A floating impact platform acceleration time sequence signal prediction method and system
Patent Information
- Application Number
- CN202611092415.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-22
- Publication Date
- 2026-11-03
AI Technical Summary
Traditional methods for assessing the impact environment of floating impact platforms suffer from limited measurement points, insufficient coverage of explosion conditions, high costs, and long cycles. Furthermore, traditional finite element simulations are difficult to perform rapid analysis of multiple operating conditions, and existing machine learning research focuses primarily on modeling frequency domain eigenvalues while neglecting time domain signal prediction.
The key parameters of the underwater explosion UNDEX model are optimized using the particle swarm optimization algorithm, a high-precision acoustic-structure interaction finite element benchmark model is established, a standardized impact response dataset is generated by batch modeling driven by parameterized scripts, and various machine learning prediction models are constructed. The models are trained and evaluated using the dataset to achieve high-precision and fast acceleration time series signal prediction.
It achieves high-precision and rapid prediction of acceleration time-series signals of floating impact platforms. The simulation response is in high agreement with the experimental data. It generates a standardized dataset covering multiple working conditions, reduces data acquisition costs, and improves prediction efficiency.
Smart Images
Figure CN123030864A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of impact environment of floating impact platforms, and specifically relates to a method and system for predicting acceleration time-series signals of floating impact platforms. Background Technology
[0002] The shock resistance of ship equipment is a key factor in assessing a ship's survivability, and its scientific and accurate evaluation and assessment have become an indispensable part of ship design. Floating impact platforms are important vehicles for evaluating the shock resistance of shipborne equipment. However, traditional floating platform impact physics tests require the construction of complex testing systems, which suffer from problems such as limited measurement points, insufficient coverage of explosive conditions, high cost, and long cycles; traditional finite element simulations also struggle to achieve rapid analysis under multiple conditions. In the field of theoretical evaluation, the impact spectrum method is currently the most widely used and classic theoretical tool for shock resistance analysis. However, due to the high dimensionality, strong nonlinearity, and time correlation of the time-domain signal of floating platform impact acceleration, it is difficult to make accurate predictions using traditional methods. Existing literature on machine learning prediction of floating platform impact response mainly focuses on the modeling and analysis of frequency domain eigenvalues, while the prediction of time-domain signals containing richer transient information is rarely addressed. Therefore, developing an efficient and reliable method for rapid prediction of the time-series signal of floating platform impact acceleration is particularly important and urgent. Summary of the Invention
[0003] The purpose of this invention is to provide a method and system for predicting the acceleration time-series signal of a floating impact platform, so as to achieve high-precision and rapid prediction of the acceleration time-domain signal of the floating platform under the same explosion and equipment operating conditions, and solve the problems of high cost and insufficient prediction accuracy of traditional test and simulation.
[0004] To address the aforementioned technical problems, this invention provides a method for predicting the acceleration time-series signal of a floating impact platform, comprising the following steps: S1. Establish an acoustic-structure coupled finite element benchmark model of the floating impact platform, and use the particle swarm optimization algorithm to optimize the key parameters of the underwater explosion UNDEX model to obtain a high-precision benchmark model. S2. Based on the high-precision benchmark model, batch modeling is driven by parameterized scripts, and a standardized impact response dataset is generated using multiple working condition parameters as variables. S3. Construct multiple machine learning prediction models, wherein the input parameters of the multiple machine learning prediction models are the working condition parameters, and the output is the acceleration time series signal of typical measuring points of the floating impact platform; S4. The standardized impact response dataset is used to train and evaluate the various machine learning prediction models to determine the prediction model, which is used to output the acceleration time series signal prediction result of the floating impact platform based on the input target working condition parameters.
[0005] According to the above scheme, the key parameters of the underwater explosion UNDEX model optimized using the particle swarm optimization algorithm include: Set the population size, number of iterations, learning factor, and inertia weight for the particle swarm optimization algorithm, wherein the inertia weight adopts a linear decreasing strategy; Underwater explosion simulation was performed using the key parameters of the UNDEX model corresponding to each particle in the particle swarm optimization algorithm. The sum of squared relative errors of the peak acceleration between the simulation results and the experimental values was used as the fitness function, and the fitness value of each particle was calculated. The individual optimal position and global optimal position of the particles are updated based on the fitness value, and the velocity and position of each particle are updated. The key parameters of the optimal underwater explosion UNDEX model are obtained through iterative optimization.
[0006] According to the above scheme, the step of driving batch modeling through parameterized scripts includes: By parsing the command log files of the finite element software using Python scripts, a script architecture is generated, an iterative loop structure is embedded, and key parameters are dynamically corrected to achieve parametric batch modeling.
[0007] According to the above scheme, the multiple operating parameters include explosive equivalent, detonation distance, angle of attack, azimuth angle, equipment mass, and spring installation frequency.
[0008] According to the above scheme, generating the standardized shock response dataset includes: Multiple sets of parameter combinations are generated in the mid-to-far field explosion domain of the multiple operating parameters using an equidistant sampling strategy. Through parameterized batch modeling and automatic post-processing, acceleration response data are extracted to form a standardized impact response dataset.
[0009] According to the above scheme, the various machine learning prediction models include the random forest model, the long short-term memory network model, and the gradient boosting machine model.
[0010] According to the above scheme, training and evaluating the various machine learning prediction models to determine the prediction model includes: The coefficient of determination was used as an evaluation index to assess the performance of each model, and the model with the highest coefficient of determination was selected as the forecast model.
[0011] According to the above scheme, the step of outputting the acceleration time-series signal prediction result of the floating impact platform based on the input target working condition parameters includes: Input the target operating condition parameters into the prediction model, and output the acceleration time series signal prediction curve of typical measuring points of the floating impact platform.
[0012] According to the above scheme, the establishment of the acoustic-structure interaction finite element reference model of the floating impact platform includes: A three-dimensional finite element model of the floating impact platform and the watershed is established. The watershed adopts an acoustic-structure coupling element, and the watershed and the floating impact platform are coupled by a tie connection.
[0013] The present invention also provides a system for predicting the acceleration timing signal of a floating impact platform, comprising: The benchmark model establishment module is used to establish an acoustic-structure coupled finite element benchmark model of the floating impact platform, and to optimize the key parameters of the underwater explosion UNDEX model using the particle swarm optimization algorithm to obtain a high-precision benchmark model. The dataset generation module is used to generate a standardized impact response dataset based on the high-precision benchmark model by driving batch modeling through parameterized scripts, using multiple working condition parameters as variables. The model building module is used to build various machine learning prediction models. The input parameters of the various machine learning prediction models are the working condition parameters, and the output is the acceleration time series signal of typical measurement points of the floating impact platform. The training and prediction module is used to train and evaluate the various machine learning prediction models using the standardized impact response dataset, determine the prediction model, and output the acceleration time series signal prediction result of the floating impact platform based on the input target working condition parameters.
[0014] Beneficial effects This invention optimizes underwater explosion model parameters using a particle swarm optimization algorithm to obtain a high-precision acoustic-structure interaction finite element benchmark model. This ensures a high degree of agreement between the simulation response and experimental data, fundamentally guaranteeing the accuracy of subsequent forecast data sources. Based on this, parameterized scripts drive batch simulations, automatically generating a standardized impact response dataset covering multi-dimensional operating conditions. This solves the problems of high data acquisition costs and scarce samples associated with traditional methods. Using this dataset, various machine learning models are trained and evaluated. By directly mapping operating condition parameters to complete acceleration time-series signals, the final forecast model only requires the input of the target operating condition to instantly output the impact response curve, achieving high-precision, high-efficiency, and rapid forecasting of the impact environment of floating impact platforms under all operating conditions. Attached Figure Description
[0015] Figure 1 This is a schematic diagram of a three-dimensional finite element model of the floating impact platform of the present invention (including the floating platform structure, flow domain and mesh generation). Figure 2 This is a graph showing the fitness iteration convergence curve of the PSO particle swarm algorithm of this invention. Figure 3 This is a flowchart of the parametric modeling process of the present invention; Figure 4 A schematic diagram defining the key parameters of this invention (blast distance, angle of attack, azimuth). Figure 5This is a comparison chart of the predicted time-domain curve and the simulated time-domain curve of the RF model of this invention under a certain working condition; Figure 6 This is a comparison chart of the predicted time-domain curve and the simulated time-domain curve of the LSTM model of this invention under a certain working condition; Figure 7 This is a comparison chart of the predicted time-domain curve and the simulated time-domain curve of the GBM model under a certain working condition. Figure 8 This is a comparison diagram of the impact spectra under simulated and experimental conditions of this invention. Detailed Implementation
[0016] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure 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 this disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.
[0017] The core concept of this invention is to address the problems of traditional underwater explosion impact environment prediction methods relying on empirical formulas and having poor prediction accuracy, and high-precision numerical simulation methods being computationally expensive and unable to meet the needs of rapid prediction under multiple operating conditions. This invention proposes a machine learning-based method for rapid and accurate prediction of acceleration time-series signals of floating impact platforms. This method first optimizes the key parameters of the underwater explosion UNDEX model using a particle swarm optimization algorithm to establish a high-precision acoustic-structure interaction finite element baseline model of the floating impact platform. Then, it uses parameterized scripts to drive batch modeling and calculation, generating a standardized impact response dataset covering multiple operating conditions. Finally, it trains and evaluates various machine learning models based on this dataset, selecting the optimal model for actual prediction, thereby achieving high-precision, high-efficiency, and rapid prediction of the impact environment of floating impact platforms under different operating conditions.
[0018] Specifically, the present invention provides a method for predicting the acceleration time series signal of a floating impact platform, comprising the following steps: S1. Establish an acoustic-structure coupled finite element benchmark model of the floating impact platform, and use the particle swarm optimization algorithm to optimize the key parameters of the underwater explosion UNDEX model to obtain a high-precision benchmark model.
[0019] In this step, the modeling process first establishes a three-dimensional finite element model of the floating impact platform and the watershed. Details of the overall three-dimensional finite element mesh of the floating impact platform and the watershed layout can be found in [link to relevant documentation]. Figure 1The floating impact platform structure comprises major components such as bulkheads, a double bottom, longitudinal / transverse bulkheads, equipment bases, and reinforcing ribs, which are discretized using Lagrange elements. The external water area is constructed as a hemispherical flow domain with a radius of 20m, discretized using AC3D8R acoustic-structure interaction elements to effectively capture the reflection, transmission, and cavitation effects of the shock wave at the fluid-structure interface. A tie connection is used to couple the flow domain to ensure the coordinated transmission of displacement and pressure between the fluid-structure interface. Through this method, an acoustic-structure interaction finite element reference model is constructed that can realistically reflect the interaction between the underwater explosion shock wave and the platform structure. In this embodiment, typical measurement points on the floating impact platform, namely nodes 2739 and 2952, are selected as the locations for subsequent extraction and analysis of impact acceleration data.
[0020] To ensure that the simulation model's response closely approximates the actual experimental results, this invention employs a particle swarm optimization algorithm to optimize key parameters of the underwater explosion UNDEX model. The specific optimization process is as follows: The particle swarm optimization (PSO) algorithm was configured with fixed hyperparameters: population size of 20, number of iterations of 30, learning factors c1=c2=2, and inertia weights decreasing linearly from an initial value of 0.9 to 0.4. Underwater explosion simulations were performed using the key parameters of the underwater explosion UNDEX model corresponding to each particle in the PSO algorithm. The sum of squared relative errors of the peak acceleration between the simulation results and experimental values was used as the fitness function, and the fitness value of each particle was calculated. The key parameters here are the Abaqus underwater explosion UNDEX model constants K, k, A, B, and Kc, which are model input parameters that directly affect the amplitude and attenuation characteristics of the shock wave load. Based on these fitness values, the individual optimal position and global optimal position of each particle were updated, along with the velocity and position of each particle. Through iterative optimization, the optimal key parameters of the underwater explosion UNDEX model were obtained when the fitness value converged to a minimum. Substituting these optimal parameters into the baseline model yielded a high-precision baseline model that closely matched the experimental data. After iterative optimization, the simulation and experimental error of the impact spectrum velocity at node 2952 is 5.62%, and the error at node 2739 is 3.75%. The model accuracy meets engineering requirements, and the iterative convergence process is as follows: Figure 2 The particle swarm optimization fitness convergence curve is shown; a comparison between the optimized simulated impact spectrum and the experimental impact spectrum is shown below. Figure 8 As shown, the high accuracy of the benchmark model can be intuitively verified.
[0021] S2. Based on the high-precision benchmark model, batch modeling is driven by parameterized scripts, and a standardized impact response dataset is generated using multiple operating condition parameters as variables.
[0022] In this step, to automate batch simulation calculations, batch modeling is first driven by parametric scripts. Specifically, a Python script parses the command log file of the finite element software Abaqus, automatically identifying and extracting key modeling and solving instructions to generate a script architecture. Based on this, an iterative loop structure is embedded, and key parameters, such as explosive equivalent and detonation distance, are dynamically adjusted to achieve parametric batch modeling. The complete automated process is as follows: Figure 3 The flowchart for parametric modeling is shown below.
[0023] Taking a typical working condition as an example, the various working condition parameters covered include explosive equivalent, detonation distance, angle of attack, azimuth, equipment mass, and spring installation frequency. The equidistant sampling ranges for each parameter within the working condition space are specifically set as follows: explosive equivalent: 20kg, 25kg, 30kg, 33kg, 40kg, 50kg, 55kg (7 ranges); detonation distance: 5m, 8m, 10m, 12m, 15m, 18m, 20m (7 ranges); angle of attack: 0°, 15°, 30°, 45°, 60°, 75° (6 ranges); azimuth: 0°, 45°, 90°, 135°, 180°, 225°, 270°, 315° (8 ranges); equipment mass: 3000kg, 5000kg, 7000kg (3 ranges); spring installation frequency: 3Hz, 6Hz, 9Hz (3 ranges). The definitions of the above geometric parameters can be found in [reference needed]. Figure 4 A diagram illustrating the definition of key parameters. By combining all parameters, a total of 9800 sets of working condition samples were generated, achieving uniform coverage of the multi-dimensional parameter space and providing sufficient sample size for subsequent data-driven modeling.
[0024] The strategy for generating the standardized shock response dataset is as follows: Within the mid-to-far field explosion conditions defined by the aforementioned multi-parameter combinations, underwater explosion simulation calculations are performed for each parameter combination, and acceleration response data for nodes 2739 and 2952 are extracted using an automatic post-processing program. These complete time-series data are then mapped one-to-one with the corresponding operating parameters to form a standardized shock response dataset, which serves as a sample library for subsequent machine learning model training.
[0025] S3. Construct multiple machine learning prediction models, wherein the input parameters of the multiple machine learning prediction models are the working condition parameters, and the output is the acceleration time series signal of typical measuring points of the floating impact platform.
[0026] In this step, to fully explore the temporal features in the data and verify the applicability of different algorithms, several machine learning prediction models were constructed. These models specifically include a random forest model, a long short-term memory network model, and a gradient booster model. The random forest model possesses good anti-overfitting capabilities and is suitable for robust regression under multi-dimensional parameter inputs; the long short-term memory network model excels at capturing long-term dependencies in time series and can accurately reproduce the temporal evolution of acceleration signals; the gradient booster model fits complex nonlinear relationships by gradually reducing residuals, exhibiting high prediction accuracy. The input parameters for these models are all the aforementioned operating condition parameters, namely, explosive equivalent, detonation distance, angle of attack, azimuth angle, equipment mass, and spring installation frequency. The model output is the complete acceleration time-series signal of typical measuring points on the floating impact platform, i.e., an acceleration curve that varies with time, rather than a single peak or characteristic value. The acceleration time-series prediction curves of the three models under a single operating condition are compared with the true values from finite element simulations. Figure 5 , Figure 6 and Figure 7 As shown, the time-domain fitting effects of different algorithms can be compared intuitively.
[0027] S4. The standardized impact response dataset is used to train and evaluate the various machine learning prediction models to determine the prediction model, which is used to output the acceleration time series signal prediction result of the floating impact platform based on the input target working condition parameters.
[0028] This step involves model training, evaluation, and final application. First, the Random Forest model, Long Short-Term Memory (LSTM) network model, and Gradient Boosting Machine (GPHM) model are trained using the 9800 standardized impact response datasets generated in step S2. After training, the coefficient of determination (R²) is used as an evaluation metric to objectively measure the predictive ability of each model. Through dataset training and validation, the R² of the Random Forest, LSM, and GPHM models are all higher than 0.9, with the GPHM model exhibiting the best fitting performance, reaching R² of 0.99, demonstrating extremely high prediction accuracy. These three models can be selected in engineering projects based on the differences in computational speed and accuracy requirements. Finally, in the prediction application stage, any new set of target operating condition parameters not included in the training is input into the prediction model. The model can then quickly calculate and output the predicted acceleration time-series signal curves for typical measurement points of the floating impact platform, completing the rapid prediction task of the impact environment.
[0029] This invention's entire method possesses three significant technical advantages: First, by automatically calibrating key parameters of the Abaqus underwater explosion UNDEX model using the PSO particle swarm optimization algorithm, it solves the problem of parameter calibration for strongly nonlinear underwater explosion simulations. After optimization, the maximum error between the simulated impact spectrum velocity and the experimental result is only 5.62%, successfully constructing a high-precision finite element benchmark model. Second, relying on Python secondary development, it achieves full automation of modeling, batch calculation, and post-processing, generating 9800 sets of standardized acceleration datasets uniformly covering all operating conditions, breaking through the bottlenecks of high cost and insufficient sample size in traditional physical experiments and single-condition simulations. Third, the constructed multi-machine learning model exhibits excellent prediction accuracy, with all models having R² > 0.9, and the optimal GBM model reaching R² of 0.99. After model training, only operating condition parameters need to be input to instantly output acceleration time history curves, significantly reducing the time cost of traditional numerical simulations and providing a reliable and efficient technical means for digital testing platforms for ship impact resistance.
[0030] The present invention also provides a system for predicting the acceleration time series signal of a floating impact platform, which corresponds to the above-mentioned prediction method and includes: The benchmark model establishment module is used to establish an acoustic-structure coupled finite element benchmark model of the floating impact platform, and to optimize the key parameters of the underwater explosion UNDEX model using the particle swarm optimization algorithm to obtain a high-precision benchmark model. The dataset generation module is used to generate a standardized impact response dataset based on the high-precision benchmark model by driving batch modeling through parameterized scripts, using multiple working condition parameters as variables. The model building module is used to build various machine learning prediction models. The input parameters of the various machine learning prediction models are the working condition parameters, and the output is the acceleration time series signal of typical measurement points of the floating impact platform. The training and prediction module is used to train and evaluate the various machine learning prediction models using the standardized impact response dataset, determine the prediction model, and output the acceleration time series signal prediction result of the floating impact platform based on the input target working condition parameters.
[0031] It should be noted that, depending on the implementation needs, the various steps / components described in this application can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.
[0032] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for predicting the acceleration time-series signal of a floating impact platform, characterized in that, Includes the following steps: S1. Establish an acoustic-structure coupled finite element benchmark model of the floating impact platform, and use the particle swarm optimization algorithm to optimize the key parameters of the underwater explosion UNDEX model to obtain a high-precision benchmark model. S2. Based on the high-precision benchmark model, batch modeling is driven by parameterized scripts, and a standardized impact response dataset is generated using multiple working condition parameters as variables. S3. Construct multiple machine learning prediction models, wherein the input parameters of the multiple machine learning prediction models are the working condition parameters, and the output is the acceleration time series signal of typical measuring points of the floating impact platform; S4. The standardized impact response dataset is used to train and evaluate the various machine learning prediction models to determine the prediction model, which is used to output the acceleration time series signal prediction result of the floating impact platform based on the input target working condition parameters.
2. The method for predicting acceleration time-series signals of a floating impact platform according to claim 1, characterized in that, The key parameters of the UNDEX model for underwater explosions, optimized using the particle swarm optimization algorithm, include: Set the population size, number of iterations, learning factor, and inertia weight for the particle swarm optimization algorithm, wherein the inertia weight adopts a linear decreasing strategy; Underwater explosion simulation was performed using the key parameters of the UNDEX model corresponding to each particle in the particle swarm optimization algorithm. The sum of squared relative errors of the peak acceleration between the simulation results and the experimental values was used as the fitness function, and the fitness value of each particle was calculated. The individual optimal position and global optimal position of the particles are updated based on the fitness value, and the velocity and position of each particle are updated. The key parameters of the optimal underwater explosion UNDEX model are obtained through iterative optimization.
3. The method for predicting acceleration time-series signals of a floating impact platform according to claim 1, characterized in that, The method of driving batch modeling through parameterized scripts includes: By parsing the command log files of the finite element software using Python scripts, a script architecture is generated, an iterative loop structure is embedded, and key parameters are dynamically corrected to achieve parametric batch modeling.
4. The method for predicting acceleration time-series signals of a floating impact platform according to claim 1, characterized in that, The various operating parameters include explosive equivalent, detonation distance, angle of attack, azimuth, equipment mass, and spring installation frequency.
5. The method for predicting acceleration time-series signals of a floating impact platform according to claim 1, characterized in that, The generation of the standardized shock response dataset includes: Multiple sets of parameter combinations are generated in the mid-to-far field explosion domain of the multiple operating parameters using an equidistant sampling strategy. Through parameterized batch modeling and automatic post-processing, acceleration response data are extracted to form a standardized impact response dataset.
6. The method for predicting acceleration time-series signals of a floating impact platform according to claim 1, characterized in that, The various machine learning prediction models include the random forest model, the long short-term memory network model, and the gradient booster model.
7. The method for predicting acceleration time-series signals of a floating impact platform according to claim 1, characterized in that, The process of training and evaluating the various machine learning prediction models to determine the prediction model includes: The coefficient of determination was used as an evaluation index to assess the performance of each model, and the model with the highest coefficient of determination was selected as the forecast model.
8. The method for predicting acceleration time-series signals of a floating impact platform according to claim 1, characterized in that, The method for outputting the acceleration time-series signal prediction result of the floating impact platform based on the input target working condition parameters includes: Input the target operating condition parameters into the prediction model, and output the acceleration time series signal prediction curve of typical measuring points of the floating impact platform.
9. The method for predicting acceleration time-series signals of a floating impact platform according to claim 1, characterized in that, The establishment of the acoustic-structure coupled finite element benchmark model for the floating impact platform includes: A three-dimensional finite element model of the floating impact platform and the watershed is established. The watershed adopts an acoustic-structure coupling element, and the watershed and the floating impact platform are coupled by a tie connection.
10. A system for predicting the acceleration timing signal of a floating impact platform, characterized in that, include: The benchmark model establishment module is used to establish an acoustic-structure coupled finite element benchmark model of the floating impact platform, and to optimize the key parameters of the underwater explosion UNDEX model using the particle swarm optimization algorithm to obtain a high-precision benchmark model. The dataset generation module is used to generate a standardized impact response dataset based on the high-precision benchmark model by driving batch modeling through parameterized scripts, using multiple working condition parameters as variables. The model building module is used to build various machine learning prediction models. The input parameters of the various machine learning prediction models are the working condition parameters, and the output is the acceleration time series signal of typical measurement points of the floating impact platform. The training and prediction module is used to train and evaluate the various machine learning prediction models using the standardized impact response dataset, determine the prediction model, and output the acceleration time series signal prediction result of the floating impact platform based on the input target working condition parameters.