to train the network parameters until convergence. After convergence, the network outputs the vibration and noise prediction data sequence for future periods.

[0010] Based on the noise prediction data sequence for future time periods, a chromosome encoding structure for vehicle scheduling schemes is designed to generate a candidate solution population. A fitness function is defined with the dual objectives of minimizing noise overscalar and maximizing operational efficiency, and the fitness function is used to evaluate individual candidates in the candidate solution population. The candidate solution population after individual evaluation is initialized as the initial solution set. Selection, crossover, and mutation genetic operations are performed on the initial solution set to drive the population to iterative evolution. When the convergence condition is met, the Pareto optimal front solution set is obtained. The optimal individual in the Pareto optimal front solution set is decoded, and the real-time optimization scheme for vehicle base operation and maintenance scheduling is output.

[0011] Preferably, the acquisition of the vehicle base multi-physics field correlation monitoring data set includes:

[0012] A track inspection vehicle and a laser displacement sensor array are deployed in the track section of the vehicle depot. Spatial positioning is performed based on the kilometer marker, and data on track gauge, elevation, and directional irregularities are collected. At the same time, millimeter-wave radar is used to scan the corrugation on the rail surface to obtain corrugation depth spectrum data. The track gauge, elevation, and directional irregularity data and corrugation depth spectrum data are stored in a unified spatiotemporal two-dimensional array structure. The time dimension corresponds to the inspection date, and the spatial dimension corresponds to the line mileage. The data is then converted into an excitation spectrum matrix, which constitutes the external excitation input of the vehicle-track coupling model.

[0013] Inertial measurement units were installed at three locations: the axle box, frame, and body of the monitored vehicle bogie. The vertical, lateral, and longitudinal vibration acceleration time history data of the vehicle were collected when it was running at low speed on the track in the depot. Zero drift elimination and trend term removal were performed on the collected vibration acceleration time history data. The dynamic displacement and dynamic velocity of the suspension components were obtained through integral calculation to form the vibration response data of the vehicle suspension system.

[0014] Vibration sensor arrays and environmental temperature and humidity sensors were deployed in key areas of the vehicle base's throat area, maintenance depot, and main structure of the office building. Dynamic characteristic parameters of the foundation soil were obtained through on-site wave velocity testing and indoor resonant column tests. Modal frequency data of the building structure were obtained through hammer impact testing or environmental vibration testing. Time series data of vehicle operating conditions were extracted from the vehicle operation management system. All data were timestamped and unified to the same time base using linear interpolation to generate a multi-physics field correlation monitoring dataset for the vehicle base. This dataset is used to provide boundary conditions for the coupling model parameters.

[0015] Preferably, based on the multiphysics-based monitoring dataset of the vehicle base, a vehicle-track coupled dynamics model is constructed and the vehicle-track subsystem matrix is ​​derived. Simultaneously, a foundation-structure coupled dynamics model is established and the soil structure subsystem matrix is ​​derived. The wheel-rail contact nonlinear force function and soil structure boundary impedance parameters are defined. The Lagrange multiplier method is used to assemble the vehicle-track subsystem matrix, soil structure subsystem matrix, wheel-rail contact nonlinear force function, and soil structure boundary impedance parameters hierarchically in a multiphysics-based co-simulation environment, forming a system of coupled dynamics equations. A variable-step implicit integration algorithm is called to solve the system of coupled dynamics equations, generating a coupled system state parameter matrix, including:

[0016] By calling the excitation spectrum matrix and the modal space vibration state vector constructed by modal filter transformation of the vibration response data of the vehicle suspension system, a vehicle-track coupled dynamic model is constructed. The vehicle system is described by multi-rigid-body dynamics, the car body, frame, and wheelset are defined as rigid bodies, the suspension elements are defined as force elements, and the track structure is discretized into beam elements and plate elements using finite element method. The vehicle-track subsystem matrix is ​​then assembled to obtain the vehicle mass matrix, which includes the vehicle mass matrix, vehicle stiffness matrix, track mass matrix, and track stiffness matrix.

[0017] By calling the dynamic characteristic parameters of the foundation soil and the modal frequency data of the building structure obtained from the multiphysics field correlation monitoring dataset of the vehicle base, a coupled dynamic model of foundation-building is established. The foundation is established as a half-space model using the boundary element method, and the building is described by the modal superposition method. The boundary impedance parameters of the soil structure are obtained by inversion of the measured frequency response function. The boundary impedance parameters of the soil structure characterize the energy transfer characteristics of the vibration wave at the interface between the foundation and the building. The soil structure subsystem matrix is ​​derived, which includes the foundation impedance matrix and the building modal mass matrix.

[0018] The vehicle track subsystem matrix and the soil structure subsystem matrix are connected through the wheel-rail contact nonlinear force function. The wheel-rail contact nonlinear force function is constructed using Hertz contact theory and Kalker creep theory. The Lagrange multiplier method is used to apply constraints at the interface nodes, and the system is assembled to form an overall coupled dynamic equation set.

[0019] The variable step size implicit integration algorithm is called to solve the system of equations to generate the coupled system state parameter matrix, which includes the system displacement, velocity and acceleration response at each time step.

[0020] Preferably, the coupled system state parameter matrix is ​​input to a preset energy field analysis module to extract the system vibration energy distribution characteristics and output vibration energy density field data; simultaneously, environmental temperature and humidity monitoring data are imported into the acoustic propagation calculation engine to construct a refraction and propagation model of sound waves in a temperature and humidity gradient field, generating sound ray bending trajectory data; and vehicle operating condition time series data are loaded into the wave equation solver to establish a dispersion characteristic model of vibration waves in the foundation soil, obtaining frequency-wavenumber domain response data; then, a finite element-boundary element coupled algorithm is used to perform spatiotemporal domain dual discretization integration on the vibration energy density field data, sound ray bending trajectory data, and frequency-wavenumber domain response data, outputting vibration and noise propagation characteristic field distribution data, including:

[0021] The system calls the state parameter matrix of the coupled system, extracts the velocity vector and stress vector of each unit node step by step, calculates the kinetic energy density and strain energy density for each unit, and accumulates them after traversing all units to obtain the total energy density field of the system. The total energy density field is then gridded according to spatial coordinates to generate vibration energy density field data.

[0022] By calling environmental temperature and humidity monitoring data, the sound velocity field distribution is calculated according to the time series. A temperature and humidity gradient field is established in the sound propagation calculation domain. The ray tracing method is used to solve the refraction path of the sound wave in the gradient field. The sound ray path from the track structure through the soil to the building structure is traced to generate sound ray bending trajectory data.

[0023] Using time-series data of vehicle operation conditions, time-varying boundary conditions are set in the wave equation solver to establish a dispersion characteristic model of vibration waves in the foundation soil. Frequency-wavenumber domain response data are obtained through wavenumber space scanning, where the frequency-wavenumber domain response data describes the propagation speed and attenuation law of vibration waves of different frequencies in the soil. The finite element-boundary element coupled algorithm is used to perform spatiotemporal dual discretization integration on the vibration energy density field data, sound ray bending trajectory data and frequency-wavenumber domain response data. The time discretization adopts the central difference scheme and the spatial discretization adopts the Gaussian integral scheme. After integration, the vibration noise propagation characteristic field distribution data are output.

[0024] Preferably, the step of importing vibration and noise propagation feature field distribution data into a spatiotemporal attention mechanism deep learning prediction framework, building a network topology to form a prediction model, and inputting historical field distribution data into the input layer of the prediction model for data standardization mapping to generate standardized feature vectors includes:

[0025] The vibration and noise propagation feature field distribution data is imported into a spatiotemporal attention mechanism deep learning prediction framework. A neural network topology containing a temporal attention module and a spatial attention module is constructed. The temporal attention module generates attention weights by calculating the dot product similarity between the query vector and the key vector. The spatial attention module extracts spatial features of the vibration and noise field through convolutional layers and generates a spatial weight matrix. The temporal attention weights and spatial attention weights are multiplied element-wise to obtain the spatiotemporal joint attention weights. This weight is applied to the value vector weighted summation to form the attention output features. After stacking multiple attention mechanisms, a batch normalization layer and an activation function layer are connected. The completed network topology is used to receive standardized feature vectors.

[0026] Historical field distribution data is input into the model's input layer, and Z-score standardization mapping is performed. The mean and standard deviation of the time series data for each spatial measurement point are calculated independently to generate standardized feature vectors.

[0027] Preferably, the step involves using a Long Short-Term Memory (LSTM) network layer to encode the temporal dependencies of the standardized feature vectors, extracting the hidden state feature sequence, and mapping the hidden state feature sequence to future noise prediction values ​​through a fully connected layer. The backpropagation algorithm is then executed to train the network parameters until convergence. The converged network outputs a future noise prediction data sequence, including:

[0028] The standardized feature vectors are input into the Long Short-Term Memory network layer for temporal dependency encoding. The number of hidden units and layers of the LSTM are set, and the temporal features are extracted from the forward and backward directions using a bidirectional LSTM structure. The bidirectional hidden state vectors are concatenated to obtain the hidden state feature sequence.

[0029] A fully connected layer is set to map the hidden state feature sequence to the noise prediction value of the future time period. The output dimension of the fully connected layer is set to the product of the number of prediction time periods and the number of measurement points. The mean square error loss function is used to calculate the error between the prediction value and the true value. The backpropagation algorithm is executed to train the network parameters to the convergence state. The convergence condition is set to the validation set loss no longer decreasing after several consecutive rounds. After the network has converged, it outputs the noise prediction data sequence of the future time period.

[0030] Preferably, based on the future noise prediction data sequence, a chromosome encoding structure for the vehicle scheduling scheme is designed to generate a candidate solution population; a fitness function is defined with the dual objectives of minimizing noise overscalar and maximizing operational efficiency, and the candidate solution population is evaluated using the fitness function; the candidate solution population after individual evaluation is initialized as the initial solution set, and selection, crossover, and mutation genetic operations are performed on the initial solution set to drive the population to iterative evolution. When the convergence condition is met, the Pareto optimal front solution set is obtained, the optimal individual in the Pareto optimal front solution set is decoded, and the real-time optimization scheme for vehicle base operation and maintenance scheduling is output, including:

[0031] Based on the noise prediction data sequence for future time periods, a chromosome coding structure for vehicle scheduling scheme is designed. The chromosome gene positions are encoded with real numbers and correspond to the train's entry and exit times and running speed levels on the designated track. A candidate solution population is generated through random initialization.

[0032] Define a dual-objective fitness function that includes minimizing noise superscalar and maximizing operational efficiency, and initialize the candidate solution population to form the initial solution set based on this function;

[0033] Tournament selection, simulated binary crossover, and polynomial mutation genetic operations are sequentially applied to the initial solution set to drive the iterative evolution of the population. When the convergence condition is met, the Pareto optimal frontier solution set is obtained, the optimal individual in the solution set is decoded, and the real-time optimization scheme for vehicle base operation and maintenance scheduling is output.

[0034] Secondly, this application also provides a four-body coupled vehicle base vibration and noise prediction and optimization system, including:

[0035] Acquisition Module: Used to acquire the multi-physics field correlation monitoring dataset of the vehicle base. This dataset includes dynamic irregularity detection data of the vehicle base track, vibration response data of the vehicle suspension system, dynamic characteristic parameters of the foundation soil, modal frequency data of the building structure, environmental temperature and humidity monitoring data, and time series data of vehicle operating conditions.

[0036] Assembly and Generation Module: Based on the multiphysics-based monitoring dataset of the vehicle base, this module constructs a vehicle-track coupled dynamics model and derives the vehicle track subsystem matrix. Simultaneously, it establishes a foundation-structure coupled dynamics model and derives the soil structure subsystem matrix, defining the wheel-rail contact nonlinear force function and soil structure boundary impedance parameters. Using the Lagrange multiplier method, the vehicle track subsystem matrix, soil structure subsystem matrix, wheel-rail contact nonlinear force function, and soil structure boundary impedance parameters are assembled hierarchically in a multiphysics-based co-simulation environment to form a system of overall coupled dynamics equations. A variable-step implicit integration algorithm is then used to solve the system of overall coupled dynamics equations, generating the coupled system state parameter matrix.

[0037] The output module is designed to input the coupled system state parameter matrix into a preset energy field analysis module, extract the system vibration energy distribution characteristics, and output vibration energy density field data. Simultaneously, it imports environmental temperature and humidity monitoring data into the acoustic propagation calculation engine to construct a refraction and propagation model of sound waves in a temperature and humidity gradient field, generating sound ray bending trajectory data. Furthermore, it loads vehicle operating condition time-series data into the wave equation solver to establish a dispersion characteristic model of vibration waves in the foundation soil, obtaining frequency-wavenumber domain response data. Finally, it employs a finite element-boundary element coupled algorithm to perform spatiotemporal dual discretization and integration operations on the vibration energy density field data, sound ray bending trajectory data, and frequency-wavenumber domain response data, outputting vibration and noise propagation characteristic field distribution data.

[0038] Training Module: This module is used to import vibration and noise propagation feature field distribution data into a spatiotemporal attention mechanism deep learning prediction framework, build a network topology to form a prediction model, input historical field distribution data into the input layer of the prediction model for data standardization mapping, and generate standardized feature vectors. A Long Short-Term Memory (LSTM) network layer is used to encode the temporal dependencies of the standardized feature vectors, extract the hidden state feature sequences, and map the hidden state feature sequences to future vibration and noise prediction values ​​through a fully connected layer. The backpropagation algorithm is executed to train the network parameters until convergence. After convergence, the network outputs a future vibration and noise prediction data sequence.

[0039] Iterative Module: Based on the noise prediction data sequence for future time periods, this module designs the chromosome encoding structure of the vehicle scheduling scheme to generate a candidate solution population. It defines a fitness function with the dual objectives of minimizing noise overscalar and maximizing operational efficiency, and uses the fitness function to evaluate individual candidates in the candidate solution population. The candidate solution population after individual evaluation is initialized as the initial solution set. Selection, crossover, and mutation genetic operations are performed on the initial solution set to drive the population to iteratively evolve. When the convergence condition is met, the Pareto optimal front solution set is obtained. The optimal individual in the Pareto optimal front solution set is decoded, and the real-time optimization scheme for vehicle base operation and maintenance scheduling is output.

[0040] Thirdly, this application also provides a four-body coupled vehicle base vibration and noise prediction and optimization device, including:

[0041] Memory, used to store computer programs;

[0042] A processor is used to implement the steps of the four-body coupled vehicle base vibration noise prediction optimization method when executing the computer program.

[0043] Fourthly, this application also provides a readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described optimization method for vibration and noise prediction of a four-body coupled vehicle base.

[0044] The beneficial effects of this invention are as follows:

[0045] This invention constructs a four-body coupled dynamic model of vehicle-track-foundation-building and integrates multi-physics field monitoring data such as track irregularity detection, vehicle suspension response, soil dynamic parameters, building modal frequencies, and environmental temperature and humidity. It systematically solves the problem of insufficient vibration and noise prediction accuracy caused by traditional methods that rely solely on single-point monitoring or single-physics field analysis. Furthermore, it employs the Lagrange multiplier method to assemble the matrices of each subsystem and nonlinear force functions at a system-level hierarchy, and efficiently solves the overall coupled dynamic equations using a variable step-size implicit integration algorithm, significantly improving the computational efficiency and stability of the dynamic response of large-scale complex systems.

[0046] This invention utilizes an energy field analysis module, an acoustic propagation calculation engine, and a wave equation solver to extract vibration energy density field, sound ray bending trajectory, and frequency-wavenumber domain response data, respectively. Based on a finite element-boundary element coupled algorithm, it performs spatiotemporal domain dual discretization integration, achieving accurate characterization of vibration and noise propagation features and a complete description of spatiotemporal evolution. Furthermore, it constructs a spatiotemporal attention mechanism deep learning prediction framework, eliminating dimensional differences in multi-source heterogeneous data through data standardization mapping. Combined with a long short-term memory network layer, it extracts temporal dependencies and generates hidden state feature sequences, effectively capturing the spatiotemporal correlation of vibration and noise field distribution and significantly improving the accuracy and generalization ability of vibration and noise prediction for future periods.

[0047] This invention designs a dual-objective genetic algorithm that uses the minimization of vibration and noise overscalar and the maximization of operational efficiency as fitness functions to evaluate individual chromosomes of scheduling schemes. Through iterative optimization using genetic operations such as selection, crossover, and mutation, a Pareto optimal frontier solution set is obtained. The vibration and noise prediction results are dynamically closed-looped with vehicle scheduling decisions, realizing a fundamental shift from passive vibration isolation to proactive prediction and optimization. This provides vehicle depots with an executable time-speed-workstation collaborative scheduling scheme that balances the dual objectives of environmental friendliness and high operational efficiency.

[0048] Other features and advantages of the invention will be set forth in the following description, or may be learned by practicing the embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of the vibration and noise prediction optimization method for a four-body coupled vehicle base as described in this embodiment of the invention.

[0051] Figure 2 This is a schematic diagram of the structure of the four-body coupled vehicle base vibration and noise prediction optimization system described in this embodiment of the invention;

[0052] Figure 3 This is a schematic diagram of the structure of the four-body coupled vehicle base vibration and noise prediction and optimization device described in this embodiment of the invention.

[0053] In the diagram: 701, Acquisition Module; 702, Assembly and Generation Module; 703, Output Construction Module; 704, Training Module; 705, Iteration Module; 800, Four-Body Coupled Vehicle Base Vibration and Noise Prediction Optimization Equipment; 801, Processor; 802, Memory; 803, Multimedia Component; 804, I / O Interface; 805, Communication Component. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0055] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0056] Example 1:

[0057] This embodiment provides a method for predicting and optimizing vibration and noise at a four-body coupled vehicle base.

[0058] See Figure 1 The figure shows that the method includes steps S100, S200, S300, S400 and S500.

[0059] S100. Obtain the multi-physics field associated monitoring dataset of the vehicle base, which includes dynamic irregularity detection data of the vehicle base track, vibration response data of the vehicle suspension system, dynamic characteristic parameters of the foundation soil, modal frequency data of the building structure, environmental temperature and humidity monitoring data, and time series data of vehicle operating conditions.

[0060] It is understood that step S100 includes S101, S102, and S103, wherein:

[0061] S101. Deploy a track inspection vehicle and a laser displacement sensor array in the track section of the vehicle depot. Use the kilometer marker as a reference for spatial positioning, collect track gauge, elevation, and directional irregularity data, and simultaneously perform millimeter-wave radar scanning on the rail surface corrugation to obtain corrugation depth spectrum data. Store the track gauge, elevation, directional irregularity data and corrugation depth spectrum data in a unified spatiotemporal two-dimensional array structure. The time dimension corresponds to the inspection date, and the spatial dimension corresponds to the line mileage. Convert the data into an excitation spectrum matrix, where the excitation spectrum matrix constitutes the external excitation input of the vehicle-track coupling model.

[0062] It should be noted that this step, through high-precision laser displacement measurement and millimeter-wave radar coordinated scanning, achieves simultaneous acquisition of short-wave and long-wave irregularities in track geometry. Kilometer marker spatial positioning ensures accurate positioning correlation between the excitation source and the track mileage. The spatiotemporal two-dimensional array structure effectively supports the subsequent time-frequency domain excitation loading of the dynamic model. The conversion of the excitation spectrum matrix transforms the measured geometric irregularities into frequency domain excitation inputs that can directly drive the vehicle-track coupling model, significantly improving the realism of the model input and the completeness of the excitation characteristics.

[0063] S102. Install inertial measurement units at three locations: the bogie axle box, frame, and car body of the monitored vehicle. Collect vertical, lateral, and longitudinal vibration acceleration time history data of the vehicle when it runs at low speed on the track in the depot. Eliminate zero drift and remove trend terms from the collected vibration acceleration time history data. Obtain the dynamic displacement and dynamic velocity of the suspension elements through integral calculation to form the vibration response data of the vehicle suspension system.

[0064] It should be noted that modal coordinate transformation is used to convert the acceleration response in physical coordinates into modal displacement and modal velocity in modal coordinates, thereby eliminating rigid displacement interference in physical measurements.

[0065] S103. Vibration sensor arrays and environmental temperature and humidity sensors are deployed in key parts of the vehicle base's throat area, maintenance depot, and main structure of the office building. Dynamic characteristic parameters of the foundation soil are obtained through on-site wave velocity testing and indoor resonant column tests. Modal frequency data of the building structure are obtained through hammer impact testing or environmental vibration testing. Time series data of vehicle operation conditions are extracted from the vehicle operation management system. All data are timestamped and unified to the same time base using linear interpolation to generate a multi-physics field associated monitoring dataset for the vehicle base. The multi-physics field associated monitoring dataset for the vehicle base is used to provide boundary conditions for the coupling model parameters.

[0066] It should be noted that a multi-point monitoring network was constructed on the critical path of vibration transmission and on sensitive buildings. The soil dynamic parameters obtained by combining on-site wave velocity testing and resonant column testing truly reflect the nonlinear dynamic characteristics and damping ratio of the foundation soil layer. The hammer impact method and the environmental vibration method complement each other to achieve high-precision identification of the building structure modal parameters. The time series data of operating conditions were directly extracted from the vehicle operation management system to ensure real-time synchronization of operating conditions and vibration source characteristics. Timestamp alignment and linear interpolation processing eliminated the sampling frequency differences of multi-source heterogeneous data. A multi-physics field correlation monitoring dataset under a unified spatiotemporal coordinate system was constructed, which provides complete, synchronous and verifiable boundary condition inputs for the four-body coupled dynamic model and greatly improves the physical consistency of model parameters.

[0067] S200: Based on the multi-physics field associated monitoring dataset of the vehicle base, a vehicle-track coupled dynamic model is constructed and the vehicle track subsystem matrix is ​​derived. At the same time, a foundation-building coupled dynamic model is established and the soil structure subsystem matrix is ​​derived. The wheel-rail contact nonlinear force function and the soil structure boundary impedance parameters are defined. The Lagrange multiplier method is used to assemble the vehicle track subsystem matrix, soil structure subsystem matrix, wheel-rail contact nonlinear force function, and soil structure boundary impedance parameters into a system-level assembly in a multi-physics field co-simulation environment, forming an overall coupled dynamic equation set. The variable step-size implicit integration algorithm is called to solve the overall coupled dynamic equation set and generate the coupled system state parameter matrix.

[0068] It should be noted that the multiphysics co-simulation environment in this application is a hybrid simulation architecture built on the open-source framework Simulink and a self-developed dynamic link library. The interface protocol is implemented using the S-Function module, and data format conversion is completed through a custom MEX file. Specifically, the vehicle-track subsystem matrix is ​​built in MATLAB / Simulink using the Simscape Multibody module. The vehicle body, frame, and wheelsets are defined as rigid bodies. The suspension force elements are simulated using a combination of nonlinear spring-damping-mass elements in SimscapeDriveline. The track beam elements and plate elements are generated into mass and stiffness matrices using PDEToolbox and then imported into Simulink's State-Space module. The foundation-building subsystem matrix is ​​calculated in a self-developed Fortran boundary element program, outputting the foundation impedance matrix and building modal matrix, which are then compiled into the dynamic link library BEM.dll through the MEX interface.

[0069] It is understood that step S200 includes S201, S202, and S203, wherein:

[0070] S201. By calling the excitation spectrum matrix and the modal space vibration state vector constructed by modal filter conversion of the vibration response data of the vehicle suspension system, a vehicle-track coupled dynamic model is constructed. The vehicle system is described by multi-rigid-body dynamics. The car body, frame, and wheelset are defined as rigid bodies, the suspension elements are defined as force elements, and the track structure is discretized into beam elements and plate elements using finite element method. The vehicle-track subsystem matrix is ​​then assembled to obtain the vehicle mass matrix, which includes the vehicle mass matrix, vehicle stiffness matrix, track mass matrix, and track stiffness matrix.

[0071] Understandably, the modal filter is constructed based on the measured frequency response function of the vehicle suspension system. Essentially, it solves for the modal participation factor using the least squares method, decoupling the acceleration responses of the axle box, frame, and body into a linear combination of rigid body modes and elastic modes. Since the rigid body buoyancy and pitching modes contribute very little to vibration noise but dominate signal energy during low-speed vehicle travel, direct modeling would obscure the crucial elastic vibration components. Therefore, this modal filter eliminates the first six rigid body modes through orthogonal projection, retaining only the seventh and higher suspension elastic mode displacements and modal velocities, forming a modal space vibration state vector. This vector serves as the dimensionality-reduced observation input to the state space, eliminating measurement noise interference and allowing subsequent modeling to focus on the degenerative evolution characteristics of the suspension parameters.

[0072] In the multi-rigid-body dynamics description of the vehicle system, the car body, frame, and wheelsets are defined as rigid bodies, while the suspension elements are defined as nonlinear force elements. For the three-part bogie commonly found in vehicle depots, each side's suspension force element includes four force element nodes: primary steel springs, primary vertical dampers, secondary air springs, and secondary lateral dampers. The steel spring force element uses a nonlinear piecewise stiffness function, considering the dead zone of the air spring height valve adjustment and the hardening effect of the rubber stack; the damper force element uses the Maxwell model, and the damping coefficient degradation with service mileage is calibrated using force-velocity hysteresis curves obtained from indoor bench tests. The mass matrix and stiffness matrix formed by the coupling of rigid and elastic bodies are arranged diagonally in blocks, facilitating subsequent interface coupling with the soil structure subsystem matrix. This subsystem matrix not only preserves the time-varying degradation characteristics of the vehicle suspension parameters but also fully includes the high-frequency vibration modes of the track structure, providing an accurate frequency domain transfer function basis for four-body coupling.

[0073] S202. By calling the dynamic characteristic parameters of the foundation soil and the modal frequency data of the building structure obtained from the multi-physics field correlation monitoring data of the vehicle base, a coupled dynamic model of foundation-building is established. The foundation is established as a half-space model using the boundary element method, and the building is described using the modal superposition method. The boundary impedance parameters of the soil structure are obtained by inversion of the measured frequency response function. The boundary impedance parameters of the soil structure characterize the energy transfer characteristics of the vibration wave at the junction of the foundation and the building. The soil structure subsystem matrix is ​​derived, which includes the foundation impedance matrix and the building modal mass matrix.

[0074] It should be noted that accelerometers are placed at the foundation-building interface, and a hammer is used to excite the top of the foundation to measure the acceleration transfer function between the foundation and the soil. The elements of the boundary impedance matrix are solved using the frequency domain least squares method. This impedance matrix is ​​a full-size matrix, but its dimension is only equal to the number of foundation nodes. It characterizes the energy transfer characteristics of vibration waves scattered from the building foundation to the ground. Its imaginary part reflects the ground radiation damping, and its real part reflects the ground stiffness. This inversion process fully considers the building foundation type (such as pile foundation, raft foundation) and soil stratification conditions. The derived soil structure subsystem matrix accurately reflects the dynamic interaction of the soil-building coupling interface, providing real boundary connection parameters for the assembly of the four-body system.

[0075] S203. The vehicle track subsystem matrix and the soil structure subsystem matrix are connected through a wheel-rail contact nonlinear force function. The wheel-rail contact nonlinear force function is constructed using Hertz contact theory and Kalker creep theory. Constraints are applied at the interface nodes using the Lagrange multiplier method, and the overall coupled dynamic equations are assembled. The calculation formula is as follows:

[0076]

[0077] In the formula, Represents the total mass matrix. Represents the generalized acceleration vector. Represents the total damping matrix. Represents the generalized velocity vector. Represents the overall stiffness matrix. Represents the generalized displacement vector. This represents the nonlinear force function of wheel-rail contact. The boundary impedance parameters of the soil structure are represented; the variable step size implicit integration algorithm is called to solve the overall coupled dynamic equations to generate the coupled system state parameter matrix, which includes the system displacement, velocity and acceleration response at each time step.

[0078] It should be noted that the wheel-rail contact is described using Hertzian contact theory to represent the normal force. The contact stiffness is related to the wheel-rail curvature radius and material parameters, while the semi-axis length of the contact patch dynamically changes with the wheel weight, enabling nonlinear simulation of wheel-rail separation and contact impact. The tangential creep force is described using Kalker's simplified theory. The creep coefficient is related to the wheel-rail material, contact patch size, and creep rate. This function can be linearized under low-speed conditions, but its complete nonlinearity must be retained when the vehicle passes through curves or switch areas. The interface constraint uses the Lagrange multiplier method. Displacement compatibility equations are set at the interface between the vehicle track subsystem matrix and the soil structure subsystem matrix. The introduction of the multiplier vector increases the scale of the system equations but maintains symmetry, avoiding the numerical stiffness problem introduced by the penalty function method. The resulting coupled system state parameter matrix contains the system displacement, velocity, and acceleration responses at each time step. Its dimensions match the measurement point layout, providing high-precision, high-frequency sampled state input for subsequent energy field analysis, and fully capturing the transient dynamic response characteristics of the four-body coupled system under time-varying conditions.

[0079] Furthermore, to address the convergence difficulties caused by the strong nonlinearity of wheel-rail contact, this application adopts a staged linearization strategy: the contact state judgment stage uses an event-driven method, determining contact when the normal gap is less than zero, and when it is greater than zero and less than zero... Determined to be a critical state, greater than The system is identified as disengaged. During the contact force calculation phase, a quasi-Newton iteration is used. If convergence is not achieved after 15 iterations, it automatically switches to a line search backtracking strategy, with the step size factor decreasing by 0.5 times, for a maximum of 5 backtracking iterations. Furthermore, in the turnout section where the wheel-rail contact nonlinearity is most severe, static equilibrium pre-calculation is performed before simulation, setting the initial displacement as a static equilibrium solution to avoid iterative oscillations caused by mismatch between initial conditions and contact state in the early stages of dynamic calculations. For the complex stiffness matrix introduced by the soil boundary impedance parameters, a complex operation separation technique is used, storing the real and imaginary parts as two separate real matrices. During the solution process, the augmented real matrix technique avoids compatibility issues of complex operations in Simulink. Every 100 steps during integration, the energy conservation error is checked. If the mechanical energy drift exceeds 5%, a step size reset and Jacobian reconstruction are triggered to ensure long-term simulation stability. The above parameters and strategies, verified by actual test data at the vehicle base, reduced the simulation time from 4 hours with a fixed step size to 0.5 hours with a variable step size, and increased the nonlinear iteration convergence rate from 73% to 98%, proving the engineering applicability of the variable step size implicit integration algorithm in the strongly nonlinear coupled problem of the vehicle base.

[0080] It should be noted that the Lagrange multiplier method is implemented as follows: The AlgebraicConstraint module is introduced into Simulink, using the interface node displacement compatibility equations between the vehicle track subsystem and the soil structure subsystem as algebraic constraints. The module automatically introduces the Lagrange multiplier λ and generates an augmented system of equations. Specifically, the constraint settings are as follows: three normal displacement compatibility constraints and two tangential creep displacement compatibility constraints are defined at the wheel-rail contact point; vertical displacement compatibility constraints are defined at the contact surface between the track foundation and the subgrade. The Jacobian matrix sparse mode is stored as a static array by pre-marking the non-zero element positions, avoiding simulation interruptions caused by dynamic memory allocation. Data format conversion uses memory-mapped file technology. The vehicle track subsystem matrix is ​​stored in shared memory in CSR format, and the soil structure subsystem matrix is ​​stored in COO format. During coupling, it is directly read through memory pointers, eliminating file I / O bottlenecks. The simulation environment is configured as a 64-bit Windows system, using the Intel MKL library for matrix operations acceleration. The single-step solution time is controlled within 0.5 seconds, meeting the real-time prediction requirements.

[0081] S300: Input the coupled system state parameter matrix into the preset energy field analysis module to extract the system vibration energy distribution characteristics and output vibration energy density field data; simultaneously, import the environmental temperature and humidity monitoring data into the acoustic propagation calculation engine to construct a refraction propagation model of sound waves in the temperature and humidity gradient field and generate sound ray bending trajectory data; load the vehicle operating condition time series data into the wave equation solver to establish a dispersion characteristic model of vibration waves in the foundation soil and obtain frequency-wavenumber domain response data; then use the finite element-boundary element coupled algorithm to perform spatiotemporal domain dual discretization integration on the vibration energy density field data, sound ray bending trajectory data and frequency-wavenumber domain response data to output vibration and noise propagation characteristic field distribution data.

[0082] It is understood that step S300 includes S301, S302, and S303, wherein:

[0083] S301. Call the state parameter matrix of the coupled system, extract the velocity vector and stress vector of each unit node step by step, calculate the kinetic energy density and strain energy density for each unit, and accumulate the total energy density field of the system after traversing all units. Then, grid the total energy density field according to spatial coordinates to generate vibration energy density field data.

[0084] It should be noted that, due to the multi-material layers of the vehicle depot track structure, including rails, track bed, and foundation, the density and elastic modulus of each element need to be assigned separately. After traversing all elements, the total energy density field of the system is obtained by weighted summation of element volumes, which physically represents the spatial distribution intensity of vibration energy. Finally, the total energy density field is meshed according to spatial coordinates and matched with the subsequent acoustic mesh to generate vibration energy density field data. This data not only reflects the intensity of the vibration source, but more importantly, its spatiotemporal gradient reveals the energy flow path, providing a non-homogeneous excitation source for acoustic radiation calculation.

[0085] S302. Call the environmental temperature and humidity monitoring data, calculate the sound velocity field distribution according to the time series, establish the temperature and humidity gradient field in the sound propagation calculation domain, use the ray tracing method to solve the refraction path of the sound wave in the gradient field, trace the sound ray path from the track structure through the soil to the building structure, and generate sound ray bending trajectory data.

[0086] It should be noted that the calculation of the sound velocity field distribution must be based on measured environmental temperature and humidity monitoring data, and a time-series segmentation processing strategy is adopted to divide the diurnal temperature and humidity variation curve into multiple time periods. The temperature and humidity gradient field is established using bilinear interpolation to form a three-dimensional sound velocity gradient field within the sound propagation calculation domain, which covers the entire propagation path from track to soil to building. When solving using the ray tracing method, a sound ray beam is emitted from the wheel-rail contact patch, with the initial direction determined according to the direction of the vibration energy density field gradient. The ray step size is adaptively adjusted: the step size is reduced to 0.1m in areas with severe temperature and humidity gradients, and increased to 1m in uniform areas. When tracing the sound ray path, the refraction angle of the ray at the soil layer interface needs to be calculated to satisfy Snell's law, while soil absorption attenuation is considered. The attenuation coefficient is determined by the damping ratio in the soil dynamic characteristic parameters. The generated sound ray bending trajectory data is stored in the form of a vector field, recording the spatial coordinates and energy attenuation rate of each ray at each time step. This data truly reflects the propagation delay and path distortion of sound waves in the non-uniform temperature and humidity field of the vehicle base, overcoming the error of the traditional assumption of straight-line sound ray propagation.

[0087] S303. Using vehicle operating condition time-series data, time-varying boundary conditions are set in the wave equation solver to establish a dispersion characteristic model of vibration waves in the foundation soil. Frequency-wavenumber domain response data is obtained through wavenumber space scanning. The frequency-wavenumber domain response data describes the propagation speed and attenuation law of vibration waves of different frequencies in the soil. The finite element-boundary element coupled algorithm is used to perform spatiotemporal dual discretization integration on the vibration energy density field data, sound ray bending trajectory data and frequency-wavenumber domain response data. The time discretization adopts the central difference scheme and the spatial discretization adopts the Gaussian integral scheme. After the integration operation, the vibration noise propagation characteristic field distribution data is output.

[0088] The formula for calculating kinetic energy density is as follows:

[0089]

[0090] In the formula, Kinetic energy density, For unit material density, The vibration velocity of a single particle;

[0091] The formula for calculating strain energy density is as follows:

[0092]

[0093] In the formula, For strain energy density, For element stress, For unit strain.

[0094] It should be noted that the vehicle operation time-series data includes train entry and exit times, travel speed, and formation. In the wave equation solver, moving load boundary conditions are set, and the load time history is obtained from the wheel-rail contact force inversion. The spatial distribution moves longitudinally along the track. In the finite element-boundary element coupled algorithm, the near-field track-soil region is discretized using finite element methods to ensure nonlinear contact accuracy, while the far-field infinite domain uses boundary element methods to avoid reflection. The coupling interface is matched by the boundary element kernel function and the finite element shape function integral. The time discretization of the spatiotemporal dual discretization integration operation uses a central difference scheme, with the time step determined according to the CFL condition. The spatial discretization uses a Gaussian integral scheme, with each spectral element configured with 6×6×6 Gaussian integration points. During integration, vibration energy density field data is used as the source term, acoustic ray bending trajectory data as path weighting, and frequency-wavenumber domain response data as the frequency domain transfer function. After three-dimensional convolution integration, the vibration and noise propagation characteristic field distribution data is output. This data is a four-dimensional tensor (time × space × frequency × noise level), which fully describes the spatiotemporal evolution of noise throughout the entire vehicle base area.

[0095] S400. The vibration and noise propagation feature field distribution data is imported into a spatiotemporal attention mechanism deep learning prediction framework to build a network topology to form a prediction model. Historical field distribution data is input into the input layer of the prediction model for data standardization mapping to generate standardized feature vectors. The standardized feature vectors are encoded with temporal dependencies using a long short-term memory network layer to extract hidden state feature sequences. The hidden state feature sequences are mapped to future vibration and noise prediction values ​​through a fully connected layer. The backpropagation algorithm is executed to train the network parameters to a convergent state. After convergence, the network outputs the future vibration and noise prediction data sequence.

[0096] It is understood that in this step, S400 includes S401, S402, and S403, wherein:

[0097] S401. Import the vibration and noise propagation feature field distribution data into the spatiotemporal attention mechanism deep learning prediction framework, and construct a neural network topology containing a temporal attention module and a spatial attention module. The temporal attention module generates attention weights by calculating the dot product similarity between the query vector and the key vector, and the spatial attention module extracts the spatial features of the vibration and noise field through convolutional layers and generates a spatial weight matrix. The temporal attention weights and spatial attention weights are multiplied element-wise to obtain the spatiotemporal joint attention weights. This weight is applied to the value vector weighted summation to form the attention output features. After stacking multiple attention mechanisms, the network is connected to a batch normalization layer and an activation function layer. The completed network topology is used to receive standardized feature vectors.

[0098] It should be noted that the temporal attention module maps the latent state features of historical time periods to query vectors and key vectors, calculates the dot product similarity between the two, and generates temporal attention weights after normalization. Each row of this weight matrix represents the attention distribution of each historical moment to the current prediction moment. The spatial attention module adopts a lightweight convolutional structure. After inputting the two-dimensional spatial distribution of the vibration and noise field, it extracts local spatial correlations through depthwise separable convolution, and then generates spatial attention weights through dimensionality reduction convolution. The convolutional kernel automatically learns the coupling strength of vibration and noise transmission in different regions. For example, when multiple tracks are arranged in parallel in the maintenance depot, the vibration of one track will propagate to the adjacent track through the soil, and the spatial attention weight is assigned a higher value at the corresponding position; the office building is far from the throat area, and the transmission path is severely attenuated through the soil, so the weight is lower at the corresponding position. The spatiotemporal joint attention weight is obtained by the outer product operation of the temporal weight and the spatial weight. After dimensional expansion, it is applied to the weighted summation of the value vector to form the attention output feature. When stacking multiple attention mechanisms, residual connections and normalization are used. Each layer is followed by a batch normalization layer to stabilize the training process, and the activation function layer selects a function that takes into account nonlinear smoothness. The completed network topology enables end-to-end modeling of the spatiotemporal coupling characteristics of the vibration and noise field. The input is the vibration and noise characteristics of spatial measurement points in historical time periods, and the output is the prediction results for future time periods. No manual design of spatiotemporal feature engineering is required, which significantly improves the adaptability and generalization ability of the model in complex scenarios of vehicle bases.

[0099] S402. Input the historical field distribution data into the model input layer, perform Z-score standardization mapping, independently calculate the mean and standard deviation of the time series data for each spatial measurement point, and generate a standardized feature vector. The formula for calculating the standardized feature vector is as follows:

[0100]

[0101] In the formula, This represents the standardized data. This represents the original vibration and noise field data. This represents the sample mean of the time series data at this measurement point. This represents the sample standard deviation of the time series data at this measurement point.

[0102] S403. Input the standardized feature vector into the Long Short-Term Memory (LSTM) network layer for temporal dependency encoding. Set the number of hidden units and layers in the LSTM. Use a bidirectional LSTM structure to extract temporal features from both the forward and backward directions. Concatenate the bidirectional hidden state vectors to obtain the hidden state feature sequence. The calculation formula for the LSTM network state update is as follows:

[0103]

[0104] In the formula, This represents the hidden state vector at the current time. This represents the standardized feature vector input at the current time. This represents the hidden state vector from the previous time step.

[0105] A fully connected layer is set to map the hidden state feature sequence to the noise prediction value of the future time period. The output dimension of the fully connected layer is set to the product of the number of prediction time periods and the number of measurement points. The mean square error loss function is used to calculate the error between the prediction value and the true value. The backpropagation algorithm is executed to train the network parameters to the convergence state. The convergence condition is set to the validation set loss no longer decreasing after several consecutive rounds. After the network has converged, it outputs the noise prediction data sequence of the future time period.

[0106] It should be noted that the fully connected layer adopts a hierarchical structure to reduce the number of parameters: the first layer maps the hidden state feature sequence to an intermediate feature space, and the second layer maps to the final prediction space. The prediction period is typically the next 12 hours, and the number of measurement points is the total number of vibration and noise measurement points within the vehicle base. This hierarchical mapping avoids the parameter explosion of a single-layer structure, while introducing intermediate nonlinearity to enhance the model's expressive power, making it suitable for the characteristics of a large number of measurement points and a long prediction period in vehicle bases. The convergence condition is set to ensure that the validation set loss no longer decreases for several consecutive rounds and that the absolute decrease is below a threshold. Simultaneously, the difference between the training and validation set losses is monitored to prevent overfitting. An early stopping mechanism immediately terminates training and saves the optimal model when the validation set loss increases. After training, the model focuses on evaluating the mid-frequency prediction error on the test set. These frequency bands correspond to the natural frequencies of the vehicle suspension and high-frequency vibrations of the track, and are the main contributing frequency bands to excessive vibration and noise at the vehicle base. When running the converged network for rolling prediction, a sliding window strategy is used to periodically update historical data, keeping the input sequence length constant to achieve continuous prediction of future periods. The final output of the predicted data sequence includes the predicted mean and confidence interval, providing a basis for risk quantification for scheduling optimization. Decision-makers can assess the prediction uncertainty based on the confidence interval width and prioritize adjusting the operation plan when the risk of excessive noise is high and the confidence level is high, thereby effectively transforming the prediction results into an executable control strategy.

[0107] S500: Based on the noise prediction data sequence for future time periods, design a chromosome encoding structure for vehicle scheduling schemes to generate a candidate solution population; define a fitness function with the dual objectives of minimizing noise overscalar and maximizing operational efficiency, and use the fitness function to evaluate individual candidates in the candidate solution population; initialize the candidate solution population after individual evaluation as the initial solution set, and perform selection, crossover, and mutation genetic operations on the initial solution set to drive the population to iteratively evolve until the convergence condition is met to obtain the Pareto optimal front solution set; decode the optimal individual in the Pareto optimal front solution set and output the real-time optimization scheme for vehicle base operation and maintenance scheduling.

[0108] It is understood that in this step, S500 includes S501, S502, and S503, wherein:

[0109] S501. Based on the noise prediction data sequence of future time periods, design the chromosome coding structure of the vehicle scheduling scheme. The chromosome gene positions are encoded with real numbers and correspond to the train's entry and exit times and running speed levels on the designated track. A candidate solution population is generated through random initialization.

[0110] It should be noted that the chromosome gene positions are encoded using real numbers. Each gene position corresponds to the train's entry and exit times and operating speed level on a specific track. Each gene position is represented by a 64-bit double-precision floating-point number. The first half of the gene position encodes the starting time of the train's track occupancy (accurate to the minute), and the second half encodes the operating speed level (discretized into three levels: low speed 5km / h, medium speed 10km / h, and high speed 15km / h). The real value range is normalized to [0,1] and then mapped to the actual time and speed level. This encoding method effectively avoids the Hamming cliff problem of binary encoding, enabling crossover and mutation operations to perform a smooth search in continuous space while maintaining computational efficiency in the decoding process.

[0111] Understandably, in this step, when randomly initializing and generating the candidate solution population, the hard constraints of vehicle depot operation must be considered: First, track capacity constraint, at the same time, a maximum of two trains can be accommodated on the same track (head-to-tail docking), and a gene position conflict detection and repair mechanism is used during initialization to ensure a feasible solution; Second, maintenance operation time window constraint, the maintenance time of each train is determined by the maintenance schedule in the vehicle operation management system, and the start and end time difference of the gene position encoding must meet the minimum maintenance time during initialization; Third, throat area route conflict constraint, the entry paths of adjacent tracks have spatial intersections, and the route occupancy time interval needs to be calculated during initialization to avoid route conflicts.

[0112] S502. Define a dual-objective fitness function that includes minimizing the noise superscalar and maximizing operational efficiency, and initialize the candidate solution population to form the initial solution set based on this function;

[0113] It should be noted that the initial solution set is generated from the candidate solution population after evaluation using a fitness function. The evaluation process requires calculating the f1 and f2 values ​​for each individual. When calculating f1, the future time period noise prediction data sequence output in step S401 is called, and the predicted value is coupled with the train operation plan in the scheduling scheme to simulate the noise distribution under different scheduling conditions. When calculating f2, chromosome gene loci are analyzed, the number of trains under daily maintenance is counted, and the efficiency coefficient model is called. The size of the initial solution set is consistent with the population size, and all individuals and their fitness values ​​are retained to provide a parent gene pool for subsequent genetic operations.

[0114] S503. The initial solution set is sequentially subjected to tournament selection, simulated binary crossover and polynomial mutation genetic operations to drive the population iterative evolution; when the convergence condition is met, the Pareto optimal frontier solution set is obtained, the optimal individual in the solution set is decoded and the real-time optimization scheme for vehicle base operation and maintenance scheduling is output.

[0115] It should be noted that during tournament selection, five individuals are randomly selected from the population each time to form a tournament group. Their Pareto dominance and non-dominance ranking are compared, and the individual with the highest ranking or the largest crowding distance is selected to enter the mating pool. This strategy ensures the inheritance of high-quality genes, maintains population diversity, and avoids premature convergence of roulette wheel selection.

[0116] In this step, a simulated binary crossover operation is applied to the parent individuals selected in the tournament, with a crossover probability set to 0.9. During crossover, the crossover point is randomly selected, and the real values ​​of the two parent gene loci are used to generate offspring gene values ​​according to the SBX formula, with a distribution exponent of 10. The generated offspring gene values ​​have a high probability of being close to the parent values, ensuring the inheritance of excellent solutions. Feasibility repair must be performed immediately after crossover: if the time encoded by the offspring gene causes a trajectory conflict, heuristic repair is used to move the gene locus value to the nearest feasible time. If a conflict still exists, it is regenerated. The polynomial mutation probability is set to 0.1, with a distribution exponent of 20. The mutation amplitude is small, mainly targeting local searches. During mutation, gene loci are randomly selected, and the real values ​​are perturbed according to a polynomial distribution. After perturbation, the time and velocity constraints are also checked; if the constraints are violated, resampling is performed until they are satisfied.

[0117] In other words, this step achieves automatic optimization of scheduling schemes through genetic evolution, deeply couples vibration and noise prediction with scheduling decisions, provides vehicle bases with a dynamic optimization tool that balances environmental protection and efficiency, and solves the problem that traditional manual scheduling cannot balance multiple objectives.

[0118] In summary, this application proposes a four-body coupled vehicle depot vibration and noise prediction and optimization method. By constructing a full-chain coupled dynamic model of vehicle-track-foundation-building, integrating multi-physics field associated monitoring data, and employing a spatiotemporal attention mechanism deep learning framework to achieve accurate temporal prediction of the vibration and noise field, a multi-objective genetic algorithm is designed based on the prediction results to actively optimize vehicle scheduling schemes. This method overcomes the limitations of traditional passive vibration isolation, shifting vibration and noise control from static management to dynamic prediction and intelligent operation and maintenance. It achieves early warning of vibration and noise exceeding standards and collaborative optimization of operation plans, providing a systematic solution for the green and intelligent operation and maintenance of vehicle depots.

[0119] Example 2:

[0120] like Figure 2 As shown, this embodiment provides a four-body coupled vehicle base vibration and noise prediction and optimization system. See [link to documentation]. Figure 2The system includes:

[0121] Acquisition Module 701: Used to acquire the multi-physics field associated monitoring dataset of the vehicle base, which includes vehicle base track dynamic irregularity detection data, vehicle suspension vibration response data, foundation soil dynamic characteristic parameters, building structure modal frequency data, environmental temperature and humidity monitoring data, and vehicle operating condition time series data.

[0122] Assembly and Generation Module 702: Based on the multi-physics field associated monitoring dataset of the vehicle base, this module constructs a vehicle-track coupled dynamics model and derives the vehicle track subsystem matrix. Simultaneously, it establishes a foundation-structure coupled dynamics model and derives the soil structure subsystem matrix, defining the wheel-rail contact nonlinear force function and soil structure boundary impedance parameters. Using the Lagrange multiplier method, the vehicle track subsystem matrix, soil structure subsystem matrix, wheel-rail contact nonlinear force function, and soil structure boundary impedance parameters are assembled hierarchically in a multi-physics field co-simulation environment to form an overall coupled dynamics equation set. A variable-step implicit integration algorithm is then called to solve the overall coupled dynamics equation set, generating the coupled system state parameter matrix.

[0123] Output module 703 is used to input the state parameter matrix of the coupled system into the preset energy field analysis module, extract the vibration energy distribution characteristics of the system, and output vibration energy density field data; at the same time, it imports environmental temperature and humidity monitoring data into the acoustic propagation calculation engine to construct a refraction and propagation model of sound waves in the temperature and humidity gradient field, and generates sound ray bending trajectory data; it also loads vehicle operating condition time series data into the wave equation solver to establish a dispersion characteristic model of vibration waves in the foundation soil, and obtains frequency-wavenumber domain response data; then, it uses a finite element-boundary element coupled algorithm to perform spatiotemporal domain dual discretization integration on the vibration energy density field data, sound ray bending trajectory data, and frequency-wavenumber domain response data, and outputs vibration and noise propagation characteristic field distribution data.

[0124] Training Module 704: This module is used to import vibration and noise propagation feature field distribution data into a spatiotemporal attention mechanism deep learning prediction framework, build a network topology to form a prediction model, input historical field distribution data into the input layer of the prediction model for data standardization mapping, and generate standardized feature vectors. A Long Short-Term Memory (LSTM) network layer is used to encode the temporal dependencies of the standardized feature vectors, extract the hidden state feature sequence, and map the hidden state feature sequence to future vibration and noise prediction values ​​through a fully connected layer. The backpropagation algorithm is executed to train the network parameters until convergence. After convergence, the network outputs a future vibration and noise prediction data sequence.

[0125] Iteration Module 705: Based on the noise prediction data sequence for future time periods, it designs the chromosome encoding structure of the vehicle scheduling scheme to generate a candidate solution population; defines a fitness function with the dual objectives of minimizing noise overscalar and maximizing operational efficiency, and uses the fitness function to evaluate individual candidates; initializes the candidate solution population after individual evaluation as the initial solution set, performs selection, crossover, and mutation genetic operations on the initial solution set to drive the population to iterative evolution, and obtains the Pareto optimal front solution set when the convergence condition is met. It decodes the optimal individual in the Pareto optimal front solution set and outputs the real-time optimization scheme for vehicle base operation and maintenance scheduling.

[0126] Specifically, the acquisition module 701 includes:

[0127] The first acquisition unit is used to deploy track inspection vehicles and laser displacement sensor arrays in the track section of the vehicle base. It performs spatial positioning based on the kilometer marker, collects track gauge, elevation, and directional irregularity data, and simultaneously performs millimeter-wave radar scanning on the rail surface corrugation to obtain corrugation depth spectrum data. The track gauge, elevation, directional irregularity data and corrugation depth spectrum data are uniformly stored as a spatiotemporal two-dimensional array structure. The time dimension corresponds to the inspection date, and the spatial dimension corresponds to the line mileage. It is then converted into an excitation spectrum matrix, which constitutes the external excitation input of the vehicle-track coupling model.

[0128] The second acquisition unit is used to install inertial measurement units at three locations: the bogie axle box, frame, and car body of the monitored vehicle. It collects the vertical, lateral, and longitudinal vibration acceleration time history data of the vehicle when it runs at low speed on the track in the depot. The acquired vibration acceleration time history data is zero drift eliminated and trend term removed. The dynamic displacement and dynamic velocity of the suspension elements are obtained through integral calculation to form the vibration response data of the vehicle suspension system.

[0129] Extraction Unit: This unit deploys vibration sensor arrays and environmental temperature and humidity sensors in key areas of the vehicle depot's throat zone, maintenance depot, and main office building structure. It acquires dynamic characteristics of the foundation soil through on-site wave velocity testing and indoor resonant column tests, obtains modal frequency data of the building structure through hammer impact or environmental vibration testing, and extracts vehicle operating condition time-series data from the vehicle operation management system. All data is timestamped and unified to the same time base using linear interpolation to generate a multi-physics field correlation monitoring dataset for the vehicle depot. This dataset provides boundary conditions for the coupling model parameters.

[0130] Specifically, the assembly generation module 702 includes:

[0131] The first calling unit is used to call the excitation spectrum matrix and the modal space vibration state vector constructed by modal filter conversion of the vibration response data of the vehicle suspension system, and to construct the vehicle-track coupled dynamic model. The vehicle system is described by multi-rigid-body dynamics. The car body, frame, and wheelset are defined as rigid bodies, the suspension elements are defined as force elements, and the track structure is discretized into beam elements and plate elements by finite element method. The vehicle track subsystem matrix is ​​obtained by assembly. The vehicle track subsystem matrix includes the vehicle mass matrix, the vehicle stiffness matrix, the track mass matrix, and the track stiffness matrix.

[0132] The second calling unit is used to call the building structure modal frequency data obtained from the dynamic characteristic parameters of the foundation soil and the multi-physics field correlation monitoring data of the vehicle base, and to establish a foundation-building coupled dynamic model. The foundation is established using the boundary element method to build a half-space model, and the building is described using the modal superposition method. The boundary impedance parameters of the soil structure are obtained by inversion of the measured frequency response function. The boundary impedance parameters of the soil structure characterize the energy transfer characteristics of the vibration wave at the junction of the foundation and the building. The soil structure subsystem matrix is ​​derived, which includes the foundation impedance matrix and the building modal mass matrix.

[0133] Equation unit construction: The vehicle track subsystem matrix and the soil structure subsystem matrix are connected through wheel-rail contact nonlinear force functions. These nonlinear force functions are constructed using Hertzian contact theory and Kalker creep theory. Constraints are applied at interface nodes using the Lagrange multiplier method, assembling to form a global coupled dynamic equation system. The calculation formulas are as follows:

[0134]

[0135] In the formula, Represents the total mass matrix. Represents the generalized acceleration vector. Represents the total damping matrix. Represents the generalized velocity vector. Represents the overall stiffness matrix. Represents the generalized displacement vector. This represents the nonlinear force function of wheel-rail contact. Indicates the boundary impedance parameters of the soil structure;

[0136] The third calling unit is used to call the variable step size implicit integration algorithm to solve the overall coupled dynamic equations and generate the coupled system state parameter matrix, which includes the system displacement, velocity and acceleration response at each time step.

[0137] Specifically, the output module 703 includes:

[0138] The fourth calling unit is used to call the state parameter matrix of the coupled system, extract the velocity vector and stress vector of each unit node step by step, calculate the kinetic energy density and strain energy density for each unit, and accumulate the total energy density field of the system after traversing all units. The total energy density field is then gridded according to spatial coordinates to generate vibration energy density field data.

[0139] The fifth calling unit is used to call environmental temperature and humidity monitoring data, calculate the sound velocity field distribution according to the time series, establish a temperature and humidity gradient field in the sound propagation calculation domain, use the ray tracing method to solve the refraction path of the sound wave in the gradient field, trace the sound ray path from the track structure through the soil to the building structure, and generate sound ray bending trajectory data.

[0140] The computational unit utilizes vehicle operating condition time-series data to set time-varying boundary conditions in the wave equation solver, establishing a dispersion characteristic model of vibration waves in the foundation soil. It obtains frequency-wavenumber domain response data through wavenumber space scanning, which describes the propagation speed and attenuation of vibration waves of different frequencies in the soil. A finite element-boundary element coupled algorithm is used to perform spatiotemporal dual discretization integration on the vibration energy density field data, acoustic ray bending trajectory data, and frequency-wavenumber domain response data. Time discretization employs a central difference scheme, and spatial discretization uses a Gaussian integral scheme. After integration, the noise propagation characteristic field distribution data is output.

[0141] The formula for calculating kinetic energy density is as follows:

[0142]

[0143] In the formula, Kinetic energy density, For unit material density, The vibration velocity of a single particle;

[0144] The formula for calculating strain energy density is as follows:

[0145]

[0146] In the formula, For strain energy density, For element stress, For unit strain.

[0147] Specifically, the training module 704 includes:

[0148] The summation unit is used to import the vibration and noise propagation feature field distribution data into the spatiotemporal attention mechanism deep learning prediction framework, and construct a neural network topology containing a temporal attention module and a spatial attention module. The temporal attention module generates attention weights by calculating the dot product similarity between the query vector and the key vector, and the spatial attention module extracts the spatial features of the vibration and noise field through convolutional layers and generates a spatial weight matrix. The temporal attention weights and spatial attention weights are multiplied element-wise to obtain the spatiotemporal joint attention weights. This weight is applied to the value vector for weighted summation to form the attention output features. After stacking multiple attention mechanisms, the network is connected to a batch normalization layer and an activation function layer. The completed network topology is used to receive standardized feature vectors.

[0149] The mapping unit is used to input historical field distribution data into the model's input layer, perform Z-score standardization mapping, independently calculate the mean and standard deviation of the time series data for each spatial measurement point, and generate standardized feature vectors. The formula for calculating the standardized feature vectors is as follows:

[0150]

[0151] In the formula, This represents the standardized data. This represents the original vibration and noise field data. This represents the sample mean of the time series data at this measurement point. This represents the sample standard deviation of the time series data at this measurement point.

[0152] Sequence units are set up: These are used to input standardized feature vectors into the Long Short-Term Memory (LSTM) network layers for temporal dependency encoding. The number of LSTM hidden layer units and the number of layers are set. A bidirectional LSTM structure is used to extract temporal features from both the forward and backward directions. The bidirectional hidden state vectors are concatenated to obtain the hidden state feature sequence. The calculation formula for the LSTM network state update is as follows:

[0153]

[0154] In the formula, This represents the hidden state vector at the current time. This represents the standardized feature vector input at the current time. This represents the hidden state vector from the previous time step.

[0155] The execution unit is set to map the hidden state feature sequence to the noise prediction value of the future time period in the fully connected layer. The output dimension of the fully connected layer is set to the product of the number of prediction time periods and the number of measurement points. The mean square error loss function is used to calculate the error between the prediction value and the true value. The backpropagation algorithm is executed to train the network parameters to the convergence state. The convergence condition is set to the validation set loss no longer decreasing after several consecutive rounds. After the network converges, it outputs the noise prediction data sequence of the future time period.

[0156] Specifically, the iteration module 705 includes:

[0157] Design Unit: Used to design the chromosome coding structure of the vehicle scheduling scheme based on the noise prediction data sequence of future time periods. The chromosome gene positions are encoded with real numbers and correspond to the train's entry and exit times and running speed levels on the specified track. A candidate solution population is generated through random initialization.

[0158] Solving unit: used to define a dual-objective fitness function that includes minimizing oscillation hyperscalar and maximizing operational efficiency, and to initialize the candidate solution population to form the initial solution set based on this function;

[0159] Decoding output unit: used to sequentially perform tournament selection, simulated binary crossover and polynomial mutation genetic operations on the initial solution set to drive the population iterative evolution; when the convergence condition is met, the Pareto optimal frontier solution set is obtained, the optimal individual in the solution set is decoded and the real-time optimization scheme of vehicle base operation and maintenance scheduling is output.

[0160] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0161] Example 3:

[0162] Corresponding to the above method embodiments, this embodiment also provides a four-body coupled vehicle base vibration and noise prediction optimization device. The four-body coupled vehicle base vibration and noise prediction optimization device described below and the four-body coupled vehicle base vibration and noise prediction optimization method described above can be referred to in correspondence.

[0163] Figure 3 This is a block diagram illustrating a four-body coupled vehicle base vibration noise prediction and optimization device 800 according to an exemplary embodiment. Figure 3 As shown, the four-body coupled vehicle base vibration and noise prediction optimization device 800 includes a processor 801 and a memory 802. The four-body coupled vehicle base vibration and noise prediction optimization device 800 also includes one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0164] The processor 801 controls the overall operation of the four-body coupled vehicle base vibration and noise prediction optimization device 800 to complete all or part of the steps in the aforementioned four-body coupled vehicle base vibration and noise prediction optimization method. The memory 802 stores various types of data to support the operation of the four-body coupled vehicle base vibration and noise prediction optimization device 800. This data may include, for example, instructions for any application or method operating on the four-body coupled vehicle base vibration and noise prediction optimization device 800, as well as application-related data, such as contact data, sent and received messages, images, audio, video, etc. The memory 802 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 802 or transmitted via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as a keyboard, mouse, or buttons. These buttons can be virtual or physical. Communication component 805 is used for wired or wireless communication between the four-body coupled vehicle base vibration prediction and optimization device 800 and other devices. Wireless communication includes, for example, Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, or an NFC module.

[0165] In an exemplary embodiment, the four-body coupled vehicle base vibration and noise prediction optimization device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described four-body coupled vehicle base vibration and noise prediction optimization method.

[0166] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the four-body coupled vehicle base noise prediction optimization method described above. For example, the computer-readable storage medium may be the memory 802 including program instructions, which may be executed by the processor 801 of the four-body coupled vehicle base noise prediction optimization device 800 to complete the four-body coupled vehicle base noise prediction optimization method described above.

[0167] Example 4:

[0168] Corresponding to the above method embodiments, this embodiment also provides a readable storage medium. The readable storage medium described below can be referred to in relation to the four-body coupled vehicle base vibration noise prediction optimization method described above.

[0169] A computer program is stored on a readable storage medium, and when the computer program is executed by a processor, it implements the steps of the four-body coupled vehicle base vibration noise prediction optimization method of the above method embodiment.

[0170] Specifically, the readable storage medium can be a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other readable storage medium capable of storing program code.

[0171] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for predicting and optimizing vibration and noise at a four-body coupled vehicle base, characterized in that, include: Acquire a multi-physics field correlation monitoring dataset for the vehicle base, which includes dynamic irregularity detection data of the vehicle base track, vibration response data of the vehicle suspension system, dynamic characteristic parameters of the foundation soil, modal frequency data of the building structure, environmental temperature and humidity monitoring data, and time series data of vehicle operating conditions. Based on the multiphysics field associated monitoring dataset of the vehicle base, a vehicle-track coupled dynamic model is constructed and the vehicle track subsystem matrix is ​​derived. At the same time, a foundation-building coupled dynamic model is established and the soil structure subsystem matrix is ​​derived. The wheel-rail contact nonlinear force function and the soil structure boundary impedance parameters are defined. The Lagrange multiplier method is used to assemble the vehicle track subsystem matrix, the soil structure subsystem matrix, the wheel-rail contact nonlinear force function, and the soil structure boundary impedance parameters into a system-level assembly in a multiphysics field co-simulation environment, forming an overall coupled dynamic equation set. The variable step-size implicit integration algorithm is called to solve the overall coupled dynamic equations and generate the state parameter matrix of the coupled system. The coupled system state parameter matrix is ​​input into a preset energy field analysis module to extract the system vibration energy distribution characteristics and output vibration energy density field data. Simultaneously, environmental temperature and humidity monitoring data is imported into the acoustic propagation calculation engine to construct a refraction and propagation model of sound waves in a temperature and humidity gradient field, generating sound ray bending trajectory data. Vehicle operating condition time series data is loaded into the wave equation solver to establish a dispersion characteristic model of vibration waves in the foundation soil, obtaining frequency-wavenumber domain response data. Then, a finite element-boundary element coupled algorithm is used to perform spatiotemporal dual discretization and integration operations on the vibration energy density field data, sound ray bending trajectory data, and frequency-wavenumber domain response data to output vibration and noise propagation characteristic field distribution data. The vibration and noise propagation feature field distribution data is imported into a spatiotemporal attention mechanism deep learning prediction framework to build a network topology to form a prediction model. Historical field distribution data is input into the input layer of the prediction model for data standardization mapping to generate standardized feature vectors. The standardized feature vectors are encoded with temporal dependencies using a long short-term memory network layer to extract the hidden state feature sequence. The hidden state feature sequence is then mapped to the vibration and noise prediction values ​​for future periods through a fully connected layer. The backpropagation algorithm is executed to train the network parameters until convergence. After convergence, the network outputs the vibration and noise prediction data sequence for future periods. Based on the noise prediction data sequence for future time periods, a chromosome encoding structure for vehicle scheduling schemes is designed to generate a candidate solution population. A fitness function is defined with the dual objectives of minimizing noise overscalar and maximizing operational efficiency, and the fitness function is used to evaluate individual candidates in the candidate solution population. The candidate solution population after individual evaluation is initialized as the initial solution set. Selection, crossover, and mutation genetic operations are performed on the initial solution set to drive the population to iterative evolution. When the convergence condition is met, the Pareto optimal front solution set is obtained. The optimal individual in the Pareto optimal front solution set is decoded, and the real-time optimization scheme for vehicle base operation and maintenance scheduling is output.

2. The four-body coupled vehicle base vibration noise prediction and optimization method according to claim 1, characterized in that, The acquired multi-physics field correlation monitoring dataset for the vehicle base includes: A track inspection vehicle and a laser displacement sensor array are deployed in the track section of the vehicle depot. Spatial positioning is performed based on the kilometer marker, and data on track gauge, elevation, and directional irregularities are collected. At the same time, millimeter-wave radar is used to scan the corrugation on the rail surface to obtain corrugation depth spectrum data. The track gauge, elevation, and directional irregularity data and corrugation depth spectrum data are stored in a unified spatiotemporal two-dimensional array structure. The time dimension corresponds to the inspection date, and the spatial dimension corresponds to the line mileage. The data is then converted into an excitation spectrum matrix, which constitutes the external excitation input of the vehicle-track coupling model. Inertial measurement units were installed at three locations: the axle box, frame, and body of the monitored vehicle bogie. The vertical, lateral, and longitudinal vibration acceleration time history data of the vehicle were collected when it was running at low speed on the track in the depot. Zero drift elimination and trend term removal were performed on the collected vibration acceleration time history data. The dynamic displacement and dynamic velocity of the suspension components were obtained through integral calculation to form the vibration response data of the vehicle suspension system. Vibration sensor arrays and environmental temperature and humidity sensors were deployed in key areas of the vehicle base's throat area, maintenance depot, and main structure of the office building. Dynamic characteristic parameters of the foundation soil were obtained through on-site wave velocity testing and indoor resonant column tests. Modal frequency data of the building structure were obtained through hammer impact testing or environmental vibration testing. Time series data of vehicle operating conditions were extracted from the vehicle operation management system. All data were timestamped and unified to the same time base using linear interpolation to generate a multi-physics field correlation monitoring dataset for the vehicle base. This dataset is used to provide boundary conditions for the coupling model parameters.

3. The method for predicting and optimizing vibration and noise at a four-body coupled vehicle base according to claim 1, characterized in that, Based on the multiphysics field correlation monitoring dataset of the vehicle base, a vehicle-track coupled dynamics model is constructed and the vehicle track subsystem matrix is ​​derived. At the same time, a foundation-building coupled dynamics model is established and the soil structure subsystem matrix is ​​derived. The wheel-rail contact nonlinear force function and the soil structure boundary impedance parameters are defined. The Lagrange multiplier method is used to assemble the vehicle track subsystem matrix, soil structure subsystem matrix, wheel-rail contact nonlinear force function and soil structure boundary impedance parameters into a system-level assembly in a multiphysics field co-simulation environment to form an overall coupled dynamics equation set. The variable-step implicit integration algorithm is used to solve the overall coupled dynamic equations, generating the state parameter matrix of the coupled system, which includes: By calling the excitation spectrum matrix and the modal space vibration state vector constructed by modal filter transformation of the vibration response data of the vehicle suspension system, a vehicle-track coupled dynamic model is constructed. The vehicle system is described by multi-rigid-body dynamics, the car body, frame, and wheelset are defined as rigid bodies, the suspension elements are defined as force elements, and the track structure is discretized into beam elements and plate elements using finite element method. The vehicle-track subsystem matrix is ​​then assembled to obtain the vehicle mass matrix, which includes the vehicle mass matrix, vehicle stiffness matrix, track mass matrix, and track stiffness matrix. By calling the dynamic characteristic parameters of the foundation soil and the modal frequency data of the building structure obtained from the multiphysics field correlation monitoring dataset of the vehicle base, a coupled dynamic model of foundation-building is established. The foundation is established as a half-space model using the boundary element method, and the building is described by the modal superposition method. The boundary impedance parameters of the soil structure are obtained by inversion of the measured frequency response function. The boundary impedance parameters of the soil structure characterize the energy transfer characteristics of the vibration wave at the interface between the foundation and the building. The soil structure subsystem matrix is ​​derived, which includes the foundation impedance matrix and the building modal mass matrix. The vehicle track subsystem matrix and the soil structure subsystem matrix are connected by a wheel-rail contact nonlinear force function. This wheel-rail contact nonlinear force function is constructed using Hertzian contact theory and Kalker creep theory. Constraints are applied at the interface nodes using the Lagrange multiplier method, assembling to form a global coupled dynamic equation system. The calculation formula is as follows: In the formula, Represents the total mass matrix. Represents the generalized acceleration vector. Represents the total damping matrix. Represents the generalized velocity vector. Represents the overall stiffness matrix. Represents the generalized displacement vector. This represents the nonlinear force function of wheel-rail contact. Indicates the boundary impedance parameters of the soil structure; The variable step-size implicit integration algorithm is called to solve the overall coupled dynamic equations and generate the coupled system state parameter matrix, which includes the system displacement, velocity and acceleration response at each time step.

4. The method for predicting and optimizing vibration and noise at a four-body coupled vehicle base according to claim 1, characterized in that, The coupled system state parameter matrix is ​​input into a preset energy field analysis module to extract the vibration energy distribution characteristics of the system and output vibration energy density field data. Simultaneously, environmental temperature and humidity monitoring data is imported into an acoustic propagation calculation engine to construct a refraction and propagation model of sound waves in a temperature and humidity gradient field, generating sound ray bending trajectory data. Vehicle operating condition time-series data is loaded into a wave equation solver to establish a dispersion characteristic model of vibration waves in the foundation soil, obtaining frequency-wavenumber domain response data. Then, a finite element-boundary element coupled algorithm is used to perform spatiotemporal domain dual discretization and integration operations on the vibration energy density field data, sound ray bending trajectory data, and frequency-wavenumber domain response data, outputting vibration and noise propagation characteristic field distribution data, including: The system calls the state parameter matrix of the coupled system, extracts the velocity vector and stress vector of each unit node step by step, calculates the kinetic energy density and strain energy density for each unit, and accumulates them after traversing all units to obtain the total energy density field of the system. The total energy density field is then gridded according to spatial coordinates to generate vibration energy density field data. By calling environmental temperature and humidity monitoring data, the sound velocity field distribution is calculated according to the time series. A temperature and humidity gradient field is established in the sound propagation calculation domain. The ray tracing method is used to solve the refraction path of the sound wave in the gradient field. The sound ray path from the track structure through the soil to the building structure is traced to generate sound ray bending trajectory data. Using time-series data of vehicle operation conditions, time-varying boundary conditions are set in the wave equation solver to establish a dispersion characteristic model of vibration waves in the foundation soil. Frequency-wavenumber domain response data are obtained through wavenumber space scanning, where the frequency-wavenumber domain response data describes the propagation speed and attenuation law of vibration waves of different frequencies in the soil. The finite element-boundary element coupled algorithm is used to perform spatiotemporal dual discretization integration on the vibration energy density field data, sound ray bending trajectory data and frequency-wavenumber domain response data. The time discretization adopts the central difference scheme and the spatial discretization adopts the Gaussian integral scheme. After integration, the vibration noise propagation characteristic field distribution data are output.

5. The four-body coupled vehicle base vibration noise prediction and optimization method according to claim 4, characterized in that, The formula for calculating kinetic energy density is as follows: In the formula, Kinetic energy density, For unit material density, The vibration velocity of a single particle; The formula for calculating strain energy density is as follows: In the formula, For strain energy density, For element stress, For unit strain.

6. The method for predicting and optimizing vibration and noise at a four-body coupled vehicle base according to claim 1, characterized in that, The process involves importing vibration and noise propagation feature field distribution data into a spatiotemporal attention mechanism deep learning prediction framework, constructing a network topology to form a prediction model, and inputting historical field distribution data into the input layer of the prediction model for data standardization mapping to generate standardized feature vectors, including: The vibration and noise propagation feature field distribution data is imported into a spatiotemporal attention mechanism deep learning prediction framework. A neural network topology containing a temporal attention module and a spatial attention module is constructed. The temporal attention module generates attention weights by calculating the dot product similarity between the query vector and the key vector. The spatial attention module extracts spatial features of the vibration and noise field through convolutional layers and generates a spatial weight matrix. The temporal attention weights and spatial attention weights are multiplied element-wise to obtain the spatiotemporal joint attention weights. This weight is applied to the value vector weighted summation to form the attention output features. After stacking multiple attention mechanisms, a batch normalization layer and an activation function layer are connected. The completed network topology is used to receive standardized feature vectors. Historical field distribution data is input into the model's input layer, and Z-score standardization mapping is performed. The mean and standard deviation of the time series data for each spatial measurement point are calculated independently to generate a standardized feature vector. The formula for calculating the standardized feature vector is as follows: In the formula, This represents the standardized data. This represents the original vibration and noise field data. This represents the sample mean of the time series data at this measurement point. This represents the sample standard deviation of the time series data at this measurement point.

7. The method for predicting and optimizing vibration and noise at a four-body coupled vehicle base according to claim 1, characterized in that, The process involves using a Long Short-Term Memory (LSTM) network layer to encode the temporal dependencies of standardized feature vectors, extracting hidden state feature sequences, and mapping these sequences to future noise prediction values ​​via a fully connected layer. Backpropagation is then executed to train the network parameters until convergence. The converged network outputs a future noise prediction data sequence, including: The standardized feature vectors are input into the Long Short-Term Memory (LSTM) network layer for temporal dependency encoding. The number of hidden units and layers in the LSTM are set, and a bidirectional LSTM structure is used to extract temporal features from both the forward and backward directions. The bidirectional hidden state vectors are concatenated to obtain the hidden state feature sequence. The calculation formula for the LSTM network state update is as follows: In the formula, This represents the hidden state vector at the current time. This represents the standardized feature vector input at the current time. This represents the hidden state vector from the previous time step. A fully connected layer is set to map the hidden state feature sequence to the noise prediction value of the future time period. The output dimension of the fully connected layer is set to the product of the number of prediction time periods and the number of measurement points. The mean square error loss function is used to calculate the error between the prediction value and the true value. The backpropagation algorithm is executed to train the network parameters to the convergence state. The convergence condition is set to the validation set loss no longer decreasing after several consecutive rounds. After the network has converged, it outputs the noise prediction data sequence of the future time period.

8. The method for predicting and optimizing vibration and noise at a four-body coupled vehicle base according to claim 1, characterized in that, Based on the future noise prediction data sequence, a chromosome coding structure for the vehicle scheduling scheme is designed to generate a candidate solution population. A fitness function is defined with the dual objectives of minimizing noise overscalar and maximizing operational efficiency, and the candidate solution population is evaluated using the fitness function. The candidate solution population after individual evaluation is initialized as the initial solution set. Selection, crossover, and mutation genetic operations are performed on the initial solution set to drive the population to iterative evolution. When the convergence condition is met, the Pareto optimal front solution set is obtained. The optimal individual in the Pareto optimal front solution set is decoded, and the real-time optimization scheme for vehicle base operation and maintenance scheduling is output, including: Based on the noise prediction data sequence for future time periods, a chromosome coding structure for vehicle scheduling scheme is designed. The chromosome gene positions are encoded with real numbers and correspond to the train's entry and exit times and running speed levels on the designated track. A candidate solution population is generated through random initialization. Define a dual-objective fitness function that includes minimizing noise superscalar and maximizing operational efficiency, and initialize the candidate solution population to form the initial solution set based on this function; Tournament selection, simulated binary crossover, and polynomial mutation genetic operations are sequentially applied to the initial solution set to drive the iterative evolution of the population. When the convergence condition is met, the Pareto optimal frontier solution set is obtained, the optimal individual in the solution set is decoded, and the real-time optimization scheme for vehicle base operation and maintenance scheduling is output.

9. A vibration and noise prediction and optimization system for a four-body coupled vehicle base, based on the vibration and noise prediction and optimization method for a four-body coupled vehicle base as described in claim 1, characterized in that, include: Acquisition Module: Used to acquire the multi-physics field correlation monitoring dataset of the vehicle base. This dataset includes dynamic irregularity detection data of the vehicle base track, vibration response data of the vehicle suspension system, dynamic characteristic parameters of the foundation soil, modal frequency data of the building structure, environmental temperature and humidity monitoring data, and time series data of vehicle operating conditions. Assembly and Generation Module: Based on the multiphysics-related monitoring dataset of the vehicle base, this module constructs a vehicle-track coupled dynamics model and derives the vehicle track subsystem matrix. Simultaneously, it establishes a foundation-structure coupled dynamics model and derives the soil structure subsystem matrix, defining the wheel-rail contact nonlinear force function and soil structure boundary impedance parameters. Using the Lagrange multiplier method, the vehicle track subsystem matrix, soil structure subsystem matrix, wheel-rail contact nonlinear force function, and soil structure boundary impedance parameters are assembled hierarchically in a multiphysics-based co-simulation environment to form a comprehensive set of coupled dynamic equations. The variable step-size implicit integration algorithm is called to solve the overall coupled dynamic equations and generate the state parameter matrix of the coupled system. The output module is designed to input the coupled system state parameter matrix into a preset energy field analysis module, extract the system vibration energy distribution characteristics, and output vibration energy density field data. Simultaneously, it imports environmental temperature and humidity monitoring data into the acoustic propagation calculation engine to construct a refraction and propagation model of sound waves in a temperature and humidity gradient field, generating sound ray bending trajectory data. Furthermore, it loads vehicle operating condition time-series data into the wave equation solver to establish a dispersion characteristic model of vibration waves in the foundation soil, obtaining frequency-wavenumber domain response data. Finally, it employs a finite element-boundary element coupled algorithm to perform spatiotemporal dual discretization and integration operations on the vibration energy density field data, sound ray bending trajectory data, and frequency-wavenumber domain response data, outputting vibration and noise propagation characteristic field distribution data. Training Module: This module is used to import vibration and noise propagation feature field distribution data into a spatiotemporal attention mechanism deep learning prediction framework, build a network topology to form a prediction model, input historical field distribution data into the input layer of the prediction model for data standardization mapping, and generate standardized feature vectors. A Long Short-Term Memory (LSTM) network layer is used to encode the temporal dependencies of the standardized feature vectors, extract the hidden state feature sequences, and map the hidden state feature sequences to future vibration and noise prediction values ​​through a fully connected layer. The backpropagation algorithm is executed to train the network parameters until convergence. After convergence, the network outputs a future vibration and noise prediction data sequence. Iterative Module: Based on the noise prediction data sequence for future time periods, this module designs the chromosome encoding structure of the vehicle scheduling scheme to generate a candidate solution population. It defines a fitness function with the dual objectives of minimizing noise overscalar and maximizing operational efficiency, and uses the fitness function to evaluate individual candidates in the candidate solution population. The candidate solution population after individual evaluation is initialized as the initial solution set. Selection, crossover, and mutation genetic operations are performed on the initial solution set to drive the population to iteratively evolve. When the convergence condition is met, the Pareto optimal front solution set is obtained. The optimal individual in the Pareto optimal front solution set is decoded, and the real-time optimization scheme for vehicle base operation and maintenance scheduling is output.

10. The four-body coupled vehicle base vibration and noise prediction and optimization system according to claim 9, characterized in that, The acquisition module includes: The first acquisition unit is used to deploy track inspection vehicles and laser displacement sensor arrays in the track section of the vehicle base. It performs spatial positioning based on the kilometer marker, collects track gauge, elevation, and directional irregularity data, and simultaneously performs millimeter-wave radar scanning on the rail surface corrugation to obtain corrugation depth spectrum data. The track gauge, elevation, directional irregularity data and corrugation depth spectrum data are uniformly stored as a spatiotemporal two-dimensional array structure. The time dimension corresponds to the inspection date, and the spatial dimension corresponds to the line mileage. It is then converted into an excitation spectrum matrix, which constitutes the external excitation input of the vehicle-track coupling model. The second acquisition unit is used to install inertial measurement units at three locations: the bogie axle box, frame, and car body of the monitored vehicle. It collects the vertical, lateral, and longitudinal vibration acceleration time history data of the vehicle when it runs at low speed on the track in the depot. The acquired vibration acceleration time history data is zero drift eliminated and trend term removed. The dynamic displacement and dynamic velocity of the suspension elements are obtained through integral calculation to form the vibration response data of the vehicle suspension system. Extraction Unit: This unit deploys vibration sensor arrays and environmental temperature and humidity sensors in key areas of the vehicle depot's throat zone, maintenance depot, and main office building structure. It acquires dynamic characteristics of the foundation soil through on-site wave velocity testing and indoor resonant column tests, obtains modal frequency data of the building structure through hammer impact or environmental vibration testing, and extracts vehicle operating condition time-series data from the vehicle operation management system. All data is timestamped and unified to the same time base using linear interpolation to generate a multi-physics field correlation monitoring dataset for the vehicle depot. This dataset provides boundary conditions for the coupling model parameters.

Images