Method for predicting and optimizing underwater load and motion trail of navigation body based on artificial intelligence

Abstract

The invention discloses a navigation body water entry load and motion trail prediction and optimization method based on artificial intelligence, and relates to the field of cross-medium water entry and exit of navigation bodies. The problems that an existing real-scale navigation body cross-medium water entry test is huge in cost, a small-scale mechanical test and other methods are adopted for the cross-medium water entry problem, but large-scale development is difficult due to high cost, and the numerical simulation calculation efficiency is low are solved. The invention discloses a deep learning-based slamming load and motion track prediction method when a navigation body enters water at a high speed, and the method comprises the steps: building a navigation body water entry slamming acceleration and motion attitude database through finite element, meshless and other numerical calculation results or existing test results; and predicting the underwater slamming load characteristics and motion trails of the navigation body under different section shape parameters and underwater parameters, and on the basis, introducing a genetic algorithm to perform reverse optimization design on the shape parameters and underwater parameters of the navigation body by combining the established model, so as to obtain the underwater slamming load characteristics and motion trails of the navigation body. And optimal shape parameters and water entry parameters with strong ballistic stability and small slamming acceleration can be quickly given.

Description

Technical Field

[0001] The present invention relates to the technical field of a navigation body entering and exiting water across a medium, and in particular to a method for predicting and optimizing a navigation body's water entry load and motion trajectory based on artificial intelligence. Background Art

[0002] The study of load characteristics and ballistic stability during the process of a vehicle entering water across a medium has important engineering application value. Due to the huge cost of the full-scale vehicle entering water across a medium test and the restrictions of safety and other conditions, the current problem of entering water across a medium mainly uses finite element or meshless methods to carry out numerical calculations and small-scale mechanistic experiments. However, small-scale scaled-down tests are difficult to carry out on a large scale due to the high cost, and numerical simulations have problems such as low computational efficiency. That is, the above methods cannot efficiently give the acceleration and motion trajectory time history curves during the vehicle entering water. At the same time, in order to ensure that the vehicle maintains the best ballistic stability and reduces the impact load during the vehicle crossing the medium, the above methods are also difficult to optimize the shape parameters and water entry parameters of the vehicle. Summary of the invention

[0003] In order to solve the problem in the prior art that the actual-scale water entry test of a real-scale navigation body across a medium is extremely costly and is limited by safety conditions, the present invention currently mainly uses finite element or meshless methods to carry out numerical calculations and small-scale mechanistic tests on the problem of water entry across a medium. In addition, small-scale scaled-down tests are difficult to carry out on a large scale due to their high cost, and numerical simulations have problems such as low computational efficiency, making it difficult to quickly give the load and motion trajectory of the navigation body after entering the water. In terms of navigation body design, how to quickly and accurately determine shape parameters and water entry parameters is also an extremely thorny issue. In order to solve the above technical problems, the present invention is implemented through the following technical solutions:

[0004] Solution 1: The present invention proposes a method for predicting and optimizing the water entry load and motion trajectory of a navigation body based on artificial intelligence, and the method comprises the following steps:

[0005] Step 1: Use finite element, meshless or experimental methods to perform numerical simulation or experimental measurement on the process of the navigation body entering the water, obtain the load characteristics and motion trajectory data during the navigation body entering the water, and obtain a sample database of the acceleration and position of the navigation body under different initial conditions by changing the four parameters of the navigation body cross-sectional ratio s, semi-cone angle α, entry speed v and entry angle β;

[0006] Step 2: Based on the sample database established in step 1, the database is divided into a training set and a validation set, and a neural network rapid prediction model for predicting the acceleration history curve and motion trajectory of the vehicle under different initial parameters is established respectively;

[0007] Step 3: The neural network rapid prediction model of the acceleration history curve and motion trajectory of the vehicle under different initial parameters obtained in step 2 is embedded into the genetic algorithm, the shape parameters and water entry parameters of the vehicle are optimized, and the water entry load and motion trajectory of the vehicle are predicted and optimized.

[0008] Further, a preferred implementation manner is provided, in which the method for optimizing the shape parameters and water entry parameters of the navigation body in step 3 is:

[0009] Input the upper and lower bounds of the four parameters of the navigation body cross-sectional ratio s, semi-cone angle α, water entry speed v and water entry angle β, randomly sample the parameter combinations in the space formed by the upper and lower bounds to form an initial population, and calculate the acceleration and ballistic trajectory based on the acceleration and motion trajectory neural network rapid prediction model established in step 2;

[0010] All parameter combinations in the formed initial population are evaluated and screened, crossover and mutation are performed on the screened population, the search scale is expanded and a new generation of individuals is generated, the parameters are iterated and gradually optimized, and finally the best parameters that meet the above optimization objectives are output.

[0011] Furthermore, a preferred implementation is provided, in which the ratio of the training set to the validation set in step 2 is no less than 3:4.

[0012] Furthermore, a preferred implementation is provided, in which the input variables of the neural network rapid prediction model for predicting the acceleration time history curve and motion trajectory of the vehicle under different initial parameters described in step 2 are the four parameters in step 1, and the output variable is the acceleration time history data, and the load characteristics and motion trajectory data during the vehicle entering the water are normalized to the minimum and maximum values.

[0013] Furthermore, a preferred embodiment is provided, in which the neural network rapid prediction model of the vehicle acceleration history curve and motion trajectory includes an input layer, an output layer and multiple hidden layers, and the input layer variables are the vehicle cross-sectional ratio s, semi-cone angle α, entry speed v and entry angle β.

[0014] Furthermore, a preferred implementation is provided, wherein the hyperparameters contained in the input layer, the output layer and the multiple hidden layers are optimized by methods such as grid search or Bayesian search; similarly, the hyperparameters in the genetic algorithm are optimized by methods such as grid search or Bayesian search.

[0015] Furthermore, a preferred embodiment is provided, in which the output layer of the neural network rapid prediction model of the vehicle acceleration time history curve described in step 2 is the vehicle acceleration time history data.

[0016] Furthermore, a preferred embodiment is provided, in which the output layer of the neural network rapid prediction model of the motion trajectory is the time history data of the position coordinates of the vehicle.

[0017] Solution 2: A computer device includes a memory and a processor, wherein the memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes any one of the methods described in Solution 1.

[0018] Solution three: a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of Solution one.

[0019] The present invention is beneficial in that:

[0020] The method described in the present invention uses a deep learning method to conduct in-depth analysis and mining of the cross-medium acceleration load characteristics and motion trajectory results of the vehicle obtained by finite element, gridless or experimental methods, and determines the optimal hyperparameters of the fully connected neural network (such as the number of hidden layers, the number of neurons, etc.) according to methods such as grid search or Bayesian search, and establishes deep neural network models for predicting the acceleration and motion trajectory of the vehicle according to the prediction targets; similarly, the hyperparameters of the genetic algorithm (such as the initial population size, the number of population iterations, etc.) can also be determined by methods such as grid search or Bayesian search.

[0021] When the database is collected based on the experimental method, the acceleration or displacement data is not ideal due to the limitation of the sampling equipment, and when non-equally spaced and non-equally long data appear, the output layer length of the fully connected neural network is inconsistent and cannot be used directly. The present invention adopts a good data processing method. For non-equally spaced time series data, the original data can be processed by linear or cubic spline interpolation, and then resampled based on equal time intervals, and then equal-spaced time series data is generated. For non-equal time series data, the original time can be divided by each movement time at the same time to form equal-length time series data. For the movement time, a fully connected neural network can be established separately to predict the movement time.

[0022] The method described in the present invention aims to achieve rapid prediction of the acceleration and motion trajectory of a navigation body during its entry into water, reduce the cost of experiments and numerical simulations, and improve prediction efficiency, while providing an automated and efficient method for optimizing the shape parameters and entry parameters of a navigation body.

[0023] The present invention introduces a genetic algorithm in the parameter space of the shape parameters and the water entry parameters of the navigation body, and based on the optimization goals of stable motion trajectory and small impact acceleration, uses the fully connected neural network established in step 2 for prediction under different initial parameters, thereby realizing the optimal design of the shape parameters and the water entry parameters of the navigation body, thereby saving calculation time and assisting in the design of the navigation body cross-section and the water entry parameters, and has high engineering application value.

[0024] The present invention can provide more intuitive information on the motion trajectory of the projectile by directly predicting the ballistic displacement, which helps to quickly understand the water entry process and trajectory change characteristics of the projectile, and is closer to application needs. In particular, in scenarios where an accurate description of the water entry trajectory is required, such as hit accuracy evaluation, more direct and engineering-significant results can be provided.

[0025] The present invention combines multiple objective functions into a single objective by weighting, making the optimization process more concise and suitable for direct solution under the framework of standard genetic algorithm. By setting the weight of the objective function, the importance of different objectives can be flexibly adjusted, avoiding the complexity of the Pareto optimal solution set in multi-objective optimization; avoiding the computational overhead of calculating the dominance relationship and maintaining the multi-objective solution set in the prior art. The optimization process is simplified and suitable for the situation where a certain objective is optimized first in engineering.

[0026] By adjusting the weights, the present invention allows users to flexibly control the relative importance of each goal according to actual needs, which enables the algorithm to better meet the priority requirements of specific applications. In some cases, only one of the goals may need to be optimized, and the weighted merging method can more directly and efficiently find the optimal solution.

[0027] The present invention is also applicable to the fields of impact load and motion trajectory prediction during the water entry of a navigation body and optimization design of the cross-sectional shape parameters and water entry parameters of the navigation body. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flowchart for quickly predicting the acceleration and motion trajectory of a navigation body entering water as described in Implementation Method 1.

[0029] Figure 2 This is a schematic diagram of the water entry process of the navigation body described in Implementation Method 11.

[0030] Figure 3 This is a schematic diagram of the rapid prediction results of the acceleration and motion trajectory of the navigation body during the water entry process described in Implementation Example 11.

[0031] Figure 3In the figure, (a) is a schematic diagram for comparing the error between the predicted time history curve results of the offset during the water entry and the test results; (b) is a schematic diagram for comparing the error between the predicted time history curve results of the horizontal displacement during the water entry and the test results; (c) is a schematic diagram for comparing the error between the predicted time history curve results of the vertical displacement during the water entry and the test results; (d) is a schematic diagram for comparing the error between the predicted time history curve results of the offset during the water entry and the test results.

[0032] Figure 4 This is a schematic diagram of the optimization target and fitness change curve in the navigation body parameter optimization design described in Implementation Method 11.

[0033] Figure 5 This is a schematic diagram of the results of the optimized design of the navigation body parameters described in Implementation Method 11. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the implementation methods of the present application clearer, the technical solutions in the implementation methods of the present application will be clearly and completely described below in conjunction with the drawings in the implementation methods of the present application. Obviously, the described implementation methods are only part of the implementation methods of the present application, not all of the implementation methods.

[0035] Embodiment 1: This embodiment provides a method for predicting and optimizing the water entry load and motion trajectory of a navigation body based on artificial intelligence, and the method comprises the following steps:

[0036] Step 1: Use finite element, meshless or experimental methods to perform numerical simulation or experimental measurement on the process of the navigation body entering the water, obtain the load characteristics and motion trajectory data during the navigation body entering the water, and obtain a sample database of the acceleration and position of the navigation body under different initial conditions by changing the four parameters of the navigation body cross-sectional ratio s, semi-cone angle α, entry speed v and entry angle β;

[0037] Step 2: Based on the sample database established in step 1, the database is divided into a training set and a validation set, and a neural network rapid prediction model for predicting the acceleration history curve and motion trajectory of the vehicle under different initial parameters is established respectively;

[0038] Step 3: The neural network rapid prediction model of the acceleration history curve and motion trajectory of the vehicle under different initial parameters obtained in step 2 is embedded into the genetic algorithm, the shape parameters and water entry parameters of the vehicle are optimized, and the water entry load and motion trajectory of the vehicle are predicted and optimized.

[0039] Implementation method 2: This implementation method is a further limitation of the method for predicting and optimizing the water entry load and motion trajectory of a navigation body based on artificial intelligence described in implementation method 1. The method for optimizing the shape parameters and water entry parameters of the navigation body in step 3 is:

[0040] Input the upper and lower bounds of the four parameters of the navigation body cross-sectional ratio s, semi-cone angle α, water entry speed v and water entry angle β, randomly sample the parameter combinations in the space formed by the upper and lower bounds to form an initial population, and calculate the acceleration and ballistic trajectory based on the acceleration and motion trajectory neural network rapid prediction model established in step 2;

[0041] All parameter combinations in the initial population formed in step 3 are evaluated and screened, crossover and mutation processing is performed on the screened population, the search scale is expanded and a new generation of individuals is generated, the parameters are iterated and gradually optimized, and finally the best parameters that meet the above optimization objectives are output.

[0042] Implementation method three: This implementation method is a further limitation of the artificial intelligence-based method for predicting and optimizing the water entry load and motion trajectory of a navigation body described in implementation method one. The ratio of the training set and the verification set described in step 2 is no less than 3:4.

[0043] Implementation method 4. This implementation method is a further limitation of the artificial intelligence-based method for predicting and optimizing the water entry load and motion trajectory of a vehicle described in implementation method 1. The input variables of the neural network rapid prediction model for predicting the acceleration time history curve and motion trajectory of the vehicle under different initial parameters described in step 2 are the four parameters in step 1, and the output variable is the acceleration time history data. The load characteristics and motion trajectory data during the water entry process of the vehicle are normalized to the minimum and maximum values.

[0044] Implementation method five. This implementation method is a further limitation of the artificial intelligence-based method for predicting and optimizing the water entry load and motion trajectory of a vehicle described in implementation method five. The neural network rapid prediction model of the vehicle acceleration history curve and motion trajectory includes an input layer, an output layer and multiple hidden layers, and the input layer variables are all the vehicle cross-sectional ratio s, semi-cone angle α, water entry speed v and water entry angle β.

[0045] Implementation method six. This implementation method is a further limitation of the artificial intelligence-based method for predicting and optimizing the water entry load and motion trajectory of a navigation body described in implementation method five. The hyperparameters contained in the input layer, output layer and multiple hidden layers (such as the number of hidden layers, the number of neurons, etc.) are obtained by optimizing through methods such as grid search or Bayesian search; similarly, the hyperparameters in the genetic algorithm (such as the initial population size, the number of population iterations, etc.) can also be optimized through methods such as grid search or Bayesian search.

[0046] Implementation method seven: This implementation method is a further limitation of the artificial intelligence-based method for predicting and optimizing the water entry load and motion trajectory of a vehicle described in implementation method one. The output layer of the neural network rapid prediction model of the vehicle acceleration time history curve described in step 2 is the vehicle acceleration time history data.

[0047] Implementation method eight. Implementation method is a further limitation of the artificial intelligence-based method for predicting and optimizing the water entry load and motion trajectory of a navigation body described in implementation method one. The output layer of the neural network rapid prediction model of the motion trajectory is the time history data of the navigation body's position coordinates.

[0048] Furthermore, a preferred implementation is provided. When the acceleration history curve and motion trajectory training data of the vehicle under different initial parameters in step 2 are non-equally spaced time series data, the original data can be processed by linear or cubic spline interpolation, and then resampled based on equal time intervals to generate equal time series data.

[0049] Furthermore, a preferred implementation is provided. When the acceleration time history curve and motion trajectory training data of the vehicle under different initial parameters in step 2 are non-equal time series data, the original time can be simultaneously divided by the respective motion time to form equal-length time series data. For the motion time, a fully connected neural network can be separately established to predict the motion time.

[0050] Embodiment 9. This embodiment proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method described in any one of embodiments 1 to 8.

[0051] Embodiment 10: This embodiment proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the method described in any one of Embodiments 1 to 8 are implemented.

[0052] Implementation eleven: This implementation provides an example, which is used to explain the above implementations one to eight. The specific example is as follows:

[0053] See also Figures 1 to 5 To illustrate this embodiment, the method described in this embodiment includes the following steps:

[0054] Taking the process of a small-scale navigation body entering water across a medium as an example, the same principle applies to the real scale. Figure 1 The implementation steps of the present invention are described as follows:

[0055] Step 1: First, numerically simulate or test the process of the navigation body entering the water using finite element, meshless or experimental methods, and collect the acceleration and motion trajectory data of the navigation body under different cross-sectional ratios s, semi-cone angles α, entry speeds v and entry angles β. Take the initial entry center as the center point, and define the x-axis passing through the center point parallel to the initial water surface to the right, and the y-axis passing through the center point perpendicular to the initial water surface and pointing to the free liquid surface as the positive direction, such as Figure 2 As shown. For the non-equally spaced data obtained, linear interpolation and equal-interval resampling methods can be used to obtain acceleration sequence data and motion trajectory sequence data at equal time intervals, including the horizontal direction dx and the vertical direction dy.

[0056] On the basis of the above, by changing any variables in the cross-sectional ratio s, the semi-cone angle α, the water entry speed v and the water entry angle β, the water entry process of the navigation body is simulated or tested multiple times, and the acceleration time series data and motion trajectory data of the navigation body at different times are recorded to establish a sample database. Then, the sample database is divided into a training set and a test set, in which the number of samples in the training set accounts for no less than 3 / 4 of the total number of samples; at the same time, in order to speed up the training process and improve the prediction accuracy, the original data is normalized to the maximum and minimum values ​​and scaled to [0,1].

[0057] Step 2: Based on the sample database established in step 1, the acceleration of the navigation body when entering the water is described in detail. The acceleration of the navigation body is a truncated cone-shaped projectile (such as Figure 2 As shown in the figure, the motion trajectory prediction is similar, only the output layer is changed to displacement time series data, and a fully connected neural network deep learning model for acceleration prediction during the entry of a navigation body into the water is established. The model mainly consists of three parts: input layer, hidden layer and output layer, among which the input layer variables include cross-sectional ratio s, semi-cone angle α, entry speed v and entry angle β, and the output layer variables are acceleration time series data.

[0058] After establishing the fully connected neural network model, the model hyperparameters need to be pre-set. The quality of the hyperparameter settings directly affects the prediction accuracy and efficiency. Therefore, grid search or Bayesian search methods can be used to find the number of hidden layers and neurons in each hidden layer, activation functions, etc., so as to improve the training efficiency and accuracy of the deep neural network. In order to prevent overfitting and avoid redundant training, an early stopping strategy is adopted. When the loss of the test set no longer decreases or even increases, the training process is terminated. By supervising the average relative error between the predicted value and the validation set data, the hyperparameters of the neural network model are finally determined. According to the deep neural network model established with the above parameters, the test set is input into the model for acceleration prediction. The prediction of the motion trajectory of the vehicle is similar to the prediction of acceleration, and the final result is as shown in the attached figure. Figure 3 The predicted results of the vehicle acceleration and motion trajectory are shown.

[0059] Step 3: Embed the predicted vehicle acceleration prediction model and motion trajectory prediction model in Step 2 into the genetic algorithm to optimize the design of the vehicle cross-sectional shape and water entry parameters. First, input the optimization range of the vehicle parameters and set the number n of parameter combinations calculated in each iteration. The larger the value of n, the higher the single calculation amount, which can be set according to the computing resource configuration. The algorithm will randomly initialize n different parameter combinations (s i , α i , v i , β i ), denoted as M0. After initialization, introduce the fast prediction model into the genetic algorithm, and use (s i , α i , v i , β i ) as the input to quickly calculate the acceleration and motion trajectory under each parameter combination. Evaluate the results under all parameter combinations based on the optimization objective. The combination with a smaller acceleration and a smaller deviation of the motion trajectory has a higher fitness. Since the difference in the x direction of the motion trajectory is not significant, the definitions of the x and y directions are as shown in Figure 2 . It is mainly manifested as a gradual deviation from the predetermined orbit or even a turn in the y direction. To more accurately capture the stability of the navigation trajectory, the displacement curve in the y direction is used for measurement. When the slope of the time history curve is smaller, the offset is also smaller and the trajectory is more stable. Therefore, extract the acceleration peak value of all individuals and the minimum value k min of the slope of the time history curve in the y direction and normalize them to be denoted as f1 and f2 respectively. The fitness function is selected as 1 / (w1×f1 + w2×f2), where w1 and w2 are weight coefficients. When more emphasis is placed on the optimization of ballistic stability, set w2 > w1. When more emphasis is placed on the optimization of load reduction, set w2 < w1. When there are no special requirements, w2 = w1 can be set. After evaluating all combinations according to the fitness, start to automatically generate a new generation of search population M1.

[0060] The generation process of the new population M1 is carried out by crossover, mutation and selection in genetics. First, the individuals in M0 are converted from decimal codes to binary codes. According to the crossover probability of 0.80, 80% of the individuals in M0 are randomly selected to crossover and exchange the coding segments, such as [0,0,0] and [1,1,1] crossover to generate [0,1,1] and [1,0,0] starting from the second position; according to the mutation probability of 0.01, 1.00% of the individuals in M0 are randomly selected to mutate the coding segments, such as [0,0,0] mutates to [0,1,0]; according to the fitness level, the individuals in M0 are randomly selected according to the fitness level, and the individuals with higher fitness have a higher probability of being selected and retained. The combinations generated by the three operations are merged to obtain the new population M1. In order to retain the best individuals in M0 and prevent individuals with high fitness from being unselected, the M0 population is merged and deduplicated with M1, and the top n combinations are selected from high to low in terms of fitness as the final M1. At this point, the generation of a new generation of population is complete. Repeat the fitness evaluation and population generation operations, and simultaneously monitor the optimization target and fitness change curve. When the fitness no longer increases, the optimization result tends to converge, and the optimization process stops.

[0061] Individuals with high fitness are selected through selection operations, and the scope of optimization is continuously expanded by combining crossover and mutation. The above population generation and fitness evaluation process is iterated until the optimization algorithm stops when the fitness curve reaches the maximum and tends to be stable. Figure 4 As shown in the figure, as the number of iterations increases, the minimum value k of the slope of the y-direction displacement history curve min The peak value of acceleration tends to the minimum, and the fitness increases to the maximum and then stabilizes. Output the best parameter combination [s, α, v, β] after optimization design, as shown in Figure 5 Optimization results of acceleration and vertical motion trajectory are shown.

[0062] In summary, deep neural networks have shown strong advantages in processing multi-source data, capturing data mapping relationships and rapid prediction. Based on the complexity of the problem being studied and the size of the data set, the present invention designs a deep neural network to obtain the intrinsic logical relationship between the input and output quantities that approximates the true solution, thereby realizing the prediction of the acceleration and motion trajectory time history curve of the navigation body during the water entry process.

[0063] The above implementation steps are used to understand the core idea of ​​the patent of this invention. For engineering and technical personnel in this field, modifications and improvements can be made based on the idea of ​​the present invention, such as using different artificial intelligence algorithms and hyperparameter optimization methods to establish a neural network model, predicting or optimizing other physical quantities such as the water entry speed of the navigation body, etc. These modifications and improvements are also within the scope of protection of the patent of this invention.

Claims

1. A method for predicting and optimizing the water entry load and motion trajectory of a navigation body based on artificial intelligence, characterized in that: The method comprises the following steps: Step 1: Use finite element, meshless or experimental methods to simulate or test the process of the navigation body entering the water, obtain the load characteristics and motion trajectory data of the navigation body during the process of entering the water, and obtain a sample database of the acceleration and position of the navigation body under different initial conditions by changing the four parameters of the navigation body cross-sectional ratio s, semi-cone angle α, entry speed v and entry angle β; Step 2: Based on the sample database obtained in step 1, the database is divided into a training set and a validation set, and a neural network rapid prediction model for predicting the acceleration history curve and motion trajectory of the vehicle under different initial parameters is established respectively; Step 3: The neural network rapid prediction model of the acceleration history curve and motion trajectory of the vehicle under different initial parameters obtained in step 2 is embedded into the genetic algorithm, the shape parameters and water entry parameters of the vehicle are optimized, and the water entry load and motion trajectory of the vehicle are predicted and optimized.

2. The method for predicting and optimizing the water entry load and motion trajectory of a navigation body based on artificial intelligence according to claim 1 is characterized in that: The method for optimizing the shape parameters and water entry parameters of the navigation body in step 3 is: Input the upper and lower bounds of the four parameters of the navigation body cross-sectional ratio s, semi-cone angle α, water entry speed v and water entry angle β, randomly sample the parameter combinations in the space formed by the upper and lower bounds to form an initial population, and calculate the acceleration and ballistic trajectory based on the acceleration and motion trajectory neural network rapid prediction model established in step 2; All parameter combinations in the initial population formed in step 3 are evaluated and screened, crossover and mutation processing is performed on the screened population, the search scale is expanded and a new generation of individuals is generated, the parameters are iterated and gradually optimized, and finally the best parameters that meet the above optimization objectives are output.

3. The method for predicting and optimizing the water entry load and motion trajectory of a navigation body based on artificial intelligence according to claim 1 is characterized in that: The ratio of the training set to the validation set described in step 2 should be no less than 3:

4.

4. The method for predicting and optimizing the water entry load and motion trajectory of a navigation body based on artificial intelligence according to claim 1 is characterized in that: The input variables of the neural network rapid prediction model for predicting the acceleration time history curve and motion trajectory of the vehicle under different initial parameters described in step 2 are the four parameters in step 1, the output variable is the acceleration time history data, and the load characteristics and motion trajectory data of the vehicle during the entry into the water are normalized to the minimum and maximum values.

5. The method for predicting and optimizing the water entry load and motion trajectory of a navigation body based on artificial intelligence according to claim 1 is characterized in that: The neural network rapid prediction model of the vehicle acceleration history curve and motion trajectory includes an input layer, an output layer and multiple hidden layers, and the input layer variables are the vehicle cross-sectional ratio s, the semi-cone angle α, the water entry speed v and the water entry angle β.

6. The method for predicting and optimizing the water entry load and motion trajectory of a navigation body based on artificial intelligence according to claim 5 is characterized in that: The hyperparameters included in the input layer, the output layer and the multiple hidden layers are optimized by methods such as grid search or Bayesian search; similarly, the hyperparameters in the genetic algorithm are optimized by methods such as grid search or Bayesian search.

7. The method for predicting and optimizing the water entry load and motion trajectory of a navigation body based on artificial intelligence according to claim 1 is characterized in that: The output layer of the neural network rapid prediction model of the vehicle acceleration time history curve described in step 2 is the vehicle acceleration time history data.

8. The method for predicting and optimizing the water entry load and motion trajectory of a navigation body based on artificial intelligence according to claim 1 is characterized in that: The output layer of the neural network rapid prediction model of motion trajectory is the time history data of the vehicle position coordinates.

9. A computer device comprising a memory and a processor, characterized in that A computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images