CN1750010A - Computer auxiliary automobile chassis type selecting method - Google Patents
Computer auxiliary automobile chassis type selecting method Download PDFInfo
- Publication number
- CN1750010A CN1750010A CN 200510061027 CN200510061027A CN1750010A CN 1750010 A CN1750010 A CN 1750010A CN 200510061027 CN200510061027 CN 200510061027 CN 200510061027 A CN200510061027 A CN 200510061027A CN 1750010 A CN1750010 A CN 1750010A
- Authority
- CN
- China
- Prior art keywords
- output
- chassis
- neural network
- type selecting
- nodes
- 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
Images
Abstract
The present invention relates to computer aided automobile chassis type selecting method. After inputting known conditions of the chassis and through nerve network algorithm to simulate expert thought, is established single hidden layer and inverse propagating nerve network possessing input layer with 13 nodes, hidden layer with 10 nodes and output layer with 22 nodes. On the basis of automobile type database with great amount of data, proper specimen database is selected, and through statistics, analysis, nerve network training and reasoning on the assemblies of destination automobile, the theoretical results of the chassis structure form and other parameters are obtained. Then, the type selection of various chassis assemblies is realized and the key parameters are determined. The present invention provides effective computer support on the automobile chassis type selecting design.
Description
Affiliated technical field
The present invention relates to the artificial intelligence field in the computer technology, mainly is the computer assisted automobile chassis type selecting method of using artificial intelligence technologys such as neural network.
Background technology
Traditional chassis selection method is to rely on deviser's rich knowledge and experience to guarantee the rationality that designs.But, various factors is carried out ten minutes at the initial stage of design and analyze all sidedly and consider that select suitable chassis each total formation type and key parameter, difficulty is quite big.And each new-type in recent years chassis assembly mechanism constantly occurs, and each novel chassis assembly mechanism constantly is used, and the working method that relies on deviser's personal experience to carry out type selecting and design merely is worthless.
Summary of the invention
Purpose of the present invention will overcome above-mentioned deficiency exactly, and a kind of computer assisted automobile chassis type selecting method is provided.Advanced artificial neural network algorithm has had lot of data for those, but the problem that its inherent laws is difficult to find the accurate mathematical description, have unique self-learning capability, can extract through the artificial neural network of learning and train and contain inner mapping and contact in data.Therefore, as support, adopt artificial neural network to carry out the trial of each assembly type selecting of chassis, important practice value is arranged in practical engineering application with up-to-date model data storehouse.
The technical solution adopted for the present invention to solve the technical problems is: this computer assisted automobile chassis type selecting method, import certain chassis known conditions, by neural network algorithm simulation expert thinking, set up single hidden layer reverse transmittance nerve network (BP net), input layer is 13 nodes, hidden layer is 10 nodes, and output layer is 22 nodes.On the basis of the database that has a large amount of model datas, select appropriate sample database, each assembly of target vehicle is added up, analyzed, carry out neural metwork training, reasoning in addition, draw The reasoning results, and on this basis, realize the type selecting and the key parameter determination of each assembly of chassis for chassis structure pattern and other parameter.
The invention has the beneficial effects as follows, describe inadequately automobile chassis type selecting design problem fully, effective computing machine support is provided for condition in the design.
Description of drawings
Fig. 1 is the neuron models of artificial neural network elementary cell.
Fig. 2 is the artificial neural network learning synoptic diagram.
Embodiment
The present invention is further described below in conjunction with accompanying drawing and example.This automobile chassis type selecting method based on neural network of the present invention the steps include:
1) sets up the artificial neuron meta-model
Figure 1 shows that the neuron models of artificial neural network elementary cell, it has three fundamentals: one group of connection weight (corresponding to the cynapse of biological neuron), the weights that strength of joint is connected by each represent that weights are for just representing excitation, for negative indication suppresses.A sum unit is used to ask for the weighted sum (linear combination) of each input information.A non-linear excitation function plays the Nonlinear Mapping effect and limits the neuron output amplitude (generally being limited between [0,1] or [1 ,+1]) within certain scope.Also has a threshold value θ in addition
k(or biasing b
k=-θ
k).More than effect can be expressed as with mathematical expression
v
k=u
k-θ
k (1)
y
k=(v
k)
X in the formula
1, x
2, x
3..., x
pBe input signal, w
K1, w
K2..., w
KpBe the weights of neuron k, u
kBe linear combination result, θ
kBe threshold value. () is an excitation function, y
kOutput for neuron k.In the present invention, excitation function () adopts the Sigmoid function, and this function has level and smooth and gradual, and keeps monotonicity, and its functional form is
Its slope of parameter alpha may command wherein.
2) determine network structure
The topological structure of neural network mainly contains two kinds from connected mode: feed-forward type network and feedback-type network, wherein the action effect of feed-forward type network mainly shows as Function Mapping, has the stronger pattern-recognition and the function of approximation of function.The present invention adopts the feed-forward type neural network, is similar to the expert and equally some specific data is approached with analogy and infer in the hope of reaching.
3) determine the learning method of artificial neural network
Adopt supervised learning (having the tutor to learn) in the present invention, this mode of learning need provide due output result to given one group input, known input-output the result of this group is called training sample set, and the learning system of neural network can be come the regulating system parameter according to the difference between known output and the actual output.
The rule of study adopts the delta learning rules, is also referred to as error correction study.Its mathematical expression is
e
k(n)=d
k(n)-y
k(n) (3)
Y wherein
k(n) be input as x for working as
k(n) time, neuron k is at n actual output constantly, d
k(n) represent due output (can provide) by training sample.
The final purpose of error correction study be make a certain based on e
k(n) it is minimum that objective function reaches, so that the actual output of each output unit approaches on certain statistical significance output should be arranged in the network.In case selected the objective function form, error correction study has just become a typical optimization problem, and objective function is that square error is minimum, and the average that therefore defines error sum of squares is
Wherein E is an expectation operator, and the process of the study of making demands before the following formula is stably wide, and concrete method can be with optimum gradient descent method.Be the statistical property that to know whole process as objective function directly,, adopt J to replace J, promptly at the instantaneous value ξ of moment n (n) for addressing this problem with J
Problem becomes asks the minimal value of ξ (n) to weight w, can get according to the gradient descent method
Δw
kj=ηe
k(n)x
j(n) (6)
Wherein η is the study step-length.
In the present invention, the neural network of being set up is wished and can be reached as the expert by training, can offer a kind of satisfied answer of user.In the initial design process, only know that a spot of parameter just can determine the major parameter in the car load design.Establishing network in neural network specifically has p input, q output, and then its effect can be regarded as by the Nonlinear Mapping of p dimension Euclidean space to q dimension Euclidean space.Make the monotone increasing continuous function of (), I for bounded, non-constant
pRepresent p dimension unit hypercube [0,1]
p, C (I
p) expression is defined in I
pThe set that last continuous function constitutes, then given any function f ∈ C (I
p) and ε>0, there are integer M and one group of real constant α
i, θ
lAnd w
Ij, i=1 wherein ... M, j=1,2 ..., p is output as network
Approximating function f () arbitrarily, promptly
|F(x
1,x
2,…,x
p)-f(x
1,x
2,…,x
p)|<ε (8)
(x
1,…,x
p)∈I
p
The feedforward network that the The above results explanation only contains a hidden layer is that a kind of general purpose function approaches device, is enough for approaching hidden layer of a continuous function.
It is early existing that multitiered network can solve this conclusion of non-linear separable problem, because study was relatively more difficult after hidden layer was arranged, limited the development of multitiered network.Solved this difficulty by backpropagation (Back Propagation) algorithm.This method wherein has two kinds of signals circulating:
Working signal (representing with solid line), it is to propagate forward up to producing the signal of actual output at output terminal after applying input signal, is the function of input and weights.
Error signal (dotting), network is actual to be exported and should have the difference between output and be error, and it is begun successively to propagate backward by output terminal.
J the unit that is located at output terminal in the n time iteration is output as y
j(n), then the error signal of this unit is
e
j(n)=d
j(n)-y
j(n) (9)
The square error of definition unit j is
Then the instantaneous value of the total square error of output terminal is
Wherein c comprises all output units.If total sample number is N in the training set, then the average of square error is
ξ
AVBe the objective function of study, the destination of study should make ξ
AVReach minimum, ξ
AVIt is the function of all weights of network and threshold value and input signal.Below the situation of the sample learning BP algorithm of deriving one by one just, j unit receives last layer signal and produces the process of error signal, makes being input as only of unit j:
P is added to the number that unit j goes up input, then has
y
j(n)=
j(v
j(n)) (13)
Ask ξ (n) to w
IjGradient
Because
Weight w
JiCorrection be
Negative sign represents that correction presses the gradient descent direction, wherein
Be called partial gradient.Discuss in two kinds of situation below:
Unit j is an output unit, then
δ
i(n)=(d
j(n)-y
j(n))
j′(v
j(n)) (19)
Unit j is a hidden unit, then
When k has during for output unit
With this formula to y
j(n) differentiate,
Because e
k(n)=d
k(n)-y
k(n)=d
k(n)-
k(v
k(n)), so
And
Wherein q is the input end number of unit k.This formula is to y
j(n) differentiate,
So have
J is a hidden unit.
According to above derivation, weight w
JiCorrection can be expressed as
δ
j(n) calculating has two kinds of situations:
When j is an output unit, δ
j(n) be
j' (vj (n)) and error signal e
j(n) long-pending.
When j is a hidden unit, δ
j(n) be
jThe weighting product of sum of the δ of ' (vj (n)) and back one deck.
4) set up artificial neural network
In the present invention, set up single hidden layer reverse transmittance nerve network (BP net) (Learning Principle is seen Fig. 2), input layer is 13 nodes, and hidden layer has 10 nodes, and output layer has 22 nodes, and the advantage of BP net algorithm is that the algorithm derivation is clear, and learning accuracy is higher.According to selected sample database, carry out neural metwork training.The parameter of input during training, output sees Table 1.With finishing normalized data neural network is trained, it is 5E-4 that allowable error is set, to reduce the error of returning actual value after wheelbase and wheelspan calculate.The selection of sample is according on the various related data bases of collecting, and the corresponding data of comprehensive domestic car and determined to screen in nearly thousand kinds of vehicles by the statistical study of all types of vehicles, is selected representational vehicle.
Sequence number | The field meaning | I/O | |
1 | Car load | Length overall | Input |
2 | Beam overall | Input | |
3 | Height overall | Input |
4 | Car category | Input | ||
5 | Gross mass | Input | ||
6 | Wheelspan | Output | ||
7 | Wheelbase | Output | ||
8 | Rear wheel tire number | Output | ||
9 | Tire radius | Output | ||
10 | Index | Top speed | Input | |
11 | Cost | Input | ||
12 | Engine | Arrange | Input | |
13 | Direction | Input | ||
14 | Shape | Input | ||
15 | Discharge capacity | Output | ||
16 | Power | Output | ||
17 | Maximum speed | Output | ||
18 | Drive form | Input | ||
19 | Power train | Variator shelves number | Output | |
20 | Base ratio | Output | ||
21 | Running gear | Suspension form | Before | Input |
22 | After | Input | ||
23 | Flexible member | Front suspension | Output | |
24 | Rear suspension | Output | ||
25 | Front alignment | Kingpin inclination | Output | |
26 | Reverse caster | Output | ||
27 | Camber | Output | ||
28 | The trailing wheel location | Kingpin inclination | Output | |
29 | Reverse caster | Output | ||
30 | Camber | Output | ||
31 | Braking system | The braking system type | Input | |
32 | The front brake type | Output | ||
33 | The rear brake type | Output | ||
34 | Turn to and be | Radius of turn | Output | |
35 | Power steering | Output |
Table 1
Claims (3)
1, a kind of computer assisted automobile chassis type selecting method, it is characterized in that: import certain chassis known conditions, by neural network algorithm simulation expert thinking, set up single hidden layer reverse transmittance nerve network, on the basis of the database that has a large amount of model datas, select sample database, each assembly of target vehicle is added up, analyzed, carry out neural metwork training, reasoning in addition, draw The reasoning results, and on this basis, realize the type selecting and the key parameter determination of each assembly of chassis for chassis structure pattern and other parameter.
2, computer assisted automobile chassis type selecting method according to claim 1 is characterized in that: the steps include:
1) sets up the artificial neuron meta-model
It has three fundamentals: one group of connection weight, the weights that strength of joint is connected by each represent that weights are the negative indication inhibition for just representing excitation, and a sum unit is used to ask for the weighted sum of each input information; A non-linear excitation function plays the Nonlinear Mapping effect and limits the neuron output amplitude within certain scope; Also has a threshold value θ in addition
k, more than effect can be expressed as with mathematical expression
v
k=u
k-θ
k (1)
y
k=(v
k)
X in the formula
1, x
2, x
3..., x
pBe input signal, w
K1, w
K2..., w
KpBe the weights of neuron k, u
kBe linear combination result, θ
kBe threshold value. () is an excitation function, y
kOutput for neuron k;
2) determine network structure: adopt the feed-forward type neural network;
Determine the learning method of artificial neural network: the learning method of determining artificial neural network: adopt the supervised learning method, given training sample set, learning system is come the regulating system parameter according to the difference between known output and the actual output, and the rule of study adopts the delta learning rules
e
k(n)=d
k(n)-y
k(n)
Y wherein
k(n) be input as x for working as
k(n) time, neuron k is at n actual output constantly, d
k(n) represent due output, can provide by training sample;
3) set up artificial neural network:
Set up single hidden layer reverse transmittance nerve network, input layer is 13 nodes, and hidden layer has 10 nodes, and output layer has 22 nodes, according to selected sample database, carries out neural metwork training; With finishing normalized data neural network is trained, it is 5E-4 that allowable error is set, the selection of sample is according on the various related data bases of collecting, the corresponding data of comprehensive domestic car and determined to screen in nearly thousand kinds of vehicles, by the statistical study of all types of vehicles, select representational vehicle.
3, computer assisted automobile chassis type selecting method according to claim 2 is characterized in that:
Excitation function () adopts the Sigmoid function, and this function has level and smooth and gradual, and keeps monotonicity, and its functional form is
Its slope of parameter alpha may command wherein.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 200510061027 CN1750010A (en) | 2005-10-09 | 2005-10-09 | Computer auxiliary automobile chassis type selecting method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 200510061027 CN1750010A (en) | 2005-10-09 | 2005-10-09 | Computer auxiliary automobile chassis type selecting method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN1750010A true CN1750010A (en) | 2006-03-22 |
Family
ID=36605449
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 200510061027 Pending CN1750010A (en) | 2005-10-09 | 2005-10-09 | Computer auxiliary automobile chassis type selecting method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN1750010A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101441728B (en) * | 2007-11-21 | 2010-09-08 | 新乡市起重机厂有限公司 | Neural network method of crane optimum design |
CN102792234A (en) * | 2010-03-12 | 2012-11-21 | 西门子公司 | Method for the computer-aided control of a technical system |
CN103455521A (en) * | 2012-06-04 | 2013-12-18 | 珠海格力电器股份有限公司 | Alternating-current contactor type selection processing method and system |
CN107480779A (en) * | 2017-08-29 | 2017-12-15 | 胡明建 | Design method that is a kind of while exporting polymorphic function artificial neuron |
CN107480778A (en) * | 2017-08-29 | 2017-12-15 | 胡明建 | Design method that is a kind of while exporting polymorphic function artificial neuron |
CN109117511A (en) * | 2018-07-17 | 2019-01-01 | 常州大学 | A kind of screw pump production system design method based on yield |
CN111325146A (en) * | 2020-02-20 | 2020-06-23 | 吉林省吉通信息技术有限公司 | Truck type and axle type identification method and system |
-
2005
- 2005-10-09 CN CN 200510061027 patent/CN1750010A/en active Pending
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101441728B (en) * | 2007-11-21 | 2010-09-08 | 新乡市起重机厂有限公司 | Neural network method of crane optimum design |
CN102792234A (en) * | 2010-03-12 | 2012-11-21 | 西门子公司 | Method for the computer-aided control of a technical system |
CN102792234B (en) * | 2010-03-12 | 2015-05-27 | 西门子公司 | Method for the computer-aided control of a technical system |
CN103455521A (en) * | 2012-06-04 | 2013-12-18 | 珠海格力电器股份有限公司 | Alternating-current contactor type selection processing method and system |
CN107480779A (en) * | 2017-08-29 | 2017-12-15 | 胡明建 | Design method that is a kind of while exporting polymorphic function artificial neuron |
CN107480778A (en) * | 2017-08-29 | 2017-12-15 | 胡明建 | Design method that is a kind of while exporting polymorphic function artificial neuron |
CN109117511A (en) * | 2018-07-17 | 2019-01-01 | 常州大学 | A kind of screw pump production system design method based on yield |
CN111325146A (en) * | 2020-02-20 | 2020-06-23 | 吉林省吉通信息技术有限公司 | Truck type and axle type identification method and system |
CN111325146B (en) * | 2020-02-20 | 2021-06-04 | 吉林省吉通信息技术有限公司 | Truck type and axle type identification method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN1750010A (en) | Computer auxiliary automobile chassis type selecting method | |
CN1043543C (en) | Criterion method and conjecture device for conjecture rule | |
CN108520155B (en) | Vehicle behavior simulation method based on neural network | |
CN110936824B (en) | Electric automobile double-motor control method based on self-adaptive dynamic planning | |
CN110443359A (en) | Neural network compression algorithm based on adaptive combined beta pruning-quantization | |
CN112801143B (en) | Steering road feel simulation method based on K-Means and Gaussian process regression | |
CN105136469A (en) | Unmanned vehicle speed control method based on PSO and RBF neutral network | |
CN107480395B (en) | Method and system for constructing vehicle steering knuckle load spectrum prediction model | |
CN112046489B (en) | Driving style identification algorithm based on factor analysis and machine learning | |
CN113204920B (en) | Intelligent vehicle lane change comfort evaluation and track planning method and device based on support vector machine | |
CN105034986A (en) | On-line identification method and on-line identification device for steering characteristics of drivers | |
Wang et al. | A holistic method of complex product development based on a neural network-aided technological evolution system | |
Liu et al. | Multi-objective optimization study of regenerative braking control strategy for range-extended electric vehicle | |
CN111507365A (en) | Confidence rule automatic generation method based on fuzzy clustering | |
CN112632706B (en) | Road feel simulation method based on GMM and CART regression tree | |
CN104677629A (en) | Fault detection method for vehicle transmission | |
CN112749508B (en) | Road feel simulation method based on GMM and BP neural network | |
CN106647279B (en) | A kind of locomotive smart steering optimized calculation method based on fuzzy rule | |
CN109214500B (en) | Transformer fault identification method based on hybrid intelligent algorithm | |
CN104680023B (en) | Oil pumper parameter optimization method based on multiobjective decision-making | |
CN112550450B (en) | Road feel simulation method based on K-Means and CART regression tree | |
Puma-Benavides et al. | Influence of the final ratio on the consumption of an electric vehicle under conditions of standardized driving cycles | |
CN109408896B (en) | Multi-element intelligent real-time monitoring method for anaerobic sewage treatment gas production | |
CN101540504B (en) | Current analytical device and method on basis of step-length variable neural network | |
CN115492928B (en) | Economical efficiency, dynamic performance and safety comprehensive optimal gear shifting rule optimization method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |