A method for predicting and optimizing the sound insulation performance of a rail vehicle

Abstract

This invention provides a method for predicting and optimizing the sound insulation performance of rail vehicles. The method includes: constructing and maintaining a unified database to centrally and standardizedly manage sound insulation material parameters and sound insulation sample data of the vehicle body structure; performing data augmentation based on initial samples, combining feature weighting processing with material layer location, and then determining feature weights through the mRMR algorithm and exponential decay weight strategy to finally train a sound insulation performance prediction model; using the database and the sound insulation performance prediction model to predict and evaluate the sound insulation performance of user-defined material combination schemes; and automatically searching for the optimal sound insulation scheme based on the database and the sound insulation performance prediction model, using an adaptively selected optimization algorithm, according to the scheme and constraints set by the user. This invention solves the problems of scattered data management, limited model prediction accuracy, reliance on experience in scheme design, and low optimization efficiency in traditional sound insulation design.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of rail vehicle design and manufacturing technology, and in particular to a method for predicting and optimizing the sound insulation performance of a rail vehicle. BACKGROUND

[0002] The sound insulation design of a rail vehicle body is a key to improving ride comfort, and its effectiveness directly depends on accurate control of sound insulation material parameters, in-depth mining of historical sample data, and continuous optimization of sound insulation schemes. However, there are several technical bottlenecks in the current industry. First, the basic parameters of sound insulation materials and real vehicle test sample data are often in a scattered and isolated management state, lacking a unified standard and format, resulting in uneven data quality and laying hidden dangers for subsequent construction of reliable prediction models. Second, in the model training stage, most methods only rely on limited original sample data, neither expanding sample diversity through data augmentation techniques nor effectively quantifying the impact of key features such as material layer position on sound insulation performance, resulting in limited prediction accuracy and generalization ability of the model.

[0003] In terms of scheme design and optimization, engineers often rely on personal experience for repeated trial and error and physical testing, lacking an intelligent tool that can quickly and accurately predict the sound insulation performance of any material combination. This makes it extremely difficult to optimize schemes under multiple constraints (such as sound insulation level, total thickness, overall weight, and cost), resulting in a long design cycle and high research and development costs. In addition, existing methods often use a single search strategy when automating optimization, which cannot dynamically adapt to efficient algorithms according to the scale of material combinations, and often fall into the dilemma of low computational efficiency or even exhaustion of system resources when facing large-scale scheme spaces. Therefore, the industry urgently needs an integrated solution that can cover the entire process of data management, model training, scheme evaluation, and intelligent optimization. SUMMARY

[0004] To solve the above problems, the present application builds a standardized data management foundation and trains a high-precision sound insulation performance prediction model based on it, achieving rapid evaluation of custom sound insulation schemes and intelligent optimization under multiple constraints, solving the problems of scattered data management, limited model prediction accuracy, reliance on experience for scheme design, and low optimization efficiency in traditional sound insulation design. The sound insulation design is transformed from the traditional mode of relying on trial and error and physical testing to the modern mode of data-driven, model evaluation, and intelligent optimization, thereby significantly improving design efficiency, shortening the research and development cycle, and reducing development costs.

[0005] According to an embodiment of the present application, a method for predicting and optimizing the sound insulation performance of a rail vehicle is provided.

[0006] In a first aspect of the present application, a method for predicting and optimizing the sound insulation performance of a rail vehicle is provided. The method comprises: Step S01: Build and maintain a unified database to centrally and standardizedly manage the sound insulation material parameters and vehicle body structure sound insulation sample data; Step S02: Perform data augmentation based on the initial samples, combine the feature weighting processing of the material layer location, and then determine the feature weights through the mRMR algorithm and the exponential decay weight strategy to finally train the sound insulation performance prediction model. Step S03: Using the database and sound insulation performance prediction model, predict and evaluate the sound insulation performance of the user-defined material combination scheme; Step S04: Based on the user-defined scheme and constraints, and using the database and sound insulation performance prediction model, the optimal sound insulation scheme is automatically searched through an adaptively selected optimization algorithm.

[0007] Furthermore, the sound insulation material parameters mentioned in step S01 include: surface density, thickness, weighted sound insulation Rw, and sound absorption coefficient NRC; The aforementioned vehicle body structure sound insulation sample data is stored according to region category and layered material layer; The area categories include: floor, side walls, and ceiling. The layered material comprises: an inner layer, a core layer, an aluminum profile layer, and an outer cover layer. The aforementioned interior layers include: a floor fabric layer, an interior panel layer, and a damping layer. The core layers include: a first core layer to an eighth core layer. The aluminum profile layer includes: an inner damping layer, an aluminum profile layer, an outer damping layer, a cavity sound-absorbing layer, and a cavity damping layer. The outer cover layer includes: a first sound-absorbing layer, a second sound-absorbing layer, and an outer cover panel layer.

[0008] Furthermore, the centralized and standardized management of sound insulation material parameters and vehicle body structure sound insulation sample data described in step S01 specifically adopts a material library management unit and a sample library management unit. The material library management unit supports classification and management by material type, allows editing of material parameters through Excel files or software interfaces, automatically verifies the validity of parameter formats, stores materials in an in-memory database by category and creates an index table, and automatically counts the number of material types and samples. The sample library management unit supports the classification and management of samples by region. When entering samples, materials are selected in the order of inner layer, core layer, aluminum profile layer, and outer cover layer. The unit automatically calculates the total surface density and total thickness of the sample, extracts the sound insulation / sound absorption characteristics of each core layer material to generate feature values, and updates the feature library synchronously.

[0009] Further, the specific steps for data augmentation based on the initial sample described in step S02 are as follows: extract the sound insulation feature value and sound absorption feature value of each layer in the core layer of the sample. The sound insulation feature value is calculated as STL_Value=M×Rw, where M is the areal density of the material layer and Rw is the weighted sound insulation of the material layer; the sound absorption feature value is calculated as SAC_Value=D×NRC, where D is the thickness of the material layer and NRC is the sound absorption coefficient of the material layer; identify non-empty material layers, and generate virtual samples by randomly adjusting the positions of the non-empty material layers without changing the order of the non-empty material layers. Each original sample then generates several virtual samples to obtain the augmented sample set.

[0010] Furthermore, the specific steps for training the sound insulation performance prediction model in step S02, which involve combining the feature weighting of the material layer location, determining the feature weights through the mRMR algorithm and the exponential decay weighting strategy, are as follows: The position-weighted score of the material layer is calculated using the material layer position-weighted scoring algorithm, and the position-weighted feature is obtained. The expression is: Ins_score=0.8+0.4×cos(2π×dis)+0.2×(1-2×dis), where dis is the ratio of the distance from the center of the material layer to the boundary of the core layer to the core layer thickness limit. The mRMR algorithm is used to rank the features of the enhanced sample set by importance, and the weighted sample feature set is obtained by weighting each feature based on the exponential decay weight strategy. The weight calculation expression is: Weights_exp(i)=exp(-decay_rate×(rank_position-1)), where decay_rate is the decay rate and rank_position is the ranking position of the feature. The weighted sample feature set is used as input, and the weighted sound insulation Rw is used as the output target. The support vector machine regression algorithm is used for training. During training, the parameters C and g of the support vector machine regression algorithm are optimized by grid search, and the model is evaluated by 3-fold cross-validation. When the mean squared error (MSE) of the model on the training set and the test set is less than 0.1 and the correlation coefficient (R²) is greater than 0.9, the model is considered to have met the standard and is saved. At the same time, the fitting results of the model on the training set and the test set, as well as the MSE and R² evaluation indicators, are output.

[0011] Furthermore, the specific steps of step S03 are as follows: The system specifies the area category and core layer thickness limit, combines materials by layer, and calculates the total thickness, total areal density and total parameters of the core layer material in real time. If the core layer thickness exceeds the limit, a prompt mechanism is triggered. Based on the loaded model and scheme parameters, the system calls the material layer location weighted scoring algorithm and the sound insulation performance prediction model to output the Rw prediction result. It supports scheme modification and re-prediction. After the prediction is completed, a Word format report containing material composition, parameter statistics and prediction results can be exported.

[0012] Furthermore, the constraints described in step S04 include: maximum limit for core layer thickness, maximum limit for overall thickness, maximum limit for core surface density, maximum limit for overall surface density, and minimum limit for weighted sound insulation.

[0013] Furthermore, the optimal sound insulation scheme described in step S04 includes: a scheme with maximum sound insulation and a lightweight scheme that meets the sound insulation limit, wherein: Under the maximum sound insulation scheme, constraints on core layer thickness, overall thickness, core surface density, and overall surface density can be set. Under the lightweight scheme that meets the sound insulation limit, it is possible to set the minimum limit constraints for core layer thickness, overall thickness, and weighted sound insulation, including two sub-modes: lightest core layer weight and lightest overall weight.

[0014] Furthermore, the specific content of the adaptive selection optimization algorithm described in step S04 is as follows: When the number of possible combinations is less than or equal to 50 million, an exhaustive algorithm is used to output the three-dimensional scatter plot of the solutions that meet the conditions and the global optimal solution. If the number of solutions that meet the conditions is greater than 1,000, then only the three-dimensional scatter plot of the first 1,000 solutions and the global optimal solution are output. When the number of possible solutions exceeds 50 million, a genetic algorithm is used, with a default iteration of 200 generations, outputting the optimization convergence curve and the global optimal solution.

[0015] In a second aspect of the invention, an apparatus for predicting and optimizing the sound insulation performance of rail vehicles is provided. The apparatus includes: Database Management Module: Used to build and maintain a unified database for centralized and standardized management of sound insulation material parameters and vehicle body structure sound insulation sample data; Model training module: used for data augmentation based on initial samples, combined with feature weighting processing of material layer location, and then determining feature weights through mRMR algorithm and exponential decay weight strategy, finally training to obtain sound insulation performance prediction model; Solution Prediction Module: Used to predict and evaluate the sound insulation performance of user-defined material combination schemes using a database and sound insulation performance prediction model; Solution Optimization Module: Based on the user-defined solutions and constraints, and using a database and sound insulation performance prediction model, the module automatically searches for the optimal sound insulation solution through an adaptive optimization algorithm.

[0016] This invention establishes a standardized data management foundation and trains a high-precision sound insulation performance prediction model based on it. This enables rapid evaluation of customized sound insulation schemes and intelligent optimization under multiple constraints. It solves the problems of scattered data management, limited model prediction accuracy, reliance on experience in scheme design, and low optimization efficiency in traditional sound insulation design. It transforms sound insulation design from a traditional model that relies on trial and error and physical testing to a modern model that is data-driven, model-evaluated, and intelligently optimized. This significantly improves design efficiency, shortens the R&D cycle, and reduces development costs.

[0017] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description.

[0018] The beneficial effects of this invention are: 1. It has achieved centralized, standardized and unified management of sound insulation material parameters and vehicle body structure sample data, fundamentally solving the problems of scattered data and chaotic formats, providing a high-quality and consistent data foundation for subsequent model training and scheme design, and significantly improving the standardization and reliability of the overall work; 2. By training the prediction model by fusing data augmentation and feature weighting processing based on the location of the material layer, the accuracy bottleneck caused by the limited sample and incomplete feature consideration of traditional methods is effectively overcome, thereby constructing a high-precision sound insulation performance prediction model and greatly reducing the dependence on expensive physical tests. 3. It provides fast and accurate customized solution evaluation capabilities. Users can flexibly combine materials through the system and obtain prediction results in real time, realizing digital rapid iteration and verification of design solutions, which greatly improves the evaluation efficiency and decision-making speed in the early stage of design. 4. It can automatically and efficiently search for the global or near-optimal sound insulation scheme under multiple objectives and constraints set by the user, and balance the optimization accuracy and computational efficiency through an adaptive algorithm selection mechanism. It systematically solves the complex optimization problem that is difficult to balance multiple requirements such as sound insulation performance, weight, and thickness in traditional empirical design. 5. A complete closed-loop workflow has been formed, from data preparation, model building, scheme evaluation to intelligent optimization, which transforms sound insulation design from an experience-based trial-and-error mode to a data and model-driven systems engineering mode. Overall, the design cycle has been significantly shortened, and R&D costs and over-reliance on personal experience have been reduced. Attached Figure Description

[0019] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. Wherein: Figure 1A flowchart illustrating a method for predicting and optimizing the sound insulation performance of rail vehicles according to an embodiment of the present invention is shown. Figure 2 A schematic diagram of the interface for adding a new sample according to an embodiment of the present invention is shown; Figure 3 A schematic diagram of the interface for predicting sound insulation using a custom scheme according to an embodiment of the present invention is shown. Figure 4 A schematic diagram of a material selection interface according to an embodiment of the present invention is shown; Figure 5 A schematic diagram showing the possible combinations of schemes according to an embodiment of the present invention is provided. Figure 6 A schematic diagram of the optimization results under multiple constraints according to an embodiment of the present invention is shown; Figure 7 A block diagram of a device for predicting and optimizing the sound insulation performance of rail vehicles according to an embodiment of the present invention is shown. Detailed Implementation

[0020] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] According to an embodiment of the present invention, a method for predicting and optimizing the sound insulation performance of rail vehicles is proposed. By constructing a standardized data management foundation and training a high-precision sound insulation performance prediction model based on it, a rapid evaluation of customized sound insulation schemes and intelligent optimization under multiple constraints are realized. This solves the problems of scattered data management, limited model prediction accuracy, reliance on experience in scheme design, and low optimization efficiency in traditional sound insulation design. It transforms sound insulation design from a traditional mode that relies on trial and error and physical testing to a modern mode that is data-driven, model-evaluated, and intelligently optimized, thereby significantly improving design efficiency, shortening the R&D cycle, and reducing development costs.

[0022] The principles and spirit of the present invention will be explained in detail below with reference to several representative embodiments.

[0023] Figure 1 This is a schematic flowchart illustrating a method for predicting and optimizing the sound insulation performance of rail vehicles according to an embodiment of the present invention. The method includes: Step S01: Build and maintain a unified database to centrally and standardizedly manage the sound insulation material parameters and vehicle body structure sound insulation sample data; Step S02: Perform data augmentation based on the initial samples, combine the feature weighting processing of the material layer location, and then determine the feature weights through the mRMR algorithm and the exponential decay weight strategy to finally train the sound insulation performance prediction model. Step S03: Using the database and sound insulation performance prediction model, predict and evaluate the sound insulation performance of the user-defined material combination scheme; Step S04: Based on the user-defined scheme and constraints, and using the database and sound insulation performance prediction model, the optimal sound insulation scheme is automatically searched through an adaptively selected optimization algorithm.

[0024] It should be noted that although the operation of the method of the present invention has been described in a specific order in the above embodiments and figures, this does not require or imply that the operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0025] To provide a clearer explanation of the above-mentioned method for predicting and optimizing the sound insulation performance of rail vehicles, a specific embodiment will be used for illustration below. However, it is worth noting that this embodiment is only for better illustrating the present invention and does not constitute an improper limitation of the present invention.

[0026] The following example will further illustrate the methods for predicting and optimizing the sound insulation performance of rail vehicles.

[0027] In this embodiment, a material library .xlsx file (containing the surface density, thickness, weighted sound insulation Rw, and noise reduction coefficient NRC parameters of sound insulation materials, sound absorption materials, damping materials, flooring, interior panels, and aluminum profiles) and an initial sample library .xlsx file (containing region categories, material combinations, Rw test results, etc.) are prepared. As shown in Table 1, the relevant parameters of the sound insulation materials in the material library file are displayed. As shown in Table 2, some data from the initial sample library are displayed. The material parameters are preferentially based on measured data or authoritative data from manufacturers.

[0028] Table 1

[0029] Table 2

[0030] Step S01: Build and maintain a unified database to centrally and standardizedly manage the parameters of sound insulation materials and the sample data of sound insulation of vehicle body structure.

[0031] The parameters of sound insulation materials include: surface density (kg / m²), thickness (mm), weighted sound insulation Rw (dB), and sound absorption coefficient NRC.

[0032] The sound insulation sample data of the vehicle body structure is stored according to the region category and the layer of materials.

[0033] Area categories include: floor, side walls, and ceiling.

[0034] The layered material includes: an inner layer, a core layer, an aluminum profile layer, and an outer cover layer. The inner layer includes 3 layers: a floor cloth layer, an inner panel layer, and a damping layer. The core layer includes 8 layers: the first core layer to the eighth core layer. The aluminum profile layer includes 5 layers: an inner damping layer, an aluminum profile layer, an outer damping layer, a cavity sound-absorbing layer, and a cavity damping layer. The outer cover layer includes 3 layers: the first sound-absorbing layer, the second sound-absorbing layer, and the outer cover panel layer.

[0035] The database supports statistical querying, adding, deleting, editing, and loading of sound insulation material parameters and sample data.

[0036] The parameters of sound insulation materials and the sample data of sound insulation of vehicle body structure are centrally and standardizedly managed, specifically by adopting a material library management unit and a sample library management unit.

[0037] The material library management unit supports categorized management by material type (sound insulation material, sound absorption material, damping material, flooring, interior panels, aluminum profiles). Material parameters can be edited through Excel files or software interfaces. The system automatically verifies the validity of parameter formats (numerical parameters must be non-negative, and text parameters must be non-empty), stores them in the in-memory database by category, and creates an index table. It also automatically counts the number of material types and samples.

[0038] The sample library management unit supports the classification and management of samples by region category (floor, side wall, roof). When entering samples, materials are selected in the order of inner layer, core layer, aluminum profile layer, and outer cover layer. The unit automatically calculates the total surface density and total thickness of the sample, extracts the sound insulation / sound absorption characteristics of each core layer material to generate feature values, and updates the feature library synchronously.

[0039] In this embodiment, the material library .xlsx file is read, and the material parameters in the file are format-validated to ensure that each column conforms to the corresponding input format or type. After successful validation, the materials are categorized by type and stored in an in-memory database, creating a parameter index table. Users can add, modify, or delete material information through the software interface or by directly editing the Excel file. Upon completion of editing, the system automatically updates the in-memory database and index table, generating an operation log. The in-memory database is traversed, and the number of valid material types and samples is counted by category, generating standardized statistical results (e.g., 14 types of sound insulation materials and 21 types of sound absorption materials), which users can view in real time.

[0040] The system reads the sample library .xlsx file, synchronously loads the associated material library data, and verifies the consistency between the material information in the sample and the material library. Users select the target area category, input the core layer thickness limit, and select materials in the order of inner layer, core layer, aluminum profile layer, and outer cover layer. The system automatically extracts the parameters of each material, calculates the total areal density and total thickness of the sample, extracts the sound insulation / absorption characteristics of the eight material layers, generates feature values, and synchronously updates the feature library. The system supports loading, viewing, and editing of samples. Users can view sample statistics through the software interface or directly open an Excel file to modify sample data; after modification, the system automatically recalculates the feature values.

[0041] In this embodiment, as Figure 2 As shown, the user selects the target area category: flooring, enters the core layer thickness limit: 60mm, selects the inner layer in sequence as 3mm TPU flooring fabric, 19.5mm wood plywood, selects the core layer in sequence as 0.8mm VINAFLEX sound insulation felt, 30mm carbon fiber cotton 10K, 0.8mm VINAFLEX sound insulation felt, 0.8mm VINAFLEX sound insulation felt, selects the aluminum profile layer in sequence as 2mm 123WF damping paste, 400BF floor aluminum profile, 2.85mm fireproof paint, 40mm carbon fiber cotton 10K, 0.8mm damping patch, selects the outer layer in sequence as 10mm carbon fiber cotton 10K, 20mm glass wool 60K, 2mm VINAFLEX sound insulation felt, and finally enters the Rw test result of 53.8dB; the user then... Once a sample is written through the software interface, the system will automatically extract the material parameters of each layer and calculate the total surface density (obtained by summing the surface densities of each layer) and total thickness (obtained by summing the thicknesses of each material layer in the inner layer, the aluminum profile, the outer cover, and the core layer thickness limit). Then, it will extract the sound insulation / sound absorption characteristics of the eight core layer material layers and generate feature values ​​(if it is a sound insulation material, its sound insulation feature value will be calculated; if it is a sound absorption material, its sound absorption feature value will be calculated. The formula for calculating the sound insulation feature value is STL_Value=M×Rw, where M is the surface density of the material layer and Rw is the weighted sound insulation of the material layer; the formula for calculating the sound absorption feature value is SAC_Value=D×NRC, where D is the thickness of the material layer and NRC is the sound absorption coefficient of the material layer), and update the feature library synchronously.

[0042] Step S02: Perform data augmentation based on the initial samples, combine the feature weighting processing of the material layer location, and then determine the feature weights through the mRMR algorithm and the exponential decay weight strategy to finally train the sound insulation performance prediction model.

[0043] The specific steps for data augmentation based on the initial samples are as follows: Extract the sound insulation and sound absorption characteristic values ​​of each layer in the sample core layer. The sound insulation characteristic value is calculated as STL_Value = M × Rw, where M is the areal density of the material layer and Rw is the weighted sound insulation of the material layer. The sound absorption characteristic value is calculated as SAC_Value = D × NRC, where D is the thickness of the material layer and NRC is the sound absorption coefficient of the material layer. Identify non-empty material layers (both STL_Value and SAC_Value are not 0). Without changing the order of the non-empty material layers, generate virtual samples by randomly adjusting their positions. Several virtual samples are then generated corresponding to each original sample, resulting in an augmented sample set. The target values ​​of the virtual samples are consistent with those of the original samples, effectively enhancing the robustness and generalization of the system.

[0044] The specific steps for training the sound insulation performance prediction model are as follows: Combining feature weighting based on the location of the material layer, determining feature weights using the mRMR algorithm and exponential decay weighting strategy, and finally training the model: The position-weighted score algorithm of the material layer is used to calculate the position-weighted score of the material layer, quantify the influence of the material layer position on the acoustic performance of the core layer, and obtain the position-weighted feature, which is expressed as: Ins_score=0.8+0.4×cos(2π×dis)+0.2×(1-2×dis), where dis is the ratio of the distance from the center of the material layer to the boundary of the core layer to the core layer thickness limit; The mRMR (maximum relevance-minimum redundancy) algorithm is used to rank the features of the enhanced sample set by importance, and each feature is weighted based on the exponential decay weight strategy (feature value × corresponding weight value) to obtain the weighted sample feature set. The weight calculation expression is: Weights_exp(i)=exp(-decay_rate×(rank_position-1)), where decay_rate is the decay rate, rank_position is the ranking position of the feature, and the weight of the feature after weighting is in the range of 0~1; The weighted sample feature set is used as input, and the weighted sound insulation Rw is used as the output target. The support vector machine regression algorithm is used for training. During training, the parameters C and g of the support vector machine regression algorithm are optimized by grid search, and the model is evaluated by 3-fold cross-validation. When the mean squared error (MSE) of the model on the training set and the test set is <0.1 and the correlation coefficient (R²) is >0.9, the model can be considered to have met the standard and is saved. At the same time, the fitting results of the model on the training set and the test set, as well as the evaluation indices of MSE and R², are output.

[0045] In this embodiment, standardized sample data from the sample library is loaded. Each sample contains eight core material layers, some of which are non-empty (represented by non-zero values), and the rest are empty (represented by 0). The non-empty material layers in each sample are extracted, maintaining their original order. Then, several positions from eight randomly selected locations (the number equal to the number of non-empty layers) are used to place these non-empty layers in sequence, filling the remaining positions with 0, thus generating a virtual sample. This process is repeated five times for each original sample, resulting in five virtual samples, ultimately forming the enhanced sample set.

[0046] The mRMR algorithm is used to rank the importance of features in the augmented sample set. Feature weights are calculated based on an exponential decay strategy (decay rate 0.1), as shown in Table 3. The weights of 36 features are given in order of importance. Then, these weight values ​​are multiplied by the corresponding feature values ​​to complete the feature weighting process. The support vector machine regression algorithm is called to optimize parameters C and g through grid search. The model is trained using 3-fold cross-validation, and the training progress is output in real time. After training, the fitting curves of the training set and the test set and the evaluation index are displayed. If the model evaluation index meets the requirements (MSE < 0.1 and R² > 0.90), the user can choose the save path and save the model as a MAT format file (the naming format is generally "date_R² value.mat") for easy access later.

[0047] Table 3

[0048] Step S03: Using the database and sound insulation performance prediction model, predict and evaluate the sound insulation performance of the user-defined material combination scheme.

[0049] The specific steps are as follows: load the material library and the trained sound insulation performance prediction model, and verify the integrity of the material parameters and the compatibility of the model; The system specifies the area category and core layer thickness limit, combines materials by layer, and calculates the total thickness, total areal density and total parameters of the core layer material in real time. If the core layer thickness exceeds the limit, a prompt mechanism is triggered. Based on the loaded model and scheme parameters, the system calls the material layer location weighted scoring algorithm and the sound insulation performance prediction model to output the Rw prediction result. It supports scheme modification and re-prediction. After the prediction is completed, a Word format report containing material composition, parameter statistics and prediction results can be exported.

[0050] In this embodiment, the material library .xlsx file and the trained sound insulation performance prediction model are loaded to verify the compatibility between the material library and the model, and to display the statistical information of the material library (such as...). Figure 3 (As shown in the upper right corner) and the model ready status.

[0051] Users select the area category (flooring), enter a core layer thickness limit of 55mm, and then select materials in the following order: inner layer, core layer, aluminum profile layer, and outer cover layer. Figure 4 As shown, clicking "Build Structure" will display the combined structural scheme below the material composition details. The system will also calculate the total thickness of the core material, the total areal density, and the total parameters of the scheme in real time, displaying them below the scheme. If the core thickness exceeds the limit, an automatic prompt will be displayed.

[0052] After the user triggers the prediction command, the system calls the trained benchmark model, combines the scheme parameters, feature weighting logic and sound insulation score calculation algorithm, and outputs the Rw prediction result of 50.74dB within 1 second.

[0053] After the prediction is completed, the user can trigger the report export command, and the system will automatically generate a Word format report, which includes material composition, parameter statistics and prediction results, and save it to the specified path.

[0054] Step S04: Based on the user-defined scheme and constraints, and using the database and sound insulation performance prediction model, the optimal sound insulation scheme is automatically searched through an adaptively selected optimization algorithm.

[0055] The constraints include: maximum core thickness, maximum overall thickness, maximum core surface density, maximum overall surface density, and minimum weighted sound insulation.

[0056] The specific content of the adaptive selection optimization algorithm is as follows: when the number of scheme combinations is ≤50 million, the exhaustive algorithm is adopted, and the optimization is generally completed within 30 seconds. The three-dimensional scatter plot of the schemes that meet the conditions (when the number of schemes that meet the conditions is >1000, only the first 1000 are output) and the global optimal solution are output. When the number of possible combinations exceeds 50 million, a genetic algorithm is used, with a default iteration of 200 generations. The optimization is usually completed within 3 minutes, and the optimization convergence curve and the global optimal solution are output.

[0057] Optimal sound insulation solutions include: the solution with the maximum sound insulation and the lightweight solution that meets the sound insulation limit, wherein: Under the maximum sound insulation scheme, constraints on core layer thickness, overall thickness, core surface density, and overall surface density can be set. Under the lightweight scheme that meets the sound insulation limit, it is possible to set the minimum limit constraints for core layer thickness, overall thickness, and weighted sound insulation, including two sub-modes: core layer lightweight (lightest core layer weight) and overall lightweight (lightest overall weight).

[0058] In this embodiment, the material library .xlsx file is loaded, and the user selects candidate materials layer by layer, such as... Figure 4 As shown, after completing the alternatives, the system will automatically calculate the number of possible combinations, which is 1,259,712. Figure 5 As shown; Load the adapted model file, the user selects the area category: floor, sets the optimization mode: maximum sound insulation, and sets the constraints: core layer thickness ≤ 60mm, overall thickness ≤ 200mm, core surface density ≤ 60kg / m², and overall surface density ≤ 100kg / m².

[0059] The system automatically selects the algorithm based on the number of possible combinations (exhaustive search algorithm is used when the number of combinations is ≤50 million). After triggering the optimization command, the optimization progress is displayed in real time. Figure 6 As shown, the optimization was completed after 16.98 seconds, and the optimal solution was output: Rw=53.05dB, core layer thickness 14.00mm, overall thickness 166.50mm, and constraint verification results (all parameters meet the set constraints).

[0060] Users can trigger a report export command, and the system will generate a Word format optimization report, which includes the material composition, parameter details and constraint satisfaction of the optimal solution, and save it to a specified path.

[0061] Based on the same inventive concept, this invention also proposes a device for predicting and optimizing the sound insulation performance of rail vehicles. The implementation of this device can be found in the implementation of the method described above; repeated details will not be repeated. Figure 7 As shown, the device includes: Database Management Module: Used to build and maintain a unified database for centralized and standardized management of sound insulation material parameters and vehicle body structure sound insulation sample data; Model training module: used for data augmentation based on initial samples, combined with feature weighting processing of material layer location, and then determining feature weights through mRMR algorithm and exponential decay weight strategy, finally training to obtain sound insulation performance prediction model; Solution Prediction Module: Used to predict and evaluate the sound insulation performance of user-defined material combination schemes using a database and sound insulation performance prediction model; Solution Optimization Module: Based on the user-defined solutions and constraints, and using a database and sound insulation performance prediction model, the module automatically searches for the optimal sound insulation solution through an adaptive optimization algorithm.

[0062] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0063] Furthermore, although the operations are described in a specific order, this should be understood as requiring that such operations be performed in the specific order shown or in sequential order, or requiring that all illustrated operations be performed to achieve the desired result. In certain environments, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the invention. Certain features described in the context of individual embodiments may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations.

[0064] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for predicting and optimizing the sound insulation performance of rail vehicles, characterized in that, The method includes: Step S01: Build and maintain a unified database to centrally and standardizedly manage the sound insulation material parameters and vehicle body structure sound insulation sample data; Step S02: Perform data augmentation based on the initial samples, combine the feature weighting processing of the material layer location, and then determine the feature weights through the mRMR algorithm and the exponential decay weight strategy to finally train the sound insulation performance prediction model. Step S03: Using the database and sound insulation performance prediction model, predict and evaluate the sound insulation performance of the user-defined material combination scheme; Step S04: Based on the user-defined scheme and constraints, and using the database and sound insulation performance prediction model, the optimal sound insulation scheme is automatically searched through an adaptively selected optimization algorithm.

2. The method for predicting and optimizing the sound insulation performance of rail vehicles according to claim 1, characterized in that, The sound insulation material parameters mentioned in step S01 include: surface density, thickness, weighted sound insulation Rw, and sound absorption coefficient NRC; The aforementioned vehicle body structure sound insulation sample data is stored according to region category and layered material layer; The area categories include: floor, side walls, and ceiling. The layered material comprises: an inner layer, a core layer, an aluminum profile layer, and an outer cover layer. The aforementioned interior layers include: a floor fabric layer, an interior panel layer, and a damping layer. The core layers include: a first core layer to an eighth core layer. The aluminum profile layer includes: an inner damping layer, an aluminum profile layer, an outer damping layer, a cavity sound-absorbing layer, and a cavity damping layer. The outer cover layer includes: a first sound-absorbing layer, a second sound-absorbing layer, and an outer cover panel layer.

3. The method for predicting and optimizing the sound insulation performance of rail vehicles according to claim 1, characterized in that, The centralized and standardized management of sound insulation material parameters and vehicle body structure sound insulation sample data mentioned in step S01 specifically adopts a material library management unit and a sample library management unit; The material library management unit supports classification and management by material type, allows editing of material parameters through Excel files or software interfaces, automatically verifies the validity of parameter formats, stores materials in an in-memory database by category and creates an index table, and automatically counts the number of material types and samples. The sample library management unit supports the classification and management of samples by region. When entering samples, materials are selected in the order of inner layer, core layer, aluminum profile layer, and outer cover layer. The unit automatically calculates the total surface density and total thickness of the sample, extracts the sound insulation / sound absorption characteristics of each core layer material to generate feature values, and updates the feature library synchronously.

4. The method for predicting and optimizing the sound insulation performance of rail vehicles according to claim 1, characterized in that, The specific steps for data augmentation based on the initial sample as described in step S02 are as follows: extract the sound insulation feature value and sound absorption feature value of each layer in the core layer of the sample. The sound insulation feature value is calculated as STL_Value=M×Rw, where M is the areal density of the material layer and Rw is the weighted sound insulation of the material layer. The sound absorption feature value is calculated as SAC_Value=D×NRC, where D is the thickness of the material layer and NRC is the sound absorption coefficient of the material layer. Identify non-empty material layers, and generate virtual samples by randomly adjusting the positions of the non-empty material layers without changing the order of the non-empty material layers. Each original sample then generates several virtual samples to obtain the augmented sample set.

5. The method for predicting and optimizing the sound insulation performance of rail vehicles according to claim 1, characterized in that, The specific steps in step S02, which involve combining the feature weighting of the material layer location, determining the feature weights through the mRMR algorithm and the exponential decay weighting strategy, and finally training the sound insulation performance prediction model, are as follows: The position-weighted score of the material layer is calculated using the material layer position-weighted scoring algorithm, and the position-weighted feature is obtained. The expression is: Ins_score=0.8+0.4×cos(2π×dis)+0.2×(1-2×dis), where dis is the ratio of the distance from the center of the material layer to the boundary of the core layer to the core layer thickness limit. The mRMR algorithm is used to rank the features of the enhanced sample set by importance, and the weighted sample feature set is obtained by weighting each feature based on the exponential decay weight strategy. The weight calculation expression is: Weights_exp(i)=exp(-decay_rate×(rank_position-1)), where decay_rate is the decay rate and rank_position is the ranking position of the feature. The weighted sample feature set is used as input, and the weighted sound insulation Rw is used as the output target. The support vector machine regression algorithm is used for training. During training, the parameters C and g of the support vector machine regression algorithm are optimized by grid search, and the model is evaluated by 3-fold cross-validation. When the mean squared error (MSE) of the model on the training set and the test set is less than 0.1 and the correlation coefficient (R²) is greater than 0.9, the model is considered to have met the standard and is saved. At the same time, the fitting results of the model on the training set and the test set, as well as the MSE and R² evaluation indicators, are output.

6. The method for predicting and optimizing the sound insulation performance of rail vehicles according to claim 1, characterized in that, The specific steps of step S03 are as follows: The system specifies the area category and core layer thickness limit, combines materials by layer, and calculates the total thickness, total areal density and total parameters of the core layer material in real time. If the core layer thickness exceeds the limit, a prompt mechanism is triggered. Based on the loaded model and scheme parameters, the system calls the material layer location weighted scoring algorithm and the sound insulation performance prediction model to output the Rw prediction result. It supports scheme modification and re-prediction. After the prediction is completed, a Word format report containing material composition, parameter statistics and prediction results can be exported.

7. The method for predicting and optimizing the sound insulation performance of rail vehicles according to claim 1, characterized in that, The constraints mentioned in step S04 include: maximum limit for core layer thickness, maximum limit for overall thickness, maximum limit for core surface density, maximum limit for overall surface density, and minimum limit for weighted sound insulation.

8. The method for predicting and optimizing the sound insulation performance of rail vehicles according to claim 7, characterized in that, The optimal sound insulation scheme mentioned in step S04 includes: the scheme with the maximum sound insulation and the lightweight scheme that meets the sound insulation limit, wherein: Under the maximum sound insulation scheme, constraints on core layer thickness, overall thickness, core surface density, and overall surface density are allowed to be set; Under the lightweight scheme that meets the sound insulation limit, it is possible to set the minimum limit constraints for core layer thickness, overall thickness, and weighted sound insulation, including two sub-modes: lightest core layer weight and lightest overall weight.

9. The method for predicting and optimizing the sound insulation performance of rail vehicles according to claim 1, characterized in that, The specific content of the adaptive selection optimization algorithm described in step S04 is as follows: When the number of possible combinations is less than or equal to 50 million, an exhaustive algorithm is used to output the three-dimensional scatter plot of the solutions that meet the conditions and the global optimal solution. If the number of solutions that meet the conditions is greater than 1,000, then only the three-dimensional scatter plot of the first 1,000 solutions and the global optimal solution are output. When the number of possible solutions exceeds 50 million, a genetic algorithm is used, with a default iteration of 200 generations, outputting the optimization convergence curve and the global optimal solution.

10. A device for predicting and optimizing the sound insulation performance of rail vehicles, characterized in that, The apparatus implements the method as described in any one of claims 1 to 9, comprising: Database Management Module: Used to build and maintain a unified database for centralized and standardized management of sound insulation material parameters and vehicle body structure sound insulation sample data; Model training module: used for data augmentation based on initial samples, combined with feature weighting processing of material layer location, and then determining feature weights through mRMR algorithm and exponential decay weight strategy, finally training to obtain sound insulation performance prediction model; Solution Prediction Module: Used to predict and evaluate the sound insulation performance of user-defined material combination schemes using a database and sound insulation performance prediction model; Solution Optimization Module: Based on the user-defined solutions and constraints, and using a database and sound insulation performance prediction model, the module automatically searches for the optimal sound insulation solution through an adaptive optimization algorithm.

Images