Vehicle body agent model construction method based on improved neural network

Through the meta-heuristic algorithm and Levi flight method, the neural network parameters are optimized, combined with gradient descent method and adaptive step size factor, the problem of insufficient convergence accuracy and waste of computing resources in the construction of body agent models is solved, achieving more efficient The body agent model construction.

CN119089569AActive Publication Date: 2024-12-06CHINA STATE SHIPBUILDING CORP NO 707 RES INST +1
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202410973287.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-12-06
Estimated Expiration
2044-07-19

AI Technical Summary

Technical Problem

When building a vehicle body agent model, the convergence accuracy is greatly affected by the initial parameters and is prone to falling into local optimization, resulting in waste of computing resources and insufficient accuracy.

Method used

The meta-heuristic algorithm is used to optimize the neural network parameters, and the optimal parameters are perturbed through the Levi flight method, and the gradient descent method is used for secondary optimization, and an adaptive step factor is introduced to reduce iterative oscillation and improve convergence speed and accuracy.

Benefits of technology

It effectively reduces the impact of the initial value on the results, avoids local optimization, improves the accuracy and computing efficiency of the body agent model, and is suitable for large-scale body structure designs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089569A_ABST
    Figure CN119089569A_ABST
Patent Text Reader

Abstract

The invention relates to a vehicle body agent model construction method based on an improved neural network, and the method comprises the steps: carrying out the sampling of a model data initial sample and the normalization processing of model data after the number of model data initial samples and the dimension of model input data are determined; secondly, performing primary optimization on neural network parameters through a meta-heuristic algorithm with high global search performance, and disturbing optimal parameters by using a Levy flight method in the optimization of the meta-heuristic algorithm; secondly, performing secondary optimization on the neural network parameters optimized by the meta-heuristic algorithm by using a gradient descent method, introducing a self-adaptive step length factor in the secondary optimization process, and gradually reducing the step length factor along with the increase of the number of iterations of the gradient descent method; the method is small in calculation amount, high in convergence speed and suitable for construction of proxy models in structural design of various large vehicle bodies.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention patent relates to the field of vehicle body structure, and in particular to a method for constructing a vehicle body proxy model based on an improved neural network. Background Art

[0002] In the conceptual design stage of the vehicle body, it is usually necessary to simulate and analyze the performance of the vehicle frame and optimize important performance indicators. If a high-precision finite element model is directly used for analysis, it will cause a huge waste of computing resources and it will be difficult to promote in engineering. In order to improve the efficiency of simulation design, a proxy model can be constructed by collecting multiple sample points, and the proxy model can be used instead of the high-precision finite element model during the optimization process. The proxy model has fast calculation speed and can handle the time-consuming optimization design problem of the vehicle frame.

[0003] At present, the commonly used methods for constructing surrogate models are mainly divided into the following categories: response surface method, neural network method and Kriging model. Response surface method and Kriging model are suitable for models with smaller dimensions. When the model dimension increases, the computational complexity of these two methods increases exponentially. In the process of body structure design, there are many variables and output performances that need to be considered, and the model dimension is large. The response surface method and Kriging model consume a lot of computation in practical applications. The neural network method has obvious advantages in dealing with nonlinear and high-dimensional problems. However, when using neural networks to construct surrogate models, there are the following problems: First, the convergence accuracy of the neural network method is greatly affected by the selection of the initial values ​​of each parameter. If the initial values ​​of the parameters of the neural network are not properly selected, it is easy to cause the neural network to converge too slowly or even fail to train; second, when optimizing parameters, the neural network method is easy to obtain local optimal parameters instead of global optimal parameters, which seriously affects the accuracy of the surrogate model.

[0004] In response to the above problems, the present invention proposes a method for constructing a car body proxy model based on an improved neural network, which can improve the convergence efficiency of the neural network method, save computing resources, and at the same time improve the accuracy of the car body proxy model. Summary of the invention

[0005] In view of the deficiencies of the prior art, the present invention proposes a method for constructing a body proxy model based on an improved neural network, which can improve the convergence efficiency of the neural network method, save computing resources, and also improve the accuracy of the body proxy model.

[0006] One of the above objects of the present invention is achieved by the following technical solution:

[0007] A method for constructing a vehicle body proxy model based on an improved neural network comprises the following steps:

[0008] S1: Determine the model input data dimension, model output data dimension and the number of initial samples of model data;

[0009] S2: Use Latin hypercube sampling to sample the initial sample of model data;

[0010] S3: normalize the model data;

[0011] S4: Determine the number of input layers, hidden layers, and output layers of the neural network;

[0012] S5: Determine the initial sample number of neural network parameters;

[0013] S6: Determine the maximum number of iterations of the meta-heuristic algorithm and the maximum number of iterations of the gradient descent method;

[0014] S7: Use meta-heuristic algorithms to optimize the parameters in neural networks;

[0015] S8: using the Levy flight method, perturb the currently generated optimal parameters;

[0016] S9: Determine whether the current number of iterations of the meta-heuristic algorithm is greater than the maximum number of iterations of the meta-heuristic algorithm. If the current number of iterations of the meta-heuristic algorithm is greater than the maximum number of iterations of the meta-heuristic algorithm, proceed to step S10; otherwise, return to step S7;

[0017] S10: taking the optimal neural network parameters outputted in step S9 as input, and optimizing them twice using the gradient descent method;

[0018] S11: Determine whether the current number of iterations of the gradient descent method is greater than the maximum number of iterations of the gradient descent method. If the current number of iterations of the gradient descent method is greater than the maximum number of iterations of the gradient descent method, proceed to step S12; otherwise, return to step S10;

[0019] S12: Output the neural network parameters to complete the construction of the proxy model.

[0020] Furthermore, in step S1, the number N of initial samples of model data is determined by the following formula:

[0021] N=2k(ub-lb)+k

[0022] In the formula, ub is the upper limit value of the input variable; lb is the lower limit value of the input variable; k is the model input data dimension; the model input data dimension k is determined according to actual conditions.

[0023] Furthermore, in step S3, the model data is mainly divided into model input data and model output data. The model input data is a matrix composed of body parameters, and the model output parameters are mainly a matrix composed of body performance. The model input data is generated by sampling in step S2, and the model input data is imported into the Hypermesh software for calculation to obtain the corresponding model output data. The i-th dimension model input data P i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN ) is normalized by the following formula:

[0024]

[0025] P′ i =(p′ i1 ,p′ i2 ,p′ i3 ,p′ i4 ,...,p′ iN )

[0026] Where p′ ij Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN )’s j-th element is normalized; min{P i} Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN ) minimum value; max{P i} Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN ) maximum value;

[0027] The r-th dimension model outputs data Q r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN ) is normalized by the following formula:

[0028]

[0029] Q′ r =(q′ r1 ,q′ r2 ,q′ r3 ,q′ r4 ,...,q′ rN )

[0030] In the formula, q′ rj Output data Q for the r-th dimension model r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN )’s j-th element is normalized; min{Q r} is the r-th dimension model output data Q r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN ) minimum value; max{Q r} is the r-th dimension model output data Q r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN ) is the maximum value among .

[0031] Further, in step S4, the number of neural network input layers is the model input data dimension; the number of neural network output layers is the model output data dimension; the number of neural network hidden layers is determined by the following formula:

[0032]

[0033] In the formula, k is the model input data dimension; m is the model output data dimension; n is the number of hidden layers of the neural network.

[0034] Furthermore, in step S5, the initial sample number of the neural network parameters is determined by the following formula:

[0035] M = 200 (n·k+m·n+k·m)

[0036] In the formula, k is the model input data dimension; m is the model output data dimension; n is the number of hidden layers of the neural network; M is the number of initial samples of the neural network parameters.

[0037] Furthermore, in step S6, the maximum number of iterations of the meta-heuristic algorithm is in the range of 100 to 500; the maximum number of iterations of the gradient descent method is maxiter 2 The maximum number of iterations of the meta-heuristic algorithm maxiter 1 The following relationship is satisfied:

[0038] maxiter 2 =100·maxiter 1

[0039] In the formula, maxiter 1 is the maximum number of iterations of the meta-heuristic algorithm; maxiter 2 is the maximum number of iterations of the gradient descent method.

[0040] Further, in step S7, the design variables are parameters in the neural network, which are mainly composed of four parts: a first weight coefficient W, a second weight coefficient E, a first bias coefficient B and a second bias coefficient D; the total number of neural network parameters is determined by the following formula:

[0041] T=n·k+m·n+m+n

[0042] Where T is the total number of neural network parameters; k is the model input data dimension; m is the model output data dimension; n is the number of hidden layers of the neural network;

[0043] The i-th design variable vector is:

[0044] The i-th design variable vector can be written as Z i =[W i ,E i ,B i ,D i ],in, Group all design variables into global variables:

[0045]

[0046] In step S7, the i-th group of neural network parameter fitness obj i Calculated by the following formula:

[0047] obj i =∑∑|F i -Q′|

[0048]

[0049] Where Q′ is the normalized model output data; F iis the output data of the i-th group of neural network parameters; the output data F of the i-th group of neural network parameters i Calculated by the following formula:

[0050]

[0051] Where W i is the first weight coefficient of the i-th group; E i is the second weight coefficient of the i-th group; B i is the first bias coefficient of the i-th group; D i is the second bias coefficient of the i-th group; P′ is the normalized model input data;

[0052] Logsig is the activation function of the neural network, and its expression is as follows:

[0053]

[0054] In the formula, e is a natural constant, and its value is 2.71828;

[0055] In step S7, the update formula in the meta-heuristic algorithm is as follows:

[0056]

[0057] Where a is the number of iterations of the meta-heuristic algorithm; is the i-th group of neural network parameters in the a+1-th iteration; is the optimal parameter for the ath iteration; TA, TS and TB are random vectors (1×T) in the range [0,1]; sum(TS) is the sum of all elements in TS; wt is a random value in the range [-5,5]; rnd is a random value in the range [0,1]; coe is a random value in the range [0,0.17]; T is the total number of neural network parameters; maxiter 1 is the maximum number of iterations of the meta-heuristic algorithm.

[0058] Further, in step S8, the Levy flight method perturbs the optimal parameters by the following formula:

[0059]

[0060] Where β is a constant variable with a value of 1.8; lfu and lfv are random vectors (1×T) in the range [0,1]; Γ is the gamma function; is the optimal parameter in the a-th iteration of the meta-heuristic algorithm; is the parameter vector of the best parameter in the a-th meta-heuristic algorithm iteration after being perturbed by the Levy flight method; if the parameter vector perturbed by the Levy flight method Better than the original parameter vector Use replace Otherwise no replacement is required.

[0061] Further, in step S10, when using the gradient descent method, the optimal neural network parameter Z output in step S9 is u It is divided into four parts according to the following method: Gradient descent first weight coefficient W u , Gradient descent second weight coefficient E u , gradient descent first bias coefficient B u And the second bias coefficient D of gradient descent u ;

[0062]

[0063] Further, in step S10, the gradient descent method updates the first weight coefficient W of the gradient descent by the following formula: u , Gradient descent second weight coefficient E u , gradient descent first bias coefficient B u And the second bias coefficient D of gradient descent u :

[0064]

[0065] In the formula, It is the first weight coefficient of gradient descent in the y+1th iteration of the gradient algorithm; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; It is the second weight coefficient of gradient descent in the y+1th iteration of the gradient algorithm; is the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; It is the first bias coefficient of gradient descent in the y+1th iteration of the gradient algorithm; is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; It is the second bias coefficient of gradient descent in the y+1th iteration of the gradient algorithm; The second bias coefficient of gradient descent in the yth iteration of the gradient algorithm; learnrate (y) is the step size factor in the yth iteration of the gradient algorithm; is the increment of the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the increment of the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the increment of the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; is the increment of the second bias coefficient of gradient descent in the yth iteration of the gradient algorithm;

[0066] The increment of the first weight coefficient of the gradient descent in the yth iteration of the gradient algorithm The increment of the second weight coefficient of the gradient descent in the yth iteration of the gradient algorithm The increment of the first bias coefficient of the gradient descent in the yth iteration of the gradient algorithm The increment of the second bias coefficient of the gradient descent in the yth iteration of the gradient algorithm Calculated by the following formula:

[0067]

[0068] In the formula, delta1 (y) delta2 is the first updated parameter in the yth iteration of the gradient algorithm; (y) is the second updated parameter in the yth iteration of the gradient algorithm; P′ is the normalized model input data; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm;

[0069] The first updated parameter delta1 in the yth iteration of the gradient algorithm (y) The second update parameter delta2 in the yth iteration of the gradient algorithm (y) The calculation is done by the following formula:

[0070]

[0071] In the formula, Q′ is the normalized model output data; is the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; P′ is the normalized model input data; is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; The output data of the neural network parameters in the yth iteration of the gradient algorithm can be obtained by the following formula:

[0072]

[0073] In the formula, is the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; P′ is the normalized model input data; is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; is the second bias coefficient of gradient descent in the yth iteration of the gradient algorithm; logsig is the activation function of the neural network, which is the same as the activation function of the neural network mentioned above;

[0074] The step size factor learnrate in the yth iteration of the gradient algorithm (y) Calculated by the following formula:

[0075]

[0076] Where y is the number of iterations of the gradient algorithm; maxiter 2 is the maximum number of iterations of the gradient descent method.

[0077] The advantages and positive effects of the present invention are:

[0078] 1. The present invention proposes a meta-inspired algorithm with strong global search performance to optimize the neural network parameters, which can effectively reduce the influence of the initial value of the neural network on the final result. In order to avoid falling into the local optimum during the parameter optimization process and improve the accuracy of the body proxy model, the present invention uses the Levy flight method to perturb the optimal parameters in the meta-inspired algorithm optimization.

[0079] 2. The present invention introduces an adaptive step factor when optimizing the neural network parameters using the gradient descent method. As the number of iterations of the gradient descent method increases, the step factor gradually decreases, effectively reducing the optimization convergence time and avoiding the oscillation of the iterative results.

[0080] 3. The patent of the present invention first uses a meta-inspiration algorithm to optimize the neural network parameters once, and then uses the gradient descent method to perform a second optimization on the neural network parameters optimized by the meta-inspiration algorithm; this method has a small amount of calculation and a fast convergence speed, and is suitable for the construction of proxy models in various large-scale vehicle body structure designs. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 This is a flow chart of a method for constructing a vehicle body proxy model based on an improved neural network according to the present invention;

[0082] Figure 2 This is an isometric view of the vehicle body frame according to an embodiment of the present invention;

[0083] Figure 3 A top view of a vehicle body frame according to an embodiment of the present invention;

[0084] Figure 4 It is an iterative convergence curve of an embodiment of the present invention. DETAILED DESCRIPTION

[0085] The technical solution of the present invention is described in more detail below in conjunction with the accompanying drawings. The present invention includes but is not limited to the following embodiments.

[0086] A method for constructing a vehicle body proxy model based on an improved neural network, see Figure 1 , the invention point is: comprising the following steps:

[0087] S1: Determine the model input data dimension, model output data dimension and the number of initial samples of model data;

[0088] S2: Use Latin hypercube sampling to sample the initial sample of model data;

[0089] S3: normalize the model data;

[0090] S4: Determine the number of input layers, hidden layers, and output layers of the neural network;

[0091] S5: Determine the initial sample number of neural network parameters;

[0092] S6: Determine the maximum number of iterations of the meta-heuristic algorithm and the maximum number of iterations of the gradient descent method;

[0093] S7: Use meta-heuristic algorithms to optimize the parameters in neural networks;

[0094] S8: using the Levy flight method, perturb the currently generated optimal parameters;

[0095] S9: Determine whether the current number of iterations of the meta-heuristic algorithm is greater than the maximum number of iterations of the meta-heuristic algorithm. If the current number of iterations of the meta-heuristic algorithm is greater than the maximum number of iterations of the meta-heuristic algorithm, proceed to step S10; otherwise, return to step S7;

[0096] S10: taking the optimal neural network parameters outputted in step S9 as input, and optimizing them twice using the gradient descent method;

[0097] S11: Determine whether the current number of iterations of the gradient descent method is greater than the maximum number of iterations of the gradient descent method. If the current number of iterations of the gradient descent method is greater than the maximum number of iterations of the gradient descent method, proceed to step S12; otherwise, return to step S10;

[0098] S12: Output the neural network parameters to complete the construction of the proxy model.

[0099] In step S1, the initial sample size N of the model data is determined by the following formula:

[0100] N=2k(ub-lb)+k

[0101] In the formula, ub is the upper limit value of the input variable; lb is the lower limit value of the input variable; k is the model input data dimension; the model input data dimension k is determined according to actual conditions.

[0102] In step S2, the hypercube Latin sampling method is a sampling method built into the MATLAB software.

[0103] In step S3, the model data is mainly divided into model input data and model output data. The model input data is a matrix composed of body parameters, and the model output parameters are mainly a matrix composed of body performance. The model input data is generated by sampling in step S2. The model input data is imported into the Hypermesh software for calculation to obtain the corresponding model output data. The i-th dimension model input data P i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN ) is normalized by the following formula:

[0104]

[0105] P′ i =(p′ i1 ,p′ i2 ,p′ i3 ,p′ i4 ,...,p′ iN )

[0106] Where p′ ij Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN )’s j-th element is normalized; min{P i} Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN ) minimum value; max{P i} Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN ) is the maximum value among .

[0107] The r-th dimension model outputs data Q r =(q r1 ,q r2 ,q r3 ,q r4 ,...,qrN ) is normalized by the following formula:

[0108]

[0109] Q′ r =(q′ r1 ,q′ r2 ,q′ r3 ,q′ r4 ,...,q′ rN )

[0110] In the formula, q′ rj Output data Q for the r-th dimension model r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN )’s j-th element is normalized; min{Q r} is the r-th dimension model output data Q r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN ) minimum value; max{Q r} is the r-th dimension model output data Q r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN ) is the maximum value among .

[0111] In step S4, the number of neural network input layers is the model input data dimension; the number of neural network output layers is the model output data dimension; the number of neural network hidden layers is determined by the following formula:

[0112]

[0113] In the formula, k is the model input data dimension; m is the model output data dimension; n is the number of hidden layers of the neural network. The number of hidden layers of the neural network is an integer, and the calculated value needs to be rounded up.

[0114] The model input data dimension is equal to the total number of design variables of the vehicle body, and the model output data dimension is equal to the number of vehicle body performance indicators of interest.

[0115] In step S5, the initial sample number of the neural network parameters is determined by the following formula:

[0116] M = 200 (n·k+m·n+k·m)

[0117] In the formula, k is the model input data dimension; m is the model output data dimension; n is the number of hidden layers of the neural network; M is the number of initial samples of the neural network parameters.

[0118] In step S6, the maximum number of iterations of the meta-heuristic algorithm is in the range of 100 to 500. The maximum number of iterations of the gradient descent method maxiter 2 The maximum number of iterations of the meta-heuristic algorithm maxiter 1 The following relationship is satisfied:

[0119] maxiter 2 =100·maxiter 1

[0120] In the formula, maxiter 1 is the maximum number of iterations of the meta-heuristic algorithm; maxiter 2 is the maximum number of iterations of the gradient descent method.

[0121] In step S7, the design variables are parameters in the neural network, which are mainly composed of four parts: a first weight coefficient W, a second weight coefficient E, a first bias coefficient B and a second bias coefficient D. The total number of neural network parameters is determined by the following formula:

[0122] T=n·k+m·n+m+n

[0123] Where T is the total number of neural network parameters; k is the model input data dimension; m is the model output data dimension; and n is the number of hidden layers in the neural network.

[0124] The i-th design variable vector is

[0125] To facilitate subsequent calculations, the i-th design variable vector can be written as Z i =[W i ,E i ,B i ,D i ],in, Group all design variables into global variables:

[0126]

[0127] In step S7, the i-th group of neural network parameter fitness obj i Calculated by the following formula:

[0128] obj i =∑∑|F i -Q′|

[0129]

[0130] Where Q′ is the normalized model output data; F i The output data of the i-th group of neural network parameters. The output data of the i-th group of neural network parameters F i Calculated by the following formula:

[0131]

[0132] Where W i is the first weight coefficient of the i-th group; E i is the second weight coefficient of the i-th group; B i is the first bias coefficient of the i-th group; D i is the second bias coefficient of the i-th group; P′ is the normalized model input data.

[0133] Logsig is the activation function of the neural network, and its expression is as follows:

[0134]

[0135] In the formula, e is a natural constant, and its value is 2.71828.

[0136] In step S7, the update formula in the meta-heuristic algorithm is as follows:

[0137]

[0138] Where a is the number of iterations of the meta-heuristic algorithm; is the i-th group of neural network parameters in the a+1-th iteration; is the optimal parameter for the ath iteration; TA, TS and TB are random vectors (1×T) in the range [0,1]; sum(TS) is the sum of all elements in TS; wt is a random value in the range [-5,5]; rnd is a random value in the range [0,1]; coe is a random value in the range [0,0.17]; T is the total number of neural network parameters; maxiter 1 is the maximum number of iterations of the meta-heuristic algorithm.

[0139] In step S8, the Levy flight method perturbs the optimal parameters by the following formula:

[0140]

[0141] Where β is a constant variable with a value of 1.8; lfu and lfv are random vectors (1×T) in the range [0,1]; Γ is the gamma function; is the optimal parameter in the a-th iteration of the meta-heuristic algorithm; is the parameter vector of the best parameter in the a-th meta-heuristic algorithm iteration after being perturbed by the Levy flight method. If the parameter vector perturbed by the Levy flight method Better than the original parameter vector Use replace Otherwise no replacement is required.

[0142] In step S10, when using the gradient descent method, the optimal neural network parameter Z output in step S9 is u It is divided into four parts according to the following method: Gradient descent first weight coefficient W u , Gradient descent second weight coefficient E u , gradient descent first bias coefficient B u And the second bias coefficient D of gradient descent u .

[0143]

[0144] 12. The method for constructing a vehicle body proxy model based on an improved neural network according to claim 1, characterized in that in step S10, the gradient descent method updates the first weight coefficient W of the gradient descent by the following formula: u , Gradient descent second weight coefficient E u , gradient descent first bias coefficient B u And the second bias coefficient D of gradient descent u :

[0145]

[0146]

[0147] In the formula, It is the first weight coefficient of gradient descent in the y+1th iteration of the gradient algorithm; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; It is the second weight coefficient of gradient descent in the y+1th iteration of the gradient algorithm; is the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; It is the first bias coefficient of gradient descent in the y+1th iteration of the gradient algorithm; is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; It is the second bias coefficient of gradient descent in the y+1th iteration of the gradient algorithm; The second bias coefficient of gradient descent in the yth iteration of the gradient algorithm; learnrate (y) is the step size factor in the yth iteration of the gradient algorithm; is the increment of the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the increment of the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the increment of the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; The increment of the second bias coefficient of gradient descent in the yth iteration of the gradient algorithm.

[0148] The increment of the first weight coefficient of the gradient descent in the yth iteration of the gradient algorithm The increment of the second weight coefficient of the gradient descent in the yth iteration of the gradient algorithm The increment of the first bias coefficient of the gradient descent in the yth iteration of the gradient algorithm The increment of the second bias coefficient of the gradient descent in the yth iteration of the gradient algorithm Calculated by the following formula:

[0149]

[0150] In the formula, delta1 (y) delta2 is the first updated parameter in the yth iteration of the gradient algorithm; (y) is the second updated parameter in the yth iteration of the gradient algorithm; P′ is the normalized model input data; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; It is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm.

[0151] The first updated parameter delta1 in the yth iteration of the gradient algorithm (y) The second update parameter delta2 in the yth iteration of the gradient algorithm (y) The calculation is done by the following formula:

[0152]

[0153] In the formula, Q′ is the normalized model output data; is the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; P′ is the normalized model input data; is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; The output data of the neural network parameters in the yth iteration of the gradient algorithm can be obtained by the following formula:

[0154]

[0155] In the formula, is the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; P′ is the normalized model input data; is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; is the second bias coefficient of gradient descent in the yth iteration of the gradient algorithm; logsig is the activation function of the neural network, which is the same as the activation function of the neural network in the previous content.

[0156] The step size factor learnrate( y ) is obtained by the following formula:

[0157]

[0158] Where y is the number of iterations of the gradient algorithm; maxiter 2 is the maximum number of iterations of the gradient descent method.

[0159] Example:

[0160] This embodiment takes the body frame structure of a six-wheeled unmanned security vehicle as an example. The isometric view of the body frame is as follows: Figure 2 As shown, the top view of the body frame is as follows Figure 3 According to symmetry and manufacturing constraints, the six-wheeled unmanned security vehicle body frame structure can be divided into 19 parts. Figure 2 and 3 The parts are marked with numbers 1-19, and the thickness of each part is used as the design variable, that is, the model input data dimension k = 19. The thickness of the body frame structure of the six-wheeled unmanned security vehicle ranges from 1mm to 4mm, that is, lb = 1, ub = 4. The initial sample size of the model data can be obtained by the following calculation formula:

[0161] N=2k(ub-lb)+k=2×19×(4-1)+19=133

[0162] In this embodiment, the vehicle body performance of interest includes the first-order modal frequency, the second-order modal frequency, the maximum displacement under static load, and the mass. Therefore, the model output data dimension m=4. The results of sampling the initial sample of model data according to the Latin super-cube sampling method are shown in Appendix 1-1 and Appendix 1-2. The model input data is imported into the Hypermesh software for calculation, and the corresponding model output data can be obtained, as shown in Appendix 2.

[0163] The model input data in Appendix 1 are normalized according to the following formula:

[0164]

[0165] P′ i =(p′ i1 ,p′ i2 ,p′ i3 ,p′ i4 ,...,p′ iN )

[0166] Where p′ ij Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN )’s j-th element is normalized; min{P i} Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN ) minimum value; max{P i} Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN ). The results of normalized model input data are shown in Appendix 3-1 and Appendix 3-2.

[0167] The model output data in Appendix 2 are normalized according to the following formula:

[0168]

[0169] Q′ r =(q′ r1 ,q′ r2 ,q′ r3 ,q′ r4 ,...,q′ rN )

[0170] In the formula, q′ rj Output data Q for the r-th dimension model r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN )’s j-th element is normalized; min{Q r} is the r-th dimension model output data Q r =(q r1,q r2 ,q r3 ,q r4 ,...,q rN ) minimum value; max{Q r} is the r-th dimension model output data Q r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN The results of normalized model output data are shown in Appendix 4.

[0171] The number of hidden layers n of the neural network can be obtained according to the following formula:

[0172]

[0173] The initial number of samples of neural network parameters can be determined by the following formula:

[0174] M=200(n·k+m·n+k·m)=200×(10×19+4×10+19×4)=61200

[0175] The maximum number of iterations of the meta-heuristic algorithm maxiter 1 The value is 600. The maximum number of iterations of the gradient descent method can be calculated according to the following calculation formula: maxiter 2 :

[0176] maxiter 2 =100·maxiter 1 =100×600=60000

[0177] The iterative convergence curve is as follows Figure 4 As shown in Figure 2, the error between the predicted value and the actual value of the vehicle body performance index decreases with the increase of the number of iterations. After iterative calculation, the first weight coefficient W of the gradient descent is obtained. u , Gradient descent second weight coefficient E u , gradient descent first bias coefficient B u And the second bias coefficient D of gradient descent u as follows:

[0178]

[0179] In order to verify the effectiveness of the modeling of the body proxy model, multiple groups of data are randomly generated for verification. The generated verification group data are shown in Appendix 5. The actual performance of the verification group data calculated in hypermesh and the predicted performance of the body proxy model proposed by the present invention are shown in Appendix 6. It can be seen from the table that the error between the predicted value and the actual value of the first-order modal frequency of the body model of each verification group is within 5%, the error between the predicted value and the actual value of the second-order modal frequency is within 10%, the error between the predicted value and the actual value of the maximum displacement is within 10%, and the error between the predicted value and the actual value of the mass is within 10%. By comparison, it can be proved that the body proxy model construction method proposed in the present invention has good prediction accuracy, and the method can be used in engineering simulation applications.

[0180] Appendix 1-1

[0181]

[0182]

[0183]

[0184]

[0185] Appendix 1-2

[0186]

[0187]

[0188]

[0189]

[0190] Schedule 2

[0191]

[0192]

[0193]

[0194]

[0195]

[0196] Appendix 3-1

[0197]

[0198]

[0199]

[0200]

[0201] Appendix 3-2

[0202]

[0203]

[0204]

[0205]

[0206] Schedule 4

[0207]

[0208]

[0209]

[0210]

[0211] Schedule 5

[0212]

[0213] Appendix 6-1

[0214]

[0215]

[0216] Appendix 6-2

[0217]

Claims

1. A method for constructing a vehicle body proxy model based on an improved neural network. It is characterized in that The following steps are involved: S1: Determine the model input data dimension, model output data dimension and the number of initial samples of model data; S2: Use Latin hypercube sampling to sample the initial sample of model data; S3: normalize the model data; S4: Determine the number of input layers, hidden layers, and output layers of the neural network; S5: Determine the initial sample number of neural network parameters; S6: Determine the maximum number of iterations of the meta-heuristic algorithm and the maximum number of iterations of the gradient descent method; S7: Use meta-heuristic algorithms to optimize the parameters in neural networks; S8: using the Levy flight method, perturb the currently generated optimal parameters; S9: Determine whether the current number of iterations of the meta-heuristic algorithm is greater than the maximum number of iterations of the meta-heuristic algorithm. If the current number of iterations of the meta-heuristic algorithm is greater than the maximum number of iterations of the meta-heuristic algorithm, proceed to step S10; Otherwise, return to step S7; S10: taking the optimal neural network parameters outputted in step S9 as input, and optimizing them twice using the gradient descent method; S11: Determine whether the current number of iterations of the gradient descent method is greater than the maximum number of iterations of the gradient descent method. If the current number of iterations of the gradient descent method is greater than the maximum number of iterations of the gradient descent method, proceed to step S12; Otherwise, return to step S10; S12: Output the neural network parameters to complete the construction of the proxy model.

2. The method for constructing a vehicle body proxy model based on an improved neural network according to claim 1, Features: In step S1, the initial sample size N of the model data is determined by the following formula: N=2k(ub-lb)+k In the formula, ub is the upper limit value of the input variable; lb is the lower limit value of the input variable; k is the model input data dimension; the model input data dimension k is determined according to actual conditions.

3. The method for constructing a vehicle body proxy model based on an improved neural network according to claim 1, Features: In step S3, the model data is mainly divided into model input data and model output data. The model input data is a matrix composed of body parameters, and the model output parameters are moments composed of body performance. The model input data is generated by sampling in step S2, and the model input data is imported into the Hypermesh software for calculation to obtain the corresponding model output data. The i-th dimension model input data P i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN ) is normalized by the following formula: P i ′=(p i ′ 1 ,p i ′ 2 ,p i ′ 3 ,p i ′ 4 ,...,p i ′ N ) Where p′ ij Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN )’s j-th element is normalized; min{P i } Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN ) minimum value; max{P i } Input data P for the i-th dimension model i =(p i1 ,p i2 ,p i3 ,p i4 ,...,p iN ) maximum value; The r-th dimension model output data Q r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN ) is normalized by the following formula: Q′ r =(q r ′ 1 ,q r ′ 2 ,q r ′ 3 ,q r ′ 4 ,...,q r ′ N ) In the formula, q′ rj Output data Q for the r-th dimension model r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN )’s j-th element is normalized; min{Q r } is the r-th dimension model output data Q r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN ) minimum value; max{Q r } is the r-th dimension model output data Q r =(q r1 ,q r2 ,q r3 ,q r4 ,...,q rN ) is the maximum value among .

4. The method for constructing a vehicle body proxy model based on an improved neural network according to claim 1, Features: In step S4, the number of neural network input layers is the model input data dimension k; the number of neural network output layers is the model output data dimension m; and the number of neural network hidden layers n is determined by the following formula: In the formula, k is the model input data dimension; m is the model output data dimension; n is the number of hidden layers of the neural network.

5. The method for constructing a vehicle body proxy model based on an improved neural network according to claim 1, Features: In step S5, the initial sample number of the neural network parameters is determined by the following formula: M = 200 (n·k+m·n+k·m) In the formula, k is the model input data dimension; m is the model output data dimension; n is the number of hidden layers of the neural network; M is the number of initial samples of the neural network parameters.

6. The method for constructing a vehicle body proxy model based on an improved neural network according to claim 1, Features: In step S6, the maximum number of iterations of the meta-heuristic algorithm is in the range of 100 to 500; the maximum number of iterations of the gradient descent method is maxiter 2 The maximum number of iterations of the meta-heuristic algorithm maxiter 1 The following relationship is satisfied: maxiter 2 =100·maxiter 1 In the formula, maxiter 1 is the maximum number of iterations of the meta-heuristic algorithm; maxiter 2 is the maximum number of iterations of the gradient descent method.

7. The method for constructing a vehicle body proxy model based on an improved neural network according to claim 1, Features: In step S7, the design variables are parameters in the neural network, which are mainly composed of four parts: a first weight coefficient W, a second weight coefficient E, a first bias coefficient B and a second bias coefficient D; the total number of neural network parameters is determined by the following formula: T=n·k+m·n+m+n Where T is the total number of neural network parameters; k is the model input data dimension; m is the model output data dimension; n is the number of hidden layers of the neural network; The i-th design variable vector is The i-th design variable vector is abbreviated as: Z i =[W i ,E i ,B i ,D i ] in, Group all design variables into global variables: In step S7, the i-th group of neural network parameter fitness obj i Calculated by the following formula: obj i =∑∑|F i -Q′| Where Q′ is the normalized model output data; F i is the output data of the i-th group of neural network parameters; the output data F of the i-th group of neural network parameters i Calculated by the following formula: Where W i is the first weight coefficient of the i-th group; E i is the second weight coefficient of the i-th group; B i is the first bias coefficient of the i-th group; D i is the second bias coefficient of the i-th group; P′ is the normalized model input data; Logsig is the activation function of the neural network, and its expression is as follows: In the formula, e is a natural constant, and its value is 2.71828; In step S7, the update formula in the meta-heuristic algorithm is as follows: Where a is the number of iterations of the meta-heuristic algorithm; is the i-th group of neural network parameters in the a+1-th iteration; is the optimal parameter for the ath iteration; TA, TS and TB are random vectors (1×T) in the range [0,1]; sum(TS) is the sum of all elements in TS; wt is a random value in the range [-5,5]; rnd is a random value in the range [0,1]; coe is a random value in the range [0,0.17]; T is the total number of neural network parameters; maxiter 1 is the maximum number of iterations of the meta-heuristic algorithm.

8. The method for constructing a vehicle body proxy model based on an improved neural network according to claim 1, Features: In step S8, the Levy flight method perturbs the optimal parameters by the following formula: Where β is a constant variable with a value of 1.8; lfu and lfv are random vectors (1×T) in the range [0,1]; Γ is the gamma function; is the optimal parameter in the a-th iteration of the meta-heuristic algorithm; is the parameter vector of the best parameter in the a-th meta-heuristic algorithm iteration after being perturbed by the Levy flight method; if the parameter vector perturbed by the Levy flight method Better than the original parameter vector Use replace Otherwise no replacement is required.

9. The method for constructing a vehicle body proxy model based on an improved neural network according to claim 1, Features: In step S10, when using the gradient descent method, the optimal neural network parameter Z output in step S9 is u It is divided into four parts according to the following method: Gradient descent first weight coefficient W u , Gradient descent second weight coefficient E u , gradient descent first bias coefficient B u And the second bias coefficient D of gradient descent u 10. The method for constructing a vehicle body proxy model based on an improved neural network according to claim 1, Features: In step S10, the gradient descent method updates the first weight coefficient W of the gradient descent by the following formula: u , Gradient descent second weight coefficient E u , gradient descent first bias coefficient B u And the second bias coefficient D of gradient descent u : In the formula, It is the first weight coefficient of gradient descent in the y+1th iteration of the gradient algorithm; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; It is the second weight coefficient of gradient descent in the y+1th iteration of the gradient algorithm; is the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; It is the first bias coefficient of gradient descent in the y+1th iteration of the gradient algorithm; is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; It is the second bias coefficient of gradient descent in the y+1th iteration of the gradient algorithm; The second bias coefficient of gradient descent in the yth iteration of the gradient algorithm; learnrate (y) is the step size factor in the yth iteration of the gradient algorithm; is the increment of the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the increment of the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the increment of the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; is the increment of the second bias coefficient of gradient descent in the yth iteration of the gradient algorithm; The increment of the first weight coefficient of the gradient descent in the yth iteration of the gradient algorithm The increment of the second weight coefficient of the gradient descent in the yth iteration of the gradient algorithm The increment of the first bias coefficient of the gradient descent in the yth iteration of the gradient algorithm The increment of the second bias coefficient of the gradient descent in the yth iteration of the gradient algorithm Calculated by the following formula: In the formula, delta1 (y) delta2 is the first updated parameter in the yth iteration of the gradient algorithm; (y) is the second updated parameter in the yth iteration of the gradient algorithm; P′ is the normalized model input data; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; The first updated parameter delta1 in the yth iteration of the gradient algorithm (y) The second update parameter delta2 in the yth iteration of the gradient algorithm (y) The calculation is done by the following formula: In the formula, Q′ is the normalized model output data; is the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; P′ is the normalized model input data; is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; The output data of the neural network parameters in the yth iteration of the gradient algorithm can be obtained by the following formula: In the formula, is the second weight coefficient of gradient descent in the yth iteration of the gradient algorithm; is the first weight coefficient of gradient descent in the yth iteration of the gradient algorithm; P′ is the normalized model input data; is the first bias coefficient of gradient descent in the yth iteration of the gradient algorithm; is the second bias coefficient of gradient descent in the yth iteration of the gradient algorithm; logsig is the activation function of the neural network; The step size factor learnrate in the yth iteration of the gradient algorithm (y) Calculated by the following formula: Where y is the number of iterations of the gradient algorithm; maxiter 2 is the maximum number of iterations of the gradient descent method.

Citation Information

Patent Citations

  • Three-dimensional positioning method based on PSO_BP neural network

    CN106912105A

  • Neural network parameter optimization method and system and electronic equipment

    CN107909142A

  • Antenna design method based on neural network

    CN109086531A

  • Overdue monitoring method for optimizing recurrent neural network based on ant colony algorithm

    CN112634018A

  • Airship comprehensive optimization method based on neural network

    CN116702331A