CN102360180B - Method for identifying brake performance of safety monitoring system for motor vehicles - Google Patents

Abstract

The invention discloses a method for identifying the brake performance of a safety monitoring system for motor vehicles. The method comprises the following steps: obtaining operating variables of acceleration, pressure and temperature input and brake performance parameter output; analyzing a redundant information relation between the operating variables by using a principal component analysis method so as to eliminate redundant data; and determining the structure of a recurrent wavelet ELman neural network, and obtaining a plurality of accurate values of brake performance parameters for the operating process by using an identification method. In the invention, through carrying out identification processing on obtained brake performance sensing quantities, accurate sensing information can be provided for a rear-end safety status forecasting module; the method can be used for carrying out identification on operating multi-parameter input information of motor vehicles, therefore, the method has strong learning and generalization capabilities; and by using the method disclosed by the invention, inevitable error accumulation caused by a specific computation model can be shielded, therefore, the accuracy of safe-operating brake performance monitoring parameters of motor vehicles can be effectively improved.

Description

The brake performance of safety monitoring system for motor vehicles discrimination method

Technical field

The present invention relates to automobile safety operational factor monitoring, relate in particular to a kind of information processing method of the automobile safety operation monitoring system braking ability identification based on Recursive Wavelet Elman network.

Background technology

Motor vehicle security of operation Condition Monitoring Technology is the Main Means that guarantees motor vehicle safe drive, is also the inexorable trend of motor vehicle security of operation detection technique development.Adopt motor vehicle security of operation condition monitoring technology to carry out dynamic monitoring to motor vehicle security of operation state and operating index, in time find and the prevention vehicle trouble, development monitoring, control, management and decision are in the safety supervisory network system of one, and be significant to the automobile safety operation.It can not only improve technical guarantee ability, the minimizing traffic hazard of automobile safety operation, and the development that promotes automotive industry and communications and transportation cause is had great meaning.Motor vehicle security of operation status monitoring mainly comprises monitoring motor vehicle (vehicle body, wheel) athletic posture parameter, dynamic load parameters, braking ability parameter.Braking ability is to estimate the most important technical indicator of motor vehicle, is one of elementary item of automotive safety detection.Motor vehicle reruns in process, can produce braking, accelerate, turn to, the operating mode such as straight-line travelling, by monitoring wheel and vehicle performance can obtain more directly, truer, more rich automobile safety operation information.

The gyro free strap down measuring system that motor vehicle running status braking ability monitoring is adopted be a kind of multivariate, non-linear, the time complication system that becomes, when considering alignment error, be installed on the sensitive direction matrix of 12 accelerometers of vehicle body and installation site matrix and all can on itself vector, subtle change be arranged, thereby the angular velocity that calculates and theoretical value have relatively large deviation, and this deviation can constantly be accumulated along with the time.The factor that affects system accuracy be still waiting to solve for other factor that affects system accuracy, but because the model of setting up is complicated, calculated amount is loaded down with trivial details except the impact of alignment error diagonal angle velocity calculated, and it is thorough to be difficult to consider.If do not need to know concrete accurate mathematical model, can consider so to carry out System Discrimination with the method for neural network, approach non-linear between system's input and output.

Wavelet transformation is a kind of signal processing method that grows up on the Fourier analysis basis, wavelet transformation can resolve into the different frequency mixed signal of various weave ins the block signal on different frequency bands, then observe respectively and process in each time-frequency region, all having the ability of characterization signal local feature in time domain and frequency domain.In view of wavelet transformation has good time-frequency local property and zoom character, and the Elman network has self study, strong robustness and Generalization Ability, and modeling just can in conjunction with both advantages, improve the accuracy of modeling wavelet transformation and Elman network integration are got up.Therefore the present invention sets forth a kind of automobile safety operation monitoring system braking ability discrimination method based on Recursive Wavelet Elman network.

The severity of braking parameter of " braking signal system of automobile " of domestic (patent of invention) number CN1605502 by to the motor vehicle high-speed travelling brake time processed classification, and take the graded signal mode to show, clear and definite car brakeing intensity, so that the rear car of running at high speed is taked to brake accordingly counter-measure timely and accurately, avoid rear car blindly to brake the danger that causes.This patent does not relate to the signal processing method of automobile braking performance discrimination.

" brake system in motor vehicle being carried out the method for functional check " of domestic (patent of invention) number CN102112351A relates to a kind of for the brake system of motor vehicle being carried out the method for functional check, at first brake fluid is moved in apotheca in the method, measure subsequently the brake pressure in brake circuit and this brake pressure is compared with reference value, wherein in the situation that produce alerting signal lower than reference value.This patent does not relate to the signal processing method of automobile braking performance identification.

Domestic (patent of invention) number CN101480946 that proposed in 2008 " a kind of based on the wheel-loaded intelligent sensing wheel brake performance monitoring methods " discloses a kind of based on the wheel-loaded intelligent sensing wheel brake performance monitoring methods, by installation intelligent sensing module on the wheel hub equatorial plane of each wheel of motor vehicle with in vehicle body installation intelligent sensing unit, transducing signal calculates through signal condition, digitizing, Attitude Algorithm, braking algorithm the main braking ability parameter that obtains wheel.The method can initiatively be estimated motor vehicle sports safety situation by parameter is carried out Multi-Sensor Data Fusion and analysis.This patent content relates to based on braking accurate model mode and carries out the braking ability calculation of parameter, the identification signal New Method for Processing based on the non-concrete model of Recursive Wavelet Elman network that does not relate to that this patent proposes.

Domestic (patent of invention) number CN101718769A " a kind of source drive-type 235U concentration recognition method based on parallel type genetic Elman neural network " mainly comprises and sets up neural network model, and Artificial Neural Network Structures is divided into three layers: data allocations layer, subnet layer, decision making package layer; The neutron pulse signal auto-correlation function of obtaining is done pre-service; Data allocations layer with the signal auto-correlation function sample input parallel type genetic Elman network after processing adopts the random multiple spot sampling of circulation that sample data is distributed; Each hereditary Elman subnet that the data that distribute are inputted respectively in the subnet layer is identified, and provided recognition result separately; The decision making package layer is made joint disposal by the recognition result to a plurality of subnets, draws the final recognition result of 235U concentration.The method is because of its higher data user rate, and novel network structure has obtained 235U concentration identification effect preferably.This patent does not relate to the signal processing method of Recursive Wavelet Elman network.

Summary of the invention

For solving because of factor insoluble error problems that parameter measurement brings to braking ability such as installation, sensing demarcation, the invention provides a kind of brake performance of safety monitoring system for motor vehicles discrimination method.Non-concrete model identification disposal route based on Recursive Wavelet Elman network can effectively improve the accuracy of braking ability measuring multiple parameters value.Described technical scheme is as follows:

The brake performance of safety monitoring system for motor vehicles discrimination method comprises:

A obtains acceleration, pressure, temperature input and braking ability parameter output function variable;

B eliminates data redundancy by redundant information relation between principal component analysis (PCA) analysis operation variable;

C determines Recursive Wavelet ELman neural network structure, and obtains a plurality of braking ability of operational process accurately parameter values by discrimination method.

The beneficial effect of technical scheme provided by the invention is:

The Recursive Wavelet Elman network identification method of setting up according to the method has generality, also exist similar problem and move other parameter at automobile safety, therefore the method can also be generalized to motor vehicle and move other parameter detecting, as aspects such as attitude, dynamic loadings.

Description of drawings

Fig. 1 is based on Recursive Wavelet Elman network brake performance of safety monitoring system for motor vehicles discrimination method modeling process flow diagram;

Fig. 2 is based on the hereditary training algorithm process flow diagram in Recursive Wavelet Elman network brake performance of safety monitoring system for motor vehicles discrimination method;

Fig. 3 is Recursive Wavelet Elman schematic network structure;

Fig. 4 a, 4b and 4c are based on the brake performance of safety monitoring system for motor vehicles discrimination method of Recursive Wavelet Elman network and implement platform illustration in kind.

Embodiment

For making the purpose, technical solutions and advantages of the present invention clearer, embodiment of the present invention is described further in detail below in conjunction with accompanying drawing:

The present embodiment provides a kind of method of brake performance of safety monitoring system for motor vehicles discrimination method.

Referring to Fig. 1, the method is based on Recursive Wavelet Elman and realizes, comprises the following steps:

Step 101 obtains acceleration, pressure, temperature input and braking ability parameter output function variable;

Step 102 is eliminated data redundancy by redundant information relation between principal component analysis (PCA) analysis operation variable.

Step 103 input data normalization is processed;

Step 104 is utilized experimental formula and experiment algorithm computational grid model hidden layer node number;

In order to make model that better approximation capability and generalization ability be arranged, adopt experimental formula and experiment to choose the method that combines and determine choosing of hidden layer neuron number, at first according to experimental formula

Be input number of nodes, n is the output node number, and l is the constant between 1-20) determine the scope of the number of a hidden layer neuron, set corresponding train epochs and training precision, then optimum hidden layer neuron number is therefrom chosen at last in preference pattern training from small to large.

Step 105 Recursive Wavelet Elman neural network model parameter initialization:

Initial setting up

Initial setting up specifically be divided into for four steps and realize:

1. produce at first at random equally distributed random number conduct on [1,1] interval

The initial setting up value;

2. then right

Carry out normalization by row:

${\mathrm{<figure-callout\; id="2"\; label="W\; ij"\; filenames="HSA00000584101000032.png"\; state="\{\{state\}\}">W</figure-callout>}}_{\mathrm{ij}}^{}=\frac{{\mathrm{<figure-callout\; id="2"\; label="W\; ij"\; filenames="HSA00000584101000032.png"\; state="\{\{state\}\}">W</figure-callout>}}_{\mathrm{ij}}^{}}{\sqrt{\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left(W_{\mathrm{ij}}^2\right)}^2}},i=1,2,\&CenterDot;\&CenterDot;\&CenterDot;,m---\left(1\right)$

3. then multiply by one again and count n with input layer, hidden layer neuron is counted m and the relevant factor of transport function:

${\mathrm{<figure-callout\; id="2"\; label="W\; ij"\; filenames="HSA00000584101000032.png"\; state="\{\{state\}\}">W</figure-callout>}}_{\mathrm{ij}}^{}=C*n^{\frac{1}{m}}*{\mathrm{<figure-callout\; id="2"\; label="W\; ij"\; filenames="HSA00000584101000032.png"\; state="\{\{state\}\}">W</figure-callout>}}_{\mathrm{ij}}^{},i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,m---\left(2\right)$

In formula, C is a constant relevant with the hidden layer transport function.The value of C is extremely important to network, through study practice repeatedly relatively, to the proper value of Morlet small echo Elman neural network between 2.3～2.6;

4. last and training sample contacts.If in input layer j neuronic input sample, maximal value is x _Jmax, minimum value is x _Jmin:

$W_{\mathrm{<figure-callout\; id="2"\; label="ij"\; filenames="HSA00000584101000032.png"\; state="\{\{state\}\}">ij</figure-callout>}}^2=\frac{2*W_{\mathrm{ij}}^2}{x_{j\max }-x_{j\min }},i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,m---\left(3\right)$

Obtain by above step

Be input layer to the initial weight of hidden layer.

Initial setting up

Obtain

After, and then carry out the hidden layer neuron threshold value

Initialization, step is as follows:

1. produce at first at random equally distributed random number conduct on [1,1] interval

The initial setting up value;

2. then multiply by an input layer again and count n, hidden layer neuron is counted m and the relevant factor of transport function;

$W_{i0}^2=C*n^{\frac{1}{m}}*W_{i0}^2,i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,m---\left(4\right)$

3. last and training sample reaches

Contact.

$W_{i0}^2=W_{i0}^2-0.5*\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{W}_{\mathrm{ij}}^2(x_{j\max }+x_{j\min }),j=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,m---\left(5\right)$

(3) initial setting up of the flexible and translation parameters of small echo

After being provided with initial weight, it is also very important that the flexible translation parameters of small echo is carried out initial setting up, generally is divided into two kinds of situations.

1. the input layer number is 1.To the neuronic flexible parameter a of each small echo _iGet identical value, and translation parameters b _i=(i-1) L/m, i=1,2 ..., m.In formula, L is number of training, and m is the neuron number of transport function place layer.

2. the nodes of input layer is greater than 1.Known by the small echo basic theory, if the time domain center of female small echo is t ^*, radius is Δ _ΨThe flexible concentrated area that ties up to time domain of small echo is:

[b+at ^*-aΔ _Ψ，b+at ^*+aΔ _Ψ]

Be the gamut that covers input vector in order to make small echo flexible, the initial value setting of flexible translation parameters must be satisfied following formula:

$b_i+a_i{t}^{*}-a_i{\mathrm{\&Delta;}}_{\mathrm{\&psi;}}=\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{W}_{\mathrm{ij}}^2{x}_{i\min }---\left(6\right)$

$b_i+a_i{t}^{*}+a_i{\mathrm{\&Delta;}}_{\mathrm{\&psi;}}=\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{W}_{\mathrm{ij}}^2{x}_{i\max }---\left(7\right)$

Can be obtained by following formula:

$a_i=\frac{\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{W}_{\mathrm{ij}}^2{x}_{i\max }-\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{W}_{\mathrm{ij}}^2{x}_{i\mathrm{<figure-callout\; id="2"\; label="min"\; filenames="HSA00000584101000032.png"\; state="\{\{state\}\}">min</figure-callout>}}}{2{\mathrm{\&Delta;}}_{\mathrm{\&psi;}}}---\left(8\right)$

$b_i=\frac{\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{W}_{\mathrm{ij}}^2{x}_{i\max }({\mathrm{\&Delta;}}_{\mathrm{\&psi;}}-t^{*})+\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{W}_{\mathrm{ij}}^2{x}_{i\min }({\mathrm{\&Delta;}}_{\mathrm{\&psi;}}+t^{*})}{2{\mathrm{\&Delta;}}_{\mathrm{\&psi;}}}---\left(9\right)$

Will use time domain center and the radius of female small echo in following formula, the center that can calculate the time-domain window of Morlet small echo according to the definition of Wavelet time-frequency parameter is 0, and radius is 0.7071.

Initial setting up

And for the weights of hidden layer to output layer Initial setting up, can complete as follows:

1. produce at first at random equally distributed random number conduct on [1,1] interval

The initial setting up value;

2. then right

Carry out normalization by row:

${\mathrm{<figure-callout\; id="1"\; label="W\; ki"\; filenames="HSA00000584101000031.png,HSA00000584101000032.png"\; state="\{\{state\}\}">W</figure-callout>}}_{\mathrm{ki}}^{}=\frac{{\mathrm{<figure-callout\; id="1"\; label="W\; ki"\; filenames="HSA00000584101000031.png,HSA00000584101000032.png"\; state="\{\{state\}\}">W</figure-callout>}}_{\mathrm{ki}}^{}}{\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}{\left(W_{\mathrm{ki}}^1\right)}^2}},k=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,p---\left(10\right)$

Step 106 builds Recursive Wavelet Elman network model;

Step 107 initialization weights also carry out self-correlation correction;

Step 108 is utilized genetic algorithm training small echo Elman network;

Judge whether small echo Elman network satisfies the setting demand, if so, execution in step 109; Otherwise adjustment training parameter.

The generalization ability of step 109 checking Recursive Wavelet Elman network model;

Judge whether generalization ability reaches requirement, is, execution in step 110; Otherwise, return to execution in step 108.

After step 110 reaches requirement, storage Recursive Wavelet Elman network model parameter.

Above-mentioned braking ability parameter comprises: the parameter such as wheel braking retarded velocity, damping force, brake power are poor, wheel slip, car load braking deceleration and car load damping force.

Figure 2 shows that based on the hereditary training algorithm flow process in Recursive Wavelet Elman network brake performance of safety monitoring system for motor vehicles discrimination method, comprise the following steps:

Contraction-expansion factor and the shift factor of step 1081 pair weights, threshold value, wavelet function are carried out real coding;

The random initial population that produces between [1,1] of step 1082;

The error function of step 1083 computational grid, the functional value of the fitness of definite individuality, error is larger, and adaptive value is less;

If step 1084 end condition satisfies, turn step 1087;

Step 1085 selects the individuality of some fitness maximums directly to entail the next generation, and all the other probability of determining by adaptive value carry out heredity;

Step 1086 is carried out the crossover and mutation operation by certain probability, produces population of future generation;

Step 1087 stops circulation, obtains best chromosome.Then be converted into contraction-expansion factor and the shift factor of weights, threshold value, wavelet function.

Be illustrated in figure 3 as Recursive Wavelet Elman network structure, the Morlet small echo excitation function that Recursive Wavelet Elman network structure adopts, each hidden neuron has two additional parameter a _i, b _i, represent respectively contraction-expansion factor and the shift factor of small echo action function.This Recursive Wavelet Elman neural network has n+1 input neuron, and p output neuron accepted a layer neuron for m hidden neuron and m.x ₀(t)=1, corresponding network weight

Representing the threshold value of hidden neuron i, is not actual outside input.H ₀(t)=1, the map network weights

Representing the threshold value of output neuron k, is not actual outside input.

x _i(t) ∈ R ⁿInscribe Recursive Wavelet Elman neural network n dimension input vector during for t;

y _k(t) ∈ R ^pInscribe the p dimension output vector of Recursive Wavelet Elman neural network during for t;

H _i(t) ∈ R ^mBe hidden layer output;

x _c(t) ∈ R ^mFor accepting layer output.

If feedback gain is α, namely each constantly accept the layer neuron be output as:

x _c，l(t)＝αH _l(t-1) (11)

The dynamic equation of these network input and output this moment can be expressed as:

Wherein:

Be the weights that are connected between hidden neuron i and output layer neuron k;

Be the weights that are connected between input layer j and hidden neuron i;

For accepting the weights that are connected between layer neuron l and hidden neuron i;

H _i(t) be the output of hidden neuron i;

Be wavelet function, a _iBe the coefficient of dilatation of small echo, b _iTranslation coefficient for small echo.

Hidden layer neuron number: i=1,2 ..., m; Input layer number: j=0,1,2 ..., n,

Accept layer neuron number a: l=1,2 ..., m; The neuron number of output layer: k=1,2 ..., p.

As shown in Fig. 4 a, 4b and 4c, implement platform in kind based on the brake performance of safety monitoring system for motor vehicles discrimination method of Recursive Wavelet Elman network.Fig. 4 a comprises central control module 3 in wheel 1, wheel intelligent sensing module 2, car, wherein wheel intelligent sensing module comprises three accelerometers, a pressure transducer and a temperature sensor, and wheel intelligent sensing module is installed on the surface, wheel hub equator of each wheel, and in car, central control module 3 is installed in car; Realize two-way communication by the Zigbee wireless radio frequency mode between central control module 3 in wheel intelligent sensing module 2 and car.In Fig. 4 b, stain is four of vehicle bodies three-dimensional acceleration intelligent sensing module installation site, adopts square four apex configuration schemes.Fig. 4 c is the One-dimensional Vertical accelerometer that is installed in one of them suspension of motor vehicle.

The above is only preferred embodiment of the present invention, and is in order to limit the present invention, within the spirit and principles in the present invention not all, any modification of doing, is equal to replacement, improvement etc., within all should being included in protection scope of the present invention.

Claims (5)

1. the brake performance of safety monitoring system for motor vehicles discrimination method, is characterized in that, described method comprises:

Steps A obtains acceleration, pressure, temperature input and braking ability parameter output function variable;

Step B eliminates data redundancy by redundant information relation between principal component analysis (PCA) analysis operation variable;

Step C determines Recursive Wavelet ELman neural network structure, and obtains a plurality of braking ability of operational process accurately parameter values by discrimination method;

Described step C specifically comprises:

Initialization Recursive Wavelet ELman neural network model parameter;

Build Recursive Wavelet ELman neural network model;

The initialization weights also carry out self-correlation correction;

Utilize genetic algorithm training Recursive Wavelet ELman neural network;

The generalization ability of checking Recursive Wavelet ELman neural network model;

Storage Recursive Wavelet ELman neural network model parameter.

2. brake performance of safety monitoring system for motor vehicles discrimination method according to claim 1, is characterized in that, described Recursive Wavelet ELman neural network structure adopts the type combination of compacting.

3. brake performance of safety monitoring system for motor vehicles discrimination method according to claim 1, is characterized in that, described genetic algorithm comprises:

Contraction-expansion factor and shift factor to weights, threshold value, wavelet function are carried out real coding;

The random initial population that produces between [1,1];

The error function of computational grid, the functional value of the fitness of definite individuality;

If end condition satisfies, stop circulation, obtain chromosome; Otherwise

Select the large individuality of some fitness directly to entail the next generation, all the other probability of determining by adaptive value carry out heredity;

Carry out the crossover and mutation operation by certain probability, produce population of future generation;

Stop circulation, obtain best chromosome, then be converted into contraction-expansion factor and the shift factor of weights, threshold value, wavelet function.

4. brake performance of safety monitoring system for motor vehicles discrimination method according to claim 1, it is characterized in that, determine also to comprise before Recursive Wavelet ELman neural network structure: determine the hidden layer neuron number, with the excitation function of Morlet small echo as hidden layer neuron; Determine also to comprise by orthogonal experimental method after Recursive Wavelet ELman neural network structure and obtain the Experiment Training sample data, these training sample data comprise training sample and desired output sample.

5. brake performance of safety monitoring system for motor vehicles discrimination method according to claim 4, is characterized in that, described hidden layer neuron number is chosen and adopted experimental formula and experiment to choose the method that combines, according to experimental formula

Determine the number range of a hidden layer neuron, and set corresponding train epochs and training precision, the hidden layer neuron number is determined in then preference pattern training from small to large; Described m is input number of nodes, and n is the output node number, and l is the constant between 1-20.

Images