CN112685834A - Collision energy absorption prediction method, medium and terminal for vehicle body front end structural component - Google Patents
Collision energy absorption prediction method, medium and terminal for vehicle body front end structural component Download PDFInfo
- Publication number
- CN112685834A CN112685834A CN202011629090.2A CN202011629090A CN112685834A CN 112685834 A CN112685834 A CN 112685834A CN 202011629090 A CN202011629090 A CN 202011629090A CN 112685834 A CN112685834 A CN 112685834A
- Authority
- CN
- China
- Prior art keywords
- output
- layer
- value
- hidden layer
- energy absorption
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 74
- 238000010521 absorption reaction Methods 0.000 title claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 62
- 238000013528 artificial neural network Methods 0.000 claims abstract description 47
- 241000287127 Passeridae Species 0.000 claims abstract description 42
- 238000012360 testing method Methods 0.000 claims abstract description 27
- 238000010606 normalization Methods 0.000 claims abstract description 15
- 238000010845 search algorithm Methods 0.000 claims abstract description 15
- 238000011156 evaluation Methods 0.000 claims abstract description 11
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000003062 neural network model Methods 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 58
- 230000008569 process Effects 0.000 claims description 27
- 230000004913 activation Effects 0.000 claims description 21
- 230000003213 activating effect Effects 0.000 claims description 9
- 125000004432 carbon atom Chemical group C* 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000012546 transfer Methods 0.000 claims description 6
- 238000004088 simulation Methods 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 239000002131 composite material Substances 0.000 claims description 3
- 238000009795 derivation Methods 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000002789 length control Methods 0.000 claims description 3
- 238000012886 linear function Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 abstract description 9
- 238000005070 sampling Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000005211 surface analysis Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Abstract
The invention relates to the field of vehicle body structure design, in particular to a method, a medium and a terminal for predicting collision energy absorption of a vehicle body front end structural component. The method comprises the following steps: s1, taking the thickness of a key part of a front-end structure of the vehicle body as an input parameter, taking the energy absorption of the key part as an output parameter, and taking a data set after normalization processing as a training sample set and a test sample set; s2, determining the number of hidden layers and the number of nodes of the hidden layers, and constructing a BP neural network structure; s3, optimizing the initial weight and the threshold of the BP neural network by using a sparrow search algorithm; s4, training a BP neural network; s5, calculating a prediction precision evaluation parameter to obtain an optimized BP neural network model; and S6, testing the optimized BP neural network, performing inverse normalization on the prediction result to obtain an energy absorption prediction value, and evaluating the network performance and the prediction precision. The method for predicting the collision energy absorption of the front end structural component of the vehicle body is efficient and accurate, and solves the problems in the prior art.
Description
Technical Field
The invention relates to the field of vehicle body structure design, in particular to a method, a medium and a terminal for predicting collision energy absorption of a vehicle body front end structural component.
Background
In the structural design of the automobile body, the structural crash resistance evaluation is an indispensable ring, in the collision process, the energy absorption of the front end part of the automobile body is an important index for evaluating the crash resistance of the automobile body structure, the energy absorption capacity of the automobile body plays an important role in protecting passengers, and the kinetic energy which is not absorbed in the collision process can be directly transmitted to a passenger compartment, so that the automobile body is seriously deformed and the passengers are seriously injured.
In the design of the vehicle body, a finite element analysis method is usually adopted for structural simulation, and the calculation of a finite element model is long in time consumption; meanwhile, results need to be optimized, and the optimization process is also a repeated iteration process. The process time for predicting the energy absorption of the front end structural component of the vehicle body by adopting the method is too long, and the method has no engineering significance; there is a need for a more efficient and accurate method of predicting energy absorption of a front end structural member of a vehicle body.
At present, the premise of realizing the energy absorption prediction of a vehicle body structure part in engineering is to construct an approximate model; response surface analysis method and Crimen method are approximate model construction methods for solving common and effective energy absorption prediction of vehicle body structure parts. However, the above methods also have their limitations: for example, as the degree of non-linearity of the processing system increases, the amount of sample point data that needs to be input increases significantly, and fitting a function often has difficulty accounting for all points. And a large number of sample points are needed to construct by simply adopting the kriging model, so that the efficiency is low, and the prediction precision can not meet the requirement.
The BP neural network has strong nonlinear fitting capability and self-adaptive capability, has wide application in the aspect of approximate model prediction, and can also be used for predicting collision energy absorption of a front-end structural component of a vehicle body. However, the initial weight and the threshold of the BP neural network are randomly selected, which easily causes the problem of falling into the local minimum, leads to unsatisfactory prediction effect and larger prediction error; therefore, how to optimize the BP neural network algorithm and overcome the problems becomes a key for solving whether the energy absorption prediction of the front-end structural component of the vehicle body is accurate or not.
Disclosure of Invention
In order to overcome the problems in the prior art, the invention provides a collision energy-absorption prediction method of a vehicle body front end structural component, and a medium and a terminal applying the method.
The invention is realized by adopting the following technical scheme:
a method for predicting collision energy absorption of a vehicle body front end structural component comprises the following steps:
s1, collecting training sample data and test sample data by taking the thickness of the key part of the front-end structure of the automobile body as an input parameter and the energy absorption of the key part of the front-end structure of the automobile as an output parameter; carrying out normalization processing on each group of data, and taking the data as a training sample set and a test sample set;
the formula for the normalization process is as follows:
in the above formula, the first and second carbon atoms are,
for normalized data, xiFor data to be normalized output from the training set, xminOutputting a minimum value, x, for the training set to be normalizedmaxOutputting a maximum value for the training set to be normalized;
s2, determining the number of input layer nodes and the number of output layer nodes according to the input parameters and the output parameters in S1, determining the number of hidden layers and the number of hidden layer nodes, selecting the type of an activation function between layers of a BP neural network, and constructing a BP neural network structure;
the BP neural network adopting a three-layer structure has the following relationship between the number of hidden layer nodes and the number of input layer nodes:
hiddennum＝2×inputnum+1
in the above formula, hiddennum is the number of hidden layer nodes, and inputnum is the number of input layer nodes;
the activating function between the input layer and the hidden layer is a hyperbolic tangent function Tanh, the activating function between the hidden layer and the output layer is a linear function Purelin, and the activating function formula between the layers is as follows:
tanh activation function:
purelin activation function: f. of2(x)＝x
In the above formula, x is the total input of any node of the hidden layer or the output layer, and f (x) is the total output of any node of the hidden layer or the output layer;
the transfer formula from the BP neural network input layer to the hidden layer is as follows:
in the above formula, H1jFor the output of the jth node of the hidden layer, j is 1,2, …, hiddenum, f1As a function of Tanh activation, ωijIs the weight between the ith node of the input layer and the jth node of the hidden layer, aiIs an input value of the ith node of the input layer, ajA threshold value of the jth node of the hidden layer;
the transfer formula from the hidden layer to the output layer is:
in the above formula, y is the output value of the output layer, f2As Purelin activation function, ωjThe weight between the jth node of the hidden layer and the output layer, and b is the threshold of the output layer;
s3, optimizing the initial weight and the threshold of the BP neural network by using a sparrow search algorithm;
the optimization process of the sparrow search algorithm comprises the following steps:
s31, initializing the population:
initializing the content includes: the population scale N, the number pNum of discoverers, the number sNum of reconnaissance, the dimension D of a target function, the maximum iteration number T, and the upper and lower limits lb and ub of an initial value;
the objective function dimension calculation formula is as follows:
D＝inputnum×hiddennum+hiddennum×outputnum+hiddennum+outputnum
in the above formula, inputnum is the number of nodes of the input layer, hiddennum is the number of nodes of the hidden layer, and outputnum is the number of nodes of the output layer;
s32, selecting a fitness function:
the fitness function is the absolute value of the relative error between the predicted output and the expected output of the output layer and is calculated by adopting the following formula:
in the above formula, erroriAs a fitness function of each sparrow, yiIn order to train the desired output of the sample,
predicting output for an output layer;
s33, fitness sorting:
calculating the fitness of each sparrow in the population, selecting the sparrow with the optimal fitness and the worst fitness in the current population, and determining the position corresponding to the sparrow with the optimal fitness and the worst fitness;
s34, updating positions of discoverers, joiners and spyware:
selecting the first pNum sparrows with better fitness in the population as discoverers, selecting the rest sparrows as participants, randomly selecting sNum sparrows in the population as scouts, and updating the position information of the discoverers, the participants and the scouts in turn;
wherein, the discoverer position updating formula is as follows:
in the above formula, xijIndicating the position information of the ith sparrow in the jth dimension, i is 1,2, …, N, j is 1,2, …, D, t is the current iteration number, itermaxAlpha is (0, 1) for the maximum iteration number of the algorithm]Q is a random number following a standard normal distribution, L is a matrix of 1 × d and the elements are all 1, R2∈[0,1]And ST ∈ [0.5,1 ]]Respectively representing an early warning value and a safety value;
the subscriber location update formula is as follows:
in the above formula, the first and second carbon atoms are,
represents the best position of the finder at the t +1 th iteration,
representing the global worst position at the t-th iteration, A represents a 1 x d matrix, where each element is randomly assigned a value of 1 or-1, and A+＝AT.(AAT)-1；
The scout position update formula is as follows:
in the above formula, the first and second carbon atoms are,
represents the global optimum position in the t iteration, beta is taken as a step length control parameter, and obeys a normal distribution random number with the mean value of 0 and the variance of 1, and K is [ -1,1]Is a very small constant, fi、fgAnd fwRespectively representing the fitness, global optimum and worst fitness of the current sparrow;
s35, optimizing the weight and the threshold;
obtaining a current optimal value, if the current optimal value is better than the optimal value of the last iteration, performing updating operation, otherwise, not performing updating operation, and continuing the iteration operation until a convergence condition is met, finally obtaining a global optimal position and an optimal fitness value, and taking the optimal position as a weight and a threshold value after the BP neural network is optimized;
s4, setting initial training parameters, and training the BP neural network by using the training sample set in S1;
s5, calculating the estimation accuracy evaluation parameter including the determination coefficient R2And the maximum relative error of the MRE and,
(3) if both the weight and the threshold meet the set value, extracting the trained weight and threshold to obtain an optimized BP neural network model;
(4) if at least one item is not satisfied, returning to S3 to train again until the coefficient R is determined2And the maximum relative error MRE both meet a set value;
and S6, testing the optimized BP neural network by using the test sample set, carrying out reverse normalization on the prediction result to obtain a key piece collision energy absorption prediction value, and evaluating the network performance and the prediction precision.
Further, in step S1, the acquired front end structural component object includes: the front anti-collision device comprises a front anti-collision beam front plate, a front anti-collision beam rear plate, an energy absorption box outer shell, an energy absorption box inner shell, a front longitudinal beam front section, a front longitudinal beam rear section upper part and a front longitudinal beam rear section lower part; the thickness value of each component is used as an input parameter, the energy absorption of each component is used as an output parameter, and simulation data are collected to form a training sample set and a testing sample set based on a finished automobile finite element model and simulation software.
Further, in the structure of the BP neural network constructed in step S2, the number of nodes in the input layer is 7, the number of nodes in the output layer is 1, the number of nodes in the hidden layer is 1, and the number of nodes in the hidden layer is 15.
Further, in step S31, the initial parameters of the sparrow search algorithm are: population size 40, finder number 8, scout number 4, objective function dimension 136, maximum iteration number 30, initial value upper and lower limits-5 and 5.
Further, in step S35, the convergence condition is that the maximum number of iterations 30 is reached.
Further, in step S4, the initial training parameter in the training process of the BP neural network is iteration number 100, the minimum error of the training target is 0.03, and the learning rate η is 0.01;
the training process comprises the following steps:
s41, carrying out forward propagation on the training sample, and calculating the absolute value of the relative error between the predicted output and the expected output of the output layer;
the forward propagation specifically includes: training samples are input from an input layer, processed through an activation function of a hidden layer, output by nodes of the hidden layer, processed through an activation function of an output layer, output by nodes of the output layer to obtain an actual output value, and a relative error absolute value of a predicted output value and an expected output value of the output node is calculated;
randomly selecting the u-th input sample and the corresponding expected output data, wherein the sum of the square of the absolute error of the predicted output and the expected output is expressed as:
in the above formula, EuPredicting the absolute value of the relative error between the output and the expected output, y, for the u-th input sampleuIs the expected output of the u-th input sampleAnd then the mixture is discharged out of the furnace,
a prediction output for the u-th input sample;
s42, judging whether the absolute value of the relative error between the predicted output and the expected output meets the requirement that the minimum error of the training target is less than 0.03;
(3) if not, the process goes to S43 to perform the back propagation process,
(4) if so, finishing the training of the BP neural network model and ending;
s43, adopting gradient descent method to perform backward propagation process, gradient being vector composed of partial derivatives of error function to weight and threshold, and using EuPerforming composite derivation on the weights to obtain a weight updating formula from the hidden layer to the output layer and a weight updating formula between the input layer and the hidden layer, and respectively updating the weights of the output layer and the weights between the input layer and the hidden layer according to the updating formulas;
wherein, the weight ω between the input layer and the hidden layerijThe update formula is:
weight omega from hidden layer to output layerjThe update formula is:
in the above formula, η is the learning rate and takes the value of 0.01;
and S44, inputting the updated weight value between the input layer and the hidden layer obtained in the S43, the weight value between the hidden layer and the output layer, and the input layer and the hidden layer in the forward propagation in the S41 to continue the forward propagation, and returning to the S41.
Further, in step S5, the prediction accuracy evaluation parameter determination coefficient R2And the maximum relative error MRE calculation formula is respectively:
in the formula, ziThe real output of the sample is obtained,
the output is predicted for the output layer(s),
is the average value of the samples, and m is the number of training samples.
Further, in the precision evaluation of the test result, at least the maximum relative error MRE < 5% is satisfied and the coefficient R is determined2The expected requirements set can be met only by being greater than 0.99.
The present invention also provides a computer-readable storage medium having stored thereon a computer program which, when being executed by a processor, carries out the steps of the method for energy absorption prediction of collision of a front end structural member of a vehicle body as described above.
The invention also provides a computer terminal which comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein when the processor executes the computer program, the steps of the collision energy-absorbing prediction method of the vehicle body front end structural component can be realized.
The collision energy-absorbing prediction method for the vehicle body front end structural component has the following beneficial effects:
the method adopts the BP neural network to process the problem of collision energy absorption prediction of the front end structural component of the vehicle body, utilizes the strong nonlinear fitting capability of the method to enable the prediction result to be more efficient, and shortens the time of model simulation and result prediction. Meanwhile, the method has good self-adaptive capacity, compared with the traditional response surface and a Krigin approximate model, the high-precision approximate model can be constructed through a small number of sample points, and the prediction precision of the algorithm can be enhanced through the experience accumulation of a training process and a prediction result, so that the method has good engineering significance.
For the BP neural network, the initial weight and the threshold are randomly selected, so that the defect of local minimum is easily caused; the method optimizes the sparrow search algorithm through the sparrow search algorithm, and the sparrow search algorithm has extremely strong global search capability and high convergence rate. The optimal characteristic of global search is optimized by using a sparrow search algorithm, the defect that a BP neural network is partially optimal in learning is avoided, the model has good convergence and adaptability and good generalization capability, the collision energy absorption prediction effect of the optimized method on the front-end structural component of the vehicle body is more accurate and stable, the prediction error of the BP neural network is reduced, the reliability of the prediction result is improved, and the design cycle and the optimization cycle of the front-end structure of the vehicle body are shortened.
Drawings
FIG. 1 is a flowchart of a method for predicting energy absorption upon collision of a vehicle body front end structural member according to embodiment 1;
FIG. 2 is a schematic view showing sampling positions of variables in the vehicle body front end structural member in the embodiment 1;
FIG. 3 is a schematic structural diagram of the BP neural network in this embodiment 1;
FIG. 4 is a comparison curve between the predicted value and the simulation value of the energy absorption of the front structure of the vehicle body of the test sample set in the embodiment 1;
fig. 5 is a variation curve of the fitness function with the decrease of the number of iterations in this embodiment 1.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Example 1
As shown in fig. 1, the method for predicting collision energy absorption of a vehicle body front end structural member according to the present embodiment includes the following steps:
s1, collecting training sample data and test sample data by taking the thickness of the key part of the front-end structure of the vehicle body as an input parameter and the energy absorption of the key part of the front-end structure of the vehicle body as an output parameter based on a finite element model and simulation software of the whole vehicle; carrying out normalization processing on each group of data, and respectively using the data as a training sample set and a test sample set;
in the collision process, the main energy absorption components at the front end of the vehicle body comprise a front anti-collision beam, a front longitudinal beam, an energy absorption box, a cover sending system, a water tank cross beam, a fender and the like, in the embodiment, the sampling positions of the components are properly reduced based on the load transmission path in the collision process and the symmetrical relation among different components in the structural components, after simplification, only the thickness values of the front anti-collision beam front plate, the front anti-collision beam rear plate, the energy absorption box outer shell, the energy absorption box inner shell, the front longitudinal beam front section, the front longitudinal beam rear section upper part and the front longitudinal beam rear section lower part shown in the figure 2 are collected, and the thickness values of the seven components are used as input parameters of a subsequent energy absorption prediction method.
The design variables corresponding to the thicknesses of the seven parts are x in sequence1～x7The initial values and value ranges of the design variables are shown in table 1;
table 1: in this embodiment, the initial value of the design variable and the value range thereof
The method comprises the steps of taking the thickness values of seven key parts as input parameters, taking the energy absorption of the seven key parts as output parameters, adopting an optimal Latin super test design method to collect 65 groups of data in a sampling space as a training sample set, and collecting 15 groups of data in the sample space as a testing sample set. Part of the training sample set is shown in table 2.
Table 2: partial training sample set data
| x1 | x2 | x3 | x4 | x5 | x6 | x7 | Energy absorbing/KJ |
| 1.612 | 1.898 | 1.102 | 1.531 | 1.633 | 1.224 | 1.061 | 21.22 |
| 1.776 | 1.408 | 1.306 | 1.857 | 1.327 | 1.959 | 1.837 | 24.1 |
| 1.531 | 1.041 | 1.776 | 1.878 | 1.735 | 1.857 | 1.429 | 23.63 |
| 1.796 | 1.531 | 1.245 | 1.959 | 2 | 1.469 | 1.531 | 24.51 |
| 1.469 | 1.204 | 1.122 | 1.367 | 1.122 | 1.184 | 1.163 | 21.76 |
| 1.898 | 1.102 | 1.531 | 1.633 | 1.714 | 1.367 | 1.98 | 21.43 |
| 1.714 | 1.224 | 1.98 | 1.082 | 1.449 | 1.204 | 1.592 | 24.29 |
| 1.224 | 1.633 | 1.816 | 1.143 | 1 | 1.49 | 1.327 | 24.57 |
| 1.939 | 1.490 | 1.469 | 1.816 | 1.367 | 1 | 1.449 | 23.7 |
| 1.51 | 1.878 | 1.265 | 1.163 | 1.163 | 1.082 | 1.633 | 25.44 |
| 1.429 | 1.163 | 1.898 | 1.735 | 1.245 | 1.245 | 1.143 | 23.16 |
| 1.082 | 1.673 | 1.714 | 1.122 | 1.531 | 1.306 | 1.959 | 22.7 |
| 1.347 | 1.061 | 1.204 | 2 | 1.408 | 1.347 | 1.571 | 23.86 |
| 1.020 | 1 | 1.51 | 1.245 | 1.531 | 1.327 | 1.51 | 22.3 |
| 1.857 | 1.796 | 1.612 | 1.306 | 1.857 | 1.061 | 1.776 | 23.17 |
| 1.918 | 1.714 | 1.673 | 1.286 | 1.143 | 1.612 | 1.939 | 22.45 |
| 1.531 | 1.939 | 2 | 1.612 | 1.306 | 1.163 | 1.612 | 24.82 |
| 1.388 | 1.837 | 1.796 | 1.98 | 1.612 | 1.531 | 1.082 | 23.01 |
| 1.184 | 2 | 1.49 | 1.51 | 1.98 | 1.408 | 1.531 | 24 |
| 1.735 | 1.571 | 1.143 | 1.184 | 1.102 | 1.898 | 1.306 | 23.47 |
And carrying out normalization processing on all the training sample sets and the test sample sets by using a normalization processing formula, and mapping the acquired data to a (-1, 1) interval through the normalization formula to obtain 65 groups of training sample sets and 15 groups of verification sample sets after the normalization processing.
In the formula (I), the compound is shown in the specification,
for normalized data, xiFor data to be normalized output from the training set, xminOutputting a minimum value, x, for the training set to be normalizedmaxOutputting a maximum value for the training set to be normalized;
in this embodiment, the test set outputs the predicted value to obtain the predicted value of energy absorption of the key component through inverse normalization.
And S2, determining the number of nodes of an input layer and the number of nodes of an output layer according to the input parameters and the output parameters in the S1, determining the number of hidden layers and the number of nodes of the hidden layers, selecting the type of an activation function between the layers of the neural network, and constructing the BP neural network structure.
In this embodiment, a BP neural network with a three-layer structure is adopted, and the number of nodes in an input layer and the number of nodes in an output layer of the BP neural network are determined to be 7 and 1 respectively according to input parameters and output parameters. The relationship between the number of hidden layer nodes and the number of input layer nodes is as follows:
hiddennum＝2×inputnum+1
in the above formula, hiddennum is the number of hidden layer nodes, and inputnum is the number of input layer nodes. Therefore, the number of hidden layer nodes in this example is 15, the structure of the BP neural network as shown in fig. 3 is constructed, in the figure, a is the number of input layer nodes, b is the number of hidden layer nodes, c is the number of output layer nodes, and the structure of the BP neural network is "7-15-1".
The activating function between the input layer and the hidden layer is a hyperbolic tangent function Tanh, the activating function between the hidden layer and the output layer is a linear function Purelin, and the activating functions between the layers are as follows:
tanh activation function:
purelin activation function: f. of2(x)＝x
In the above formula, x is the total input of any node in the hidden layer or the output layer, and f (x) is the total output of any node in the hidden layer or the output layer.
In this embodiment, the transfer formula from the input layer to the hidden layer is:
in the above formula, H1jThe output of the jth node of the hidden layer, j ═ 1,2, …,15, f1As a function of Tanh activation, ωijIs the weight between the ith node of the input layer and the jth node of the hidden layer, aiIs an input value of the ith node of the input layer, ajIs the threshold of the jth node of the hidden layer.
The transfer formula from the hidden layer to the output layer is:
in the above formula, y is the output value of the output layer, f2As Purelin activation function, ωjThe weight between the jth node of the hidden layer and the output layer, and b is the threshold of the output layer.
And S3, optimizing the initial weight and the threshold of the BP neural network by using a sparrow search algorithm.
The specific process of optimizing the sparrow search algorithm comprises the following steps:
s31, initializing the population:
the initialization content comprises a population size N, the number pNum of discoverers, the number sNum of reconnaissance, a target function dimension D, a maximum iteration number T and upper and lower limits lb and ub of an initial value.
The initial parameters of the sparrow search algorithm are: the population scale is 40, the number of discoverers is 8, the number of reconnaissants is 4, the dimension D of an objective function is 30, the maximum iteration number is 30, and the upper and lower limits of an initial value are-5 and 5.
The objective function dimension calculation formula is as follows:
D＝inputnum×hiddennum+hiddennum×outputnum+hiddennum+outputnum
wherein inputnum is the number of nodes of the input layer, hiddennum is the number of nodes of the hidden layer, and outputnum is the number of nodes of the output layer.
Based on the BP neural network structure with the "7-15-1" structure shown in fig. 3, there are 120 weight values 7 × 15+15 × 1 and 16 threshold values 15+1, so the objective function dimension D is 120+16 — 136.
S32, selecting a fitness function:
the fitness function is the absolute value of the relative error between the predicted output and the expected output of the output layer and is calculated by adopting the following formula:
in the above formula, erroriAs a fitness function of each sparrow, yiIn order to train the desired output of the sample,
and predicting output for the output layer.
S33, fitness sorting:
calculating the fitness of each sparrow in the population, selecting the sparrows with the optimal fitness and the worst fitness in the current population, and determining the positions corresponding to the sparrows with the optimal fitness and the worst fitness.
S34, updating positions of discoverers, joiners and spyware:
selecting the first 8 sparrows with better fitness in the population as discoverers, the rest sparrows as participants, randomly selecting 4 sparrows in the population as scouts, sequentially updating the position information of the discoverers, the participants and the scouts,
the finder location update formula is as follows:
in the above formula, xijIndicating the position information of the ith sparrow in the jth dimension, i is 1,2, …,40, j is 1,2, …,136, t is the current iteration number, itermaxIs 30, alpha is (0, 1)]Q is a random number following a standard normal distribution, L is a matrix of 1 × d and the elements are all 1, R2∈[0,1]And ST ∈ [0.5,1 ]]Respectively representing an early warning value and a safety value.
The subscriber location update formula is as follows:
in the above formula, the first and second carbon atoms are,
represents the best position of the finder at the t +1 th iteration,
denotes the t-th timeGlobal worst position in iteration, A represents a 1 x d matrix, where each element is randomly assigned a value of 1 or-1, and A+＝AT.(AAT)-1。
The scout position update formula is as follows:
in the above formula, the first and second carbon atoms are,
represents the global optimum position in the t iteration, beta is taken as a step length control parameter, and obeys a normal distribution random number with the mean value of 0 and the variance of 1, and K is [ -1,1]Is a very small constant, fi、fgAnd fwRespectively representing the fitness, the global optimum and the worst fitness of the current sparrow.
S35, weight and threshold optimization:
obtaining a current optimal value, if the current optimal value is better than the optimal value of the last iteration, carrying out updating operation, otherwise, not carrying out updating operation, and continuing the iteration operation until a convergence condition is met: and when the maximum iteration number is up to 30, finally obtaining a global optimal position and an optimal fitness value, and taking the optimal position as a weight and a threshold value after the BP neural network is optimized. In this embodiment, the weight and the threshold of the BP neural network optimized by the sparrow search algorithm are as follows:
weight omega between 1 st node of input layer and j th node of hidden layer1jSequentially comprises the following steps: -0.272, -0.451, -0.297, -0.364, -0.441, -0.384, -0.600, -0.539, -0.325, -0.547, -0.613, -0.437, -0.582, -0.732, -0.291
Weight omega between 2 nd node of input layer and j th node of hidden layer2jSequentially comprises the following steps: -0.361, -0.339, 0.185, -0.658, -0.532, -0.648, -0.455, -0.463, -0.271, -0.894, -0.386, -0.636, 0.086, -0.681, -0.401
Weight omega between the 4 th node of the input layer and the jth node of the hidden layer4jSequentially comprises the following steps: -0.789, -0.782, -0.464, -0.703, -0.573, -0.464, -0.361, -0.708, -0.563, -0.681, -0.813, -0.408, -0.489, -0.413, -0.625
Weight omega between 5 th node of input layer and j th node of hidden layer5jSequentially comprises the following steps: -0.655, -0.400, -0.377, -0.378, -0.187, -0.571, -0.429, -0.609, -0.405, -0.203, -0.535, -0.617, -0.462, -0.552, -0.986
Weight omega between the 6 th node of the input layer and the jth node of the hidden layer6jSequentially comprises the following steps: -0.721, -0.612, -0.529, -0.469, -0.649, -0.533, -0.695, -0.428, -0.625, -0.265, -0.779, -0.648, -0.265, -0.204
Weight omega between 7 th node of input layer and j th node of hidden layer7jSequentially comprises the following steps: -0.115, -0.752, -0.265, -0.459, -0.159, -0.325, -0.782, 0.236, -0.425, -0.489, -0.236, -0.456, -0.885, 0.235, 0.456
Weight omega between jth node of hidden layer and output layerjSequentially comprises the following steps: -0.741, -0.658, -0.451, -0.382, -0.364, -0.413, -0.648, -0.236, -0.659
Threshold a of the hidden layerjSequentially comprises the following steps: -0.367, -0.564, -0.256, -0.235, -0.123, -0.432, -0.276, -0.432, -0.298, -0.574, -0.119, -0.658, -0.513, -0.497, -0.553
The output node threshold b is: -0.511.
Wherein j is 15.
And S4, setting initial training parameters including iteration times, minimum error of a training target and learning rate, and training the BP neural network by adopting a training sample.
In the step, data in a training sample set is used for training the optimized BP neural network, the initial training parameter is iteration times 100, the minimum error of a training target is 0.03, and the learning rate eta is 0.01.
The specific training process comprises the following steps:
and S41, carrying out forward propagation on the training samples, and calculating the absolute value of the relative error between the predicted output and the expected output of the output layer. The forward propagation specifically includes: training samples are input from an input layer, processed through an activation function of a hidden layer, output by nodes of the hidden layer, processed through the activation function of an output layer, output by nodes of the output layer to obtain actual output values, and the relative error absolute values of the predicted output values and the expected output values of the output nodes are calculated.
Randomly selecting the u-th input sample and the corresponding expected output data, wherein the relative error absolute value formula of the predicted output and the expected output is as follows:
in the above formula, EuPredicting the absolute value of the relative error between the output and the expected output, y, for the u-th input sampleuFor the desired output of the u-th input sample,
is the predicted output of the u-th input sample.
S42, judging whether the absolute value of the relative error between the predicted output and the expected output meets the requirement that the minimum error of the training target is less than 0.03; if not, the process goes to S43 to perform back propagation, and if so, the training of the BP neural network model is completed and the process is finished.
S43, performing back propagation process by gradient descent method (gradient is vector composed of partial derivative of error function to weight and threshold), and using EuAnd (4) obtaining a weight updating formula from the hidden layer to the output layer and a weight updating formula between the input layer and the hidden layer by carrying out composite derivation on the weights, and respectively updating the weights of the output layer and the weights between the input layer and the hidden layer according to the updating formulas.
Wherein the input layer and the hidden layerWeight ω between layersijThe update formula is:
weight omega from hidden layer to output layerjThe update formula is:
in the above formula, η is the learning rate of 0.01;
and S44, inputting the updated weight value between the input layer and the hidden layer obtained in the S43, the weight value between the hidden layer and the output layer, and the input layer and the hidden layer in the forward propagation in the S41 to continue the forward propagation, and returning to the S41.
S5, calculating prediction accuracy evaluation parameter determination coefficient R2And a maximum relative error MRE, where R is a requirement2> 0.99 and MRE < 5%:
(1) if both the weight and the threshold meet the set value, extracting the trained weight and threshold to obtain an optimized BP neural network model;
(2) if at least one item is not satisfied, returning to S3 to train again until the coefficient R is determined2And the maximum relative error MRE satisfies a set value;
in this step, a prediction accuracy evaluation parameter determination coefficient R2And the maximum relative error MRE is calculated as:
in the formula, ziThe real output of the sample is obtained,
the output is predicted for the output layer(s),
is the average value of the samples, and m is the number of training samples.
S6, testing the optimized BP neural network by using the test sample set, carrying out reverse normalization on the prediction result to obtain a key part energy absorption prediction value, and calculating a prediction precision evaluation parameter decision coefficient R2And the maximum relative error MRE, and evaluating the network performance and the prediction accuracy.
The sample set data is tested as shown in table 3,
table 3: 15 sets of test sample set data
| x1 | x2 | x3 | x4 | x5 | x6 | x7 | Energy absorbing/KJ |
| 1.895 | 1.105 | 1.684 | 1 | 1.526 | 1.526 | 1.579 | 25.17 |
| 1.368 | 1.895 | 1.579 | 1.053 | 1.316 | 1.684 | 1.053 | 22.18 |
| 1.474 | 1.632 | 2 | 1.316 | 1.211 | 1.053 | 1.737 | 25.29 |
| 1.105 | 1.579 | 1.789 | 1.105 | 1.947 | 1.632 | 1.684 | 23.77 |
| 1.053 | 1.684 | 1.842 | 1.737 | 1.684 | 1.211 | 1.105 | 20.73 |
| 1.684 | 1.789 | 1.316 | 1.158 | 1.789 | 1 | 1.421 | 24.8 |
| 1 | 1.842 | 1.211 | 1.579 | 1.368 | 1.263 | 1.789 | 23.9 |
| 1.460 | 1.25 | 1.12 | 1.365 | 1.222 | 1.284 | 1.16 | 21.66 |
| 1.516 | 1.878 | 1.265 | 1.263 | 1.163 | 1.082 | 1.63 | 25.4 |
| 1.160 | 1.531 | 1.857 | 1.939 | 1.204 | 1.599 | 1.735 | 24.57 |
| 1.895 | 1.105 | 1.684 | 1 | 1.526 | 1.526 | 1.579 | 25.17 |
| 1.368 | 1.895 | 1.579 | 1.053 | 1.316 | 1.684 | 1.053 | 22.18 |
| 1.632 | 1.421 | 1.105 | 1.526 | 2 | 1.579 | 2 | 22.56 |
| 1.579 | 1.368 | 1.947 | 1.684 | 1.053 | 1.895 | 1.368 | 24.73 |
| 1.857 | 1.79 | 1.612 | 1.306 | 1.859 | 1.061 | 1.776 | 23.19 |
Based on the data of the test sample set, the original BP neural network energy absorption predicted value and the optimized BP neural network energy absorption predicted value in the embodiment are statistically compared with the simulated value of the structural component, and a curve as shown in FIG. 4 is drawn; the fitness function is shown in fig. 5.
As can be seen from the analysis of fig. 4 and 5: the optimized BP neural network has good prediction capability; when the number of iterations reaches 5 times, the best fitness value is already reached in the embodiment, which reflects that the method in the embodiment has super-strong iteration optimization capability.
Through calculation, in this embodiment, the prediction result determination coefficient R of the test sample set20.99152, MRE 2.01% was obtained from MATLAB. Both reach the expected value and meet the precision requirement. This reflects that the method in this example has high energy absorption prediction accuracy for the test sample set, and the prediction result is stable.
Example 2
The present embodiment provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program realizes the steps of the collision energy absorption prediction method for a vehicle body front end structural member according to embodiment 1.
When the method for predicting collision energy absorption of a vehicle body front end structural component in embodiment 1 is applied, the method can be applied in the form of software, for example, a program designed to be independently run by a computer-readable storage medium, which may be a usb flash disk, designed as a usb shield, and a program designed to start the whole method by external triggering through the usb flash disk.
Example 3
The present embodiments provide a computer terminal comprising a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the program to realize the steps of the collision energy absorption prediction method of the vehicle body front end structural member according to embodiment 1.
The collision energy absorption prediction method for the vehicle body front end structural component in the embodiment 1 can be applied in the form of software when being applied, for example, a program designed to run independently is installed on a computer terminal, and the computer terminal may be a computer, a smart phone, or the like. Or it can be designed as embedded running program and installed on computer terminal, such as single-chip computer.
The present invention is not limited to the above preferred embodiments, and any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A method for predicting collision energy absorption of a vehicle body front end structural component is characterized by comprising the following steps:
s1, collecting training sample data and test sample data by taking the thickness of the key part of the front-end structure of the vehicle body as an input parameter and taking the energy absorption of the key part of the front-end structure of the vehicle body as an output parameter; carrying out normalization processing on each group of data, and taking the data as a training sample set and a test sample set;
the formula for the normalization process is as follows:
in the above formula, the first and second carbon atoms are,
for normalized data, xiFor data to be normalized output from the training set, xminOutputting a minimum value, x, for the training set to be normalizedmaxOutputting a maximum value for the training set to be normalized;
s2, determining the number of input layer nodes according to the input parameters in S1, determining the number of output layer nodes according to the output parameters, determining the number of hidden layers and the number of hidden layer nodes, selecting the type of an activation function between layers of the BP neural network, and constructing a BP neural network structure;
the BP neural network adopting a three-layer structure has the following relationship between the number of hidden layer nodes and the number of input layer nodes:
hiddennum＝2×inputnum+1
in the above formula, hiddennum is the number of hidden layer nodes, and inputnum is the number of input layer nodes;
the activating function between the input layer and the hidden layer is a hyperbolic tangent function Tanh, the activating function between the hidden layer and the output layer is a linear function Purelin, and the activating function formula between the layers is as follows:
tanh activation function:
purelin activation function: f. of2(x)＝x
In the above formula, x is the total input of any node of the hidden layer or the output layer, and f (x) is the total output of any node of the hidden layer or the output layer;
the transfer formula from the BP neural network input layer to the hidden layer is as follows:
in the above formula, H1jFor the output of the jth node of the hidden layer, j is 1,2, …, hiddenum, f1As a function of Tanh activation, ωijIs the weight between the ith node of the input layer and the jth node of the hidden layer, aiIs an input value of the ith node of the input layer, ajA threshold value of the jth node of the hidden layer;
the transfer formula from the hidden layer to the output layer is:
wherein y is the output value of the output layer, f2As Purelin activation function, ωjThe weight between the jth node of the hidden layer and the output layer, and b is the threshold of the output layer;
s3, optimizing the initial weight and the threshold of the BP neural network by using a sparrow search algorithm;
the optimization process of the sparrow search algorithm comprises the following steps:
s31, initializing the population:
initializing the content includes: the population scale N, the number pNum of discoverers, the number sNum of reconnaissance, the dimension D of a target function, the maximum iteration number T, and the upper and lower limits lb and ub of an initial value;
the objective function dimension calculation formula is as follows:
D＝inputnum×hiddennum+hiddennum×outputnum+hiddennum+outputnum
in the above formula, inputnum is the number of nodes of the input layer, hiddennum is the number of nodes of the hidden layer, and outputnum is the number of nodes of the output layer;
s32, selecting a fitness function:
the fitness function is the absolute value of the relative error between the predicted output and the expected output of the output layer and is calculated by adopting the following formula:
in the above formula, erroriAs a fitness function of each sparrow, yiIn order to train the desired output of the sample,
predicting output for an output layer;
s33, fitness sorting:
calculating the fitness of each sparrow in the population, selecting the sparrow with the optimal fitness and the worst fitness in the current population, and determining the position corresponding to the sparrow with the optimal fitness and the worst fitness;
s34, updating positions of discoverers, joiners and spyware:
selecting the first pNum sparrows with better fitness in the population as discoverers, selecting the rest sparrows as participants, randomly selecting sNum sparrows in the population as scouts, sequentially updating the position information of the discoverers, the participants and the scouts,
wherein, the discoverer position updating formula is as follows:
in the above formula, xijIndicating the position information of the ith sparrow in the jth dimension, i is 1,2, …, N, j is 1,2, …, D, t is the current iteration number, itermaxAlpha is (0, 1) for the maximum iteration number of the algorithm]Q is a random number following a standard normal distribution, L is a matrix of 1 × d and the elements are all 1, R2∈[0,1]And ST ∈ [0.5,1 ]]Respectively representing an early warning value and a safety value;
the subscriber location update formula is as follows:
in the above formula, the first and second carbon atoms are,
represents the best position of the finder at the t +1 th iteration,
representing the global worst position at the t-th iteration, A represents a 1 x d matrix, where each element is randomly assigned a value of 1 or-1, and A+＝AT.(AAT)-1；
The scout position update formula is as follows:
in the above formula, the first and second carbon atoms are,
represents the global optimum position in the t iteration, beta is taken as a step length control parameter, and obeys a normal distribution random number with the mean value of 0 and the variance of 1, and K is [ -1,1]Is a very small constant, fi、fgAnd fwRespectively representing the fitness, global optimum and worst fitness of the current sparrow;
s35, weight and threshold optimization:
obtaining a current optimal value, if the current optimal value is better than the optimal value of the last iteration, performing updating operation, otherwise, not performing updating operation, and continuing the iteration operation until a convergence condition is met, finally obtaining a global optimal position and an optimal fitness value, and taking the optimal position as a weight and a threshold value after the BP neural network is optimized;
s4, setting initial training parameters, and training the BP neural network by using the training sample set in S1;
s5, calculating the estimation accuracy evaluation parameter including the determination coefficient R2And the maximum relative error of the MRE and,
(1) if both the weight and the threshold meet the set value, extracting the trained weight and threshold to obtain an optimized BP neural network model;
(2) if at least one item is not satisfied, returning to S3 to train again until the coefficient R is determined2And the maximum relative error MRE both meet a set value;
and S6, testing the optimized BP neural network by using the test sample set, carrying out reverse normalization on the prediction result to obtain a key part energy absorption prediction value at the front end of the vehicle body, and evaluating the network performance and the prediction precision.
2. The method for predicting collision energy absorption of a vehicle body front end structural member according to claim 1, characterized in that: in step S1, the acquired front-end structural component object includes: the front anti-collision device comprises a front anti-collision beam front plate, a front anti-collision beam rear plate, an energy absorption box outer shell, an energy absorption box inner shell, a front longitudinal beam front section, a front longitudinal beam rear section upper part and a front longitudinal beam rear section lower part; the thickness value of each component is used as an input parameter, the energy absorption of each component is used as an output parameter, and simulation data are collected to form a training sample set and a testing sample set based on a finished automobile finite element model and simulation software.
3. The method for predicting collision energy absorption of a vehicle body front end structural member according to claim 1, characterized in that: in the structure of the BP neural network constructed in step S2, the number of nodes in the input layer is 7, the number of nodes in the output layer is 1, the number of nodes in the hidden layer is 1, and the number of nodes in the hidden layer is 15.
4. The method for predicting collision energy absorption of a vehicle body front end structural member according to claim 1, characterized in that: in step S31, the initial parameters of the sparrow search algorithm are: the population scale is 40, the number of discoverers is 8, the number of reconnaissance is 4, the dimensionality of an objective function is 136, the maximum iteration number is 30, and the upper limit and the lower limit of an initial value are-5 and 5.
5. The method for predicting collision energy absorption of a vehicle body front end structural member according to claim 1, characterized in that: in step S35, the convergence condition is that the maximum number of iterations 30 is reached.
6. The method for predicting collision energy absorption of a vehicle body front end structural member according to claim 1, characterized in that: in step S4, the initial training parameter is iteration number 100, the minimum error of the training target is 0.03, and the learning rate η is 0.01;
the training process comprises the following steps:
s41, carrying out forward propagation on the training sample, and calculating the absolute value of the relative error between the predicted output and the expected output of the output layer;
the forward propagation specifically includes: training samples are input from an input layer, processed through an activation function of a hidden layer, output by nodes of the hidden layer, processed through an activation function of an output layer, output by nodes of the output layer to obtain an actual output value, and a relative error absolute value of a predicted output value and an expected output value of the output node is calculated;
randomly selecting the u-th input sample and the corresponding expected output data, wherein the relative error absolute value formula of the predicted output and the expected output is as follows:
in the above formula, EuPredicting the absolute value of the relative error between the output and the expected output, y, for the u-th input sampleuFor the desired output of the u-th input sample,
a prediction output for the u-th input sample;
s42, judging whether the absolute value of the relative error between the predicted output and the expected output meets the requirement that the minimum error of the training target is less than 0.03;
(1) if not, the process goes to S43 to perform the back propagation process,
(2) if so, finishing the training of the BP neural network model;
s43, adopting gradient descent method to perform backward propagation process, gradient being vector composed of partial derivatives of error function to weight and threshold, and using EuThe composite derivation of the weights is used to obtain the hidden layer to the output layerUpdating a formula of the weight value, and updating a formula of the weight value between the input layer and the hidden layer, wherein the weight value between the output layer and the hidden layer is updated according to the updating formulas;
wherein, the weight ω between the input layer and the hidden layerijThe update formula is:
weight omega from hidden layer to output layerjThe update formula is:
in the above formula, η is the learning rate and takes the value of 0.01;
and S44, inputting the updated weight value between the input layer and the hidden layer obtained in the S43, the weight value between the hidden layer and the output layer, and the input layer and the hidden layer in the forward propagation in the S41 to continue the forward propagation, and returning to the S41.
7. The method for predicting collision energy absorption of a vehicle body front end structural member according to claim 1, characterized in that: in step S5, the prediction accuracy evaluation parameter determination coefficient R2And the maximum relative error MRE calculation formula is respectively:
in the formula, ziThe real output of the sample is obtained,
to be transportedThe output of the out-of-layer prediction,
is the average value of the samples, and m is the number of training samples.
8. The method of predicting collision energy absorption of a vehicle body front end structural member according to claim 7, characterized in that: in the precision evaluation of the test result, at least the maximum relative error MRE < 5% is satisfied and the coefficient R is determined2The expected requirements set can be met only by being greater than 0.99.
9. A computer-readable storage medium, on which a computer program is stored, which program, when being executed by a processor, is adapted to carry out the steps of the method for energy-absorption-by-collision prediction of a vehicle body front end structural component according to any one of claims 1 to 8.
10. A computer terminal comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor when executing the program implements the steps of the method for energy absorption collision prediction of a vehicle body front end structural member according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011629090.2A CN112685834A (en) | 2020-12-31 | 2020-12-31 | Collision energy absorption prediction method, medium and terminal for vehicle body front end structural component |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011629090.2A CN112685834A (en) | 2020-12-31 | 2020-12-31 | Collision energy absorption prediction method, medium and terminal for vehicle body front end structural component |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112685834A true CN112685834A (en) | 2021-04-20 |
Family
ID=75455855
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011629090.2A Pending CN112685834A (en) | 2020-12-31 | 2020-12-31 | Collision energy absorption prediction method, medium and terminal for vehicle body front end structural component |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112685834A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113239599A (en) * | 2021-06-15 | 2021-08-10 | 江苏理工学院 | Intelligent tire wear life estimation method and device based on BP neural network |
| CN114417712A (en) * | 2022-01-01 | 2022-04-29 | 西北工业大学 | Airship propeller reliability estimation method based on chaos initialization SSA-BP neural network |
| CN115982862A (en) * | 2023-02-22 | 2023-04-18 | 成都赛力斯科技有限公司 | Complete vehicle dynamic parameter optimization method, model training method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101241521A (en) * | 2008-03-06 | 2008-08-13 | 上海交通大学 | Coachbuilt body combination property index modelling approach based on support vector machine |
| DE102011077518A1 (en) * | 2011-05-17 | 2012-11-22 | Robert Bosch Gmbh | Device for adaptively reducing collision energy generated during collision of vehicle, has actuator adjusting energy absorption properties of device based on signal representing collision energy |
| CN102945327A (en) * | 2012-11-21 | 2013-02-27 | 湖南大学 | Multi-target reliability optimization technique for direct impact safety of automobile |
| CN111324989A (en) * | 2020-03-19 | 2020-06-23 | 重庆大学 | GA-BP neural network-based gear contact fatigue life prediction method |
-
2020
- 2020-12-31 CN CN202011629090.2A patent/CN112685834A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101241521A (en) * | 2008-03-06 | 2008-08-13 | 上海交通大学 | Coachbuilt body combination property index modelling approach based on support vector machine |
| DE102011077518A1 (en) * | 2011-05-17 | 2012-11-22 | Robert Bosch Gmbh | Device for adaptively reducing collision energy generated during collision of vehicle, has actuator adjusting energy absorption properties of device based on signal representing collision energy |
| CN102945327A (en) * | 2012-11-21 | 2013-02-27 | 湖南大学 | Multi-target reliability optimization technique for direct impact safety of automobile |
| CN111324989A (en) * | 2020-03-19 | 2020-06-23 | 重庆大学 | GA-BP neural network-based gear contact fatigue life prediction method |
Non-Patent Citations (4)
| Title |
|---|
| JIANKAI XUE等: "A novel swarm intelligence optimization approach: sparrow search algorithm", 《SYSTEMS SCIENCE & CONTROL ENGINEERING》 * |
| 辛丕海等: "基于RBF近似模型的汽车前部部件耐撞性优化", 《农业装备与车辆工程》 * |
| 韦灼彬等: "《新型可持续海洋骨料混凝土-珊瑚混凝土孔隙结构及性能》", 31 July 2020, 天津：天津大学出版社 * |
| 高剑武等: "基于PSO-SVR近似模型的汽车正碰可靠性优化", 《合肥工业大学学报（自然科学版）》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113239599A (en) * | 2021-06-15 | 2021-08-10 | 江苏理工学院 | Intelligent tire wear life estimation method and device based on BP neural network |
| CN114417712A (en) * | 2022-01-01 | 2022-04-29 | 西北工业大学 | Airship propeller reliability estimation method based on chaos initialization SSA-BP neural network |
| CN114417712B (en) * | 2022-01-01 | 2023-07-04 | 西北工业大学 | Airship propeller reliability estimation method based on chaos initialization SSA-BP neural network |
| CN115982862A (en) * | 2023-02-22 | 2023-04-18 | 成都赛力斯科技有限公司 | Complete vehicle dynamic parameter optimization method, model training method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112685834A (en) | Collision energy absorption prediction method, medium and terminal for vehicle body front end structural component | |
| CN108960426B (en) | Road slope comprehensive estimation system based on BP neural network | |
| CN109581190B (en) | Optimal diagnosis tree generation method for circuit fault diagnosis | |
| CN107139873B (en) | Automobile rear bumper with function gradient negative Poisson's ratio structure and optimization method | |
| CN105224872A (en) | A kind of user's anomaly detection method based on neural network clustering | |
| Zhang et al. | Vehicle crash accident reconstruction based on the analysis 3D deformation of the auto-body | |
| CN108268728A (en) | Automobile tail gate structural optimization method based on two-step Modified particle swarm optimization algorithm | |
| CN112597610B (en) | Optimization method, device and equipment for lightweight design of mechanical arm structure | |
| Fang et al. | Fatigue optimization with combined ensembles of surrogate modeling for a truck cab | |
| CN111580151A (en) | SSNet model-based earthquake event time-of-arrival identification method | |
| CN111667189A (en) | Construction engineering project risk prediction method based on one-dimensional convolutional neural network | |
| CN110110480B (en) | Optimized design method of carbon fiber bumper anti-collision beam structure considering layering compatibility | |
| CN114996995A (en) | Metamaterial vibration isolation unit performance forecasting method and system | |
| CN110020466A (en) | Negative poisson's ratio structure energy-absorption box optimization design method based on agent model | |
| CN112068088A (en) | Radar radiation source threat assessment method based on optimized BP neural network | |
| CN111832704A (en) | Design method of convolution input type nested recurrent neural network | |
| CN114239948B (en) | Deep traffic flow prediction method, medium and equipment based on time sequence decomposition unit | |
| CN115577615A (en) | Automobile multi-body dynamics modeling method and system | |
| CN103034754B (en) | The data processing subdivided modeling method of the Decoupling Mode of body lightening design | |
| CN105631093A (en) | M-BSWA multi-target optimization based mechanical structure design method | |
| Li et al. | Lightweight optimal design of a rear bumper system based on surrogate models | |
| Zhang et al. | Geometry parameter optimization method for automobile energy-absorbing box | |
| CN115659694A (en) | Method and system for evaluating and optimizing damping effect of whole vehicle | |
| CN115114819B (en) | Failure surface simulation method of planar woven panel honeycomb sandwich structure based on adaboost algorithm | |
| CN110659820A (en) | Feature selection device for traffic accident analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210420 |
|
| RJ01 | Rejection of invention patent application after publication |