CN102779287A - Ink key opening forecasting method having increment type learning capacity - Google Patents

Abstract

The invention relates to a method for presetting digital printing ink and provides an ink key opening forecasting method based on a Fuzzy ART-BP mixed neural network algorithm of a Fuzzy ART neural network and a BP neural network. An input vector is stably classified by fully utilizing the self-learning, self-organizing and information fuzzy processing capacities of the Fuzzy ART neural network by the network; for each classification, nonlinear mapping is performed on the input and output vectors of a training sample by utilizing the BP neural network, namely, a mapping relation between image-text digital information and ink key control parameter of the training sample is established by taking printing on-site temperature, humidity, a printer rotating speed and a website area rate corresponding to an ink area as input vectors and an ink key opening value as an output vector; and a converged network is used for forecasting the ink key opening value of a new sample. The network learning is higher in pertinence, the iteration times of the BP network is reduced, the network has the increment type learning capacity and the generalization of the network is increased.

Description

A kind of black key aperture Forecasting Methodology with incremental learning ability

Technical field

The invention belongs to digital printed field, be specifically related to a kind of black key aperture Forecasting Methodology with incremental learning ability.

Background technology

Increase along with what digital document in the preceding field of seal used, digital flow process is applied to typography more and more, and the effect of digital flow process in CTP (CTP) technology is more and more important.Simultaneously the short slab that faces of printing enterprise, complicacy and fast print production live and get more and more, so also printing enterprise is had higher requirement.Concerning printing enterprise, shorten an effective way of printing setup time and exactly printing ink is provided with in advance.Pre-estimate black groove the best and go out China ink amount and carry out Mo Jian and preset, can practice thrift a lot of printing machines start setup time, reduce production costs, improve printing quality and efficient, can also significantly reduce the waste of paper.

The method of printing machine control ink usage is actually and is divided into a lot of long and narrow zone-ink-covered area or black roads to printable part on the forme in the direction of vertical long side; The China ink amount of each ink-covered area is can how much accurately regulating according to the shared area percentage-dot area percentage of picture and text part in this ink-covered area area on the forme; The area percentage of picture and text part is high more, and the China ink amount that then needs is many more.Ink pre-setting is exactly before beginning to print, to obtain the information such as dot area percentage in each ink-covered area according to film, forme or other carriers, and sets up the funtcional relationship between dot area percentage and the black key aperture, and then the China ink amount that goes up of each ink-covered area on the initial setting printing machine.The ink pre-setting technology is the representative art that digitizing technique gets into the print production link, is one of gordian technique important in the digital printing workflow, to printing quality and printing efficiency decisive role.

Traditional ink pre-setting technology based on the BP neural network algorithm does not possess the incremental learning ability (on-line study) to the training sample data; And a little less than the generalization ability; Can destroy the pattern that network has been remembered during to new samples data training study, cause the black key aperture of network to predict the outcome not accurate enough.Use Fuzzy ART (fuzzy self-adaption resonance neural network) and BP (Back Propagation) neural network for addressing this problem this paper, two kinds of neural networks are carried out integrated application form a kind of Fuzzy ART-BP composite nerve network with incremental learning ability.Yet; Can save the adjustment time of printing machine start effectively based on the ink pre-setting technology of Fuzzy ART-BP composite nerve network algorithm; Reduce paper, printing ink waste that start is prepared; Reduce the labour intensity of press operator simultaneously, overcome the press operator rule of thumb and regulated and control drawback that the China ink amount brings can not realize incremental learning during to new samples data learning training with the BP neural network drawback, improved the technological precision of prediction of ink pre-setting.

Summary of the invention

The present invention relates to a kind of black key aperture Forecasting Methodology with incremental learning ability.Domestic press is in order further to improve printing quality and production efficiency; Some enterprises have introduced external various ink pre-setting system in succession; Before start, adjust the black key of printing machine in advance, but the ink pre-setting system is unsatisfactory in practical application, do not consider between black key influence each other, the influence of printing condition; Can not realize incremental learning to the learning training of new samples data simultaneously, cause the result of use that does not reach expection.

Method of the present invention is that the printed sheet that meets the GB printing standard with field density (evenly and do not have the surface color density that prints) blankly is a training sample; Utilization Fuzzy ART-BP composite nerve network has the tutor to train to training sample and the sample of not training and training is carried out black key aperture prediction, and wherein the BP neural network is selected 3 layers (input layer, hidden layer and output layers) for use.At first with client's original copy digitizing; Promptly obtain complete space of a whole page data message; Transform the layout information that produces low resolution, i.e. dot area percentage then through producing lattice information after RIP (raster image processor) rasterization process, and with the lattice information that produces through software.To print the input raw data of field condition (comprising scene temperature, on-the-spot humidity, the printing machine rotating speed) dot area percentage corresponding as Fuzzy ART-BP composite nerve network with the ink-covered area; And to the input raw data carry out [0; 1] delivers to the input layer of Fuzzy ART-BP composite nerve network after the normalization; With the output layer raw data of the corresponding black key aperture of training sample as Fuzzy ART-BP composite nerve network; Also the output raw data is carried out delivering to after [0,1] normalization is handled the output layer of Fuzzy ART-BP composite nerve network simultaneously, wherein the hidden layer node number of BP neural network is set in 21-35; Carry out the optimum adjustment according to training result, finally choose hidden layer node several 23.

Call Fuzzy ART-BP algorithm routine to qualified actual printed sheet training study; Thereby set up the Nonlinear Mapping relation of printed sheet picture and text numerical information and printing condition (scene temperature, on-the-spot humidity, printing machine rotating speed) and black key aperture; Fuzzy ART neural network is at first carried out the self-adaption cluster operation to printed sheet picture and text numerical information and printing condition; Carry out BP neural metwork training study to sorted data, when BP neural network convergence error during less than 10e-4, the BP neural network finally restrains; Preserve the weights and the threshold value of Nonlinear Mapping, and the weights of Fuzzy ART neural network are to database.If when the convergence error of BP neural network is not less than 10e-4, continue training sample iterative computation, until convergence error less than 10e-4.Fuzzy ART-BP composite nerve network with training is predicted not training sample, and black key aperture predicted value is sent in the real-time data base of printing machine through network or storage medium, and then by control desk control press printing.This method can effectively shorten the setup time of starting shooting, and improves printing efficiency and quality, realizes the new samples data are carried out the study of increment type.

Fuzzy ART-BP composite nerve network algorithm when training be adopt tutor's mode of learning arranged; Its main thought is: the sample vector of input study; At first carry out self-adaptation competition cluster through Fuzzy ART neural network; Revise the weights of the affiliated classification of input vector; Use back-propagation algorithm BP that the weights of BP neural network are carried out adjustment training repeatedly with threshold value with regard to classification under the input vector then, make that vector and the desired output vector of actual output are approaching as much as possible, when the error sum of squares of network output layer is trained completion during less than the error of appointment; Preserve the weights and the threshold value of network this moment, the study of Fuzzy ART-BP composite nerve network training finishes.

In order to use the digitizing ink pre-setting technology based on Fuzzy ART-BP composite nerve network algorithm better, this paper has provided Fuzzy ART-BP composite nerve network algorithm detailed steps, the concrete introduction as follows:

(1) parameter introduction

Input vector P=(a of Fuzzy ART-BP composite nerve network ₁, a ₂... A _n) ^T, 20≤n≤30; Be dot area percentage, scene temperature, on-the-spot humidity and the normalized data of printing machine rotating speed of ink-covered area.

Fuzzy ART-BP composite nerve network desired output vector T=(s ₁, s ₂... S _q) ^T, q=n-3; The normalized data of Jimo key aperture;

Hidden layer unit input vector S=(s ₁, s ₂... S _p) ^T, p gets 21-35; Output vector B=(b ₁, b ₂... B _p) ^T, p gets 21-35.

Output layer unit input vector L=(l ₁, l ₂... L _q) ^T, q=n-3; Actual output vector C=(c ₁, c ₂... C _q) ^T, q=n-3.

The connection power W of input layer to hidden layer _I1j1,

i1=1,2,…,p，j1=1,2,…,n。

The connection power V of hidden layer to output layer _Ti1,

i1=1,2,…,p，t=1,2,…,q。

The output threshold value θ of each unit of hidden layer _I1, ${\mathrm{\θ}}_{i1}=\left[\begin{array}{c}{\mathrm{\θ}}_1\\ {\mathrm{\θ}}_2\\ .\\ .\\ .\\ {\mathrm{\θ}}_{i1}\end{array}\right],$ I1=1,2 ..., p.

The output threshold value y of each unit of output layer _t, $y_t=\left[\begin{array}{c}{y}_1\\ {\mathrm{<figure-callout\; id="2"\; label="y"\; filenames="HDA00001677996300031.png,HDA00001677996300032.png"\; state="\{\{state\}\}">y</figure-callout>}}_2\\ .\\ .\\ .\\ y_t\end{array}\right],$ T=1,2 ..., q.

α is the factor of momentum of BP neural network, 0 < α < 1.

β is the learning rate of BP neural network, 0 < β < 1.

(2) the concrete learning process of Fuzzy ART-BP composite nerve network

1) sample of choosing one group of input, actual output is to P=(a ₁, a ₂... A _n) ^T, T=(s ₁, s ₂... S _q) ^T, and sample handled normalization, offer the input layer and the output layer of Fuzzy ART-BP composite nerve network then respectively.

2) to the input vector P=(a of the input layer of Fuzzy ART-BP composite nerve network ₁, a ₂... A _n) ^TCarry out the input vector I of complementary operation as Fuzzy ART neural network, can suppress the hyperplasia of Fuzzy ART neural network classification so effectively, concrete operations are suc as formula shown in (1-1).

I=(a ₁,a _2,，a ₃，L,a _n,1-a ₁，1-a ₂,1-a ₃，L,1-a _n) （1-1）

2) principle of " the victor is a king " through Fuzzy ART neural network is classified to input vector I then.

3) set up the BP neural network of respective classes numbering with regard to classification numbering J under the input vector I, and, promptly connect weights W to each to the BP neural network initialization of classification J _I1j1And V _Ti1, threshold value θ _I1With y _tGive the random number in (1,1), set the convergence error ε of BP neural network algorithm at last;

4) will with the sample of the corresponding original input vector P of input vector I that obtains after the complementary operation and its desired output to P=(a ₁, a ₂... A _n) ^T, T=(s ₁, s ₂... S _q) ^TBe input to the BP neural network of classification J after the normalization.

5) with input layer sample data P=(a ₁, a ₂... A _n) ^T, connect weights W _I1j1With threshold value θ _I1Calculate the input s of each unit of hidden layer _I1, use s then _I1Calculate the output b of each unit of hidden layer through transport function _I1, transport function is selected the sigmoid function for use, and its form is:

s _I1Computing formula is suc as formula shown in (1-2), b _I1Computing formula is suc as formula shown in (1-3).

$s_{i1}=\underset{\mathrm{<figure-callout\; id="1"\; label="j"\; filenames="HDA00001677996300032.png,HDA00001677996300042.png"\; state="\{\{state\}\}">j</figure-callout>}1=1}{\overset{n}{\mathrm{\&Sigma;}}}(w_{i1j1}{a}_{j1}-{\mathrm{\&theta;}}_{i1})---(1-2)$

$b_{i1}=\left[\begin{array}{c}f\left(s_1\right)\\ f\left(s_2\right)\\ .\\ .\\ .\\ f\left(s_{i1}\right)\end{array}\right]---(1-3)$

6) utilize the output b of hidden layer _I1, connect power V _Ti1Threshold value y _t, calculate the output l of each unit of output layer _t, shown in (1-4), utilize transport function to calculate the response c of each unit of output layer then _t, transport function is selected the sigmoid function for use, and its form is:

Shown in (1-5).

$l_t=\underset{i1=1}{\overset{p}{\mathrm{\&Sigma;}}}(V_{\mathrm{ti}1}{b}_{i1}-y_t)---(1-4)$

$c_t=\left[\begin{array}{c}f\left(l_1\right)\\ f\left(l_2\right)\\ .\\ .\\ .\\ f\left(l_t\right)\end{array}\right]---(1-5)$

7) utilize the actual output C=(c of Fuzzy ART-BP composite nerve network ₁, c ₂... C _q) ^TDesired output T=(s with network ₁, s ₂... S _q) ^T, error of calculation E, shown in (1-6), if E less than the convergence error ε that sets, Fuzzy ART-BP composite nerve network convergence then, finishing iteration is also preserved weights W _I1j1, V _Ti1With threshold value θ _I1, y _tOtherwise whether the continuation step 8) revises weights and threshold matrix continued error in judgement E less than the convergence error of setting.

$E=\frac{1}{2}\underset{t=1}{\overset{q}{\mathrm{\&Sigma;}}}{(s_t-c_t)}^2---(1-6)$

8) utilize the actual output C=(c of Fuzzy ART-BP composite nerve network ₁, c ₂... C _q) ^T, the desired output T=(s of network ₁, s ₂... S _q) ^T, the vague generalization error d of each unit of calculating output layer _t, shown in (1-7).

d _t=(s _t-c _t)·c _t·(1-c _t) （1-7）

9) utilize the vague generalization error d of each unit of output layer of the BP network of Fuzzy ART-BP composite nerve network _t, connect power V _Ti1Output b with hidden layer _I1Calculate the vague generalization error e of each unit of hidden layer _I1, shown in (1-8).

$e_{i1}=[\underset{t=1}{\overset{q}{\mathrm{\&Sigma;}}}{d}_t\&CenterDot;V_{\mathrm{ti}1}]\&CenterDot;b_{i1}\&CenterDot;(1-b_{i1})---(1-8)$

10) utilize the vague generalization error d of each unit of output layer _tOutput b with each unit of hidden layer _I1Revise and connect power V _Ti1, shown in (1-9), threshold value y _tModification is suc as formula shown in (1-10).

V _ti1′=αV _ti1+βd _tb _i1 （1-9）

y _t′=y _t+βd _t （1-10）

11) utilize the vague generalization error e of each unit of hidden layer _I1, the input P=(a of each unit of input layer ₁, a ₂... A _n) revise and connect weights W _I1j1, shown in (1-11), threshold value θ _I1Modification is suc as formula shown in (1-12).

w _i1j1′=αw _i1j1+βe _i1a _j1 （1-11）

θ _i1′=θ _i1+βe _i1 （1-12）

12) with input layer sample data P=(a ₁, a ₂... A _n) ^T, connect weights W _I1j1' and threshold value θ _I1The input s of each unit of ' calculating hidden layer _I1', suc as formula (1-13), use s then _I1' calculate the output b of each unit of hidden layer through transport function _I1', suc as formula (1-14), transport function is selected the sigmoid function for use, and its form is:

${s_{i1}}^{\&prime;}=\underset{\mathrm{<figure-callout\; id="1"\; label="j"\; filenames="HDA00001677996300032.png,HDA00001677996300042.png"\; state="\{\{state\}\}">j</figure-callout>}1=1}{\overset{n}{\mathrm{\&Sigma;}}}({w_{i1\mathrm{<figure-callout\; id="1"\; label="j"\; filenames="HDA00001677996300032.png,HDA00001677996300042.png"\; state="\{\{state\}\}">j</figure-callout>}1}}^{\&prime;}{a}_{j1}-{{\mathrm{\&theta;}}_{i1}}^{\&prime;})---(1-13)$

${b_{i1}}^{\&prime;}=\left[\begin{array}{c}f\left({{\mathrm{<figure-callout\; id="1"\; label="s"\; filenames="HDA00001677996300032.png,HDA00001677996300042.png"\; state="\{\{state\}\}">s</figure-callout>}}_1}^{\&prime;}\right)\\ f\left({{\mathrm{<figure-callout\; id="2"\; label="s"\; filenames="HDA00001677996300031.png,HDA00001677996300032.png"\; state="\{\{state\}\}">s</figure-callout>}}_2}^{\&prime;}\right)\\ .\\ .\\ .\\ f\left({s_{i1}}^{\&prime;}\right)\end{array}\right]---(1-14)$

13) utilize the output b of hidden layer _I1', connect power V _Ti1' threshold value y _t'.Calculate the output l of each unit of output layer _t', suc as formula (1-15), utilize transport function to calculate the response c of each unit of output layer then _t', transport function is selected the sigmoid function for use, and its form is: Suc as formula (1-16).

${l_t}^{\&prime;}=\underset{i1=1}{\overset{p}{\mathrm{\&Sigma;}}}({{\mathrm{<figure-callout\; id="1"\; label="V\; ti"\; filenames="HDA00001677996300032.png,HDA00001677996300042.png"\; state="\{\{state\}\}">V</figure-callout>}}_{\mathrm{ti}}}^{\&prime;}{{\mathrm{<figure-callout\; id="1"\; label="b\; i"\; filenames="HDA00001677996300032.png,HDA00001677996300042.png"\; state="\{\{state\}\}">b</figure-callout>}}_i}^{\&prime;}-{y_t}^{\&prime;})---(1-15)$

${c_t}^{\&prime;}=\left[\begin{array}{c}f\left({{\mathrm{<figure-callout\; id="1"\; label="l"\; filenames="HDA00001677996300032.png,HDA00001677996300042.png"\; state="\{\{state\}\}">l</figure-callout>}}_1}^{\&prime;}\right)\\ f\left({{\mathrm{<figure-callout\; id="2"\; label="l"\; filenames="HDA00001677996300031.png,HDA00001677996300032.png"\; state="\{\{state\}\}">l</figure-callout>}}_2}^{\&prime;}\right)\\ .\\ .\\ .\\ f\left({l_t}^{\&prime;}\right)\end{array}\right]---(1-16)$

14) utilize the actual output C ' of network=(c ' ₁, c ' ₂... C ' _q) ^T, the desired output T=(s of network ₁, s ₂... S _q) ^T, calculating target function (error) E ', suc as formula (1-17), if E ' less than the convergence error ε that sets, network convergence then, finishing iteration is also preserved weights and threshold value; Otherwise return step 8), revise weights and threshold matrix, continue iterative computation.

$E^{\&prime;}=\frac{1}{2}\underset{t=1}{\overset{q}{\mathrm{\&Sigma;}}}{(s_t-{c_t}^{\&prime;})}^2---(1-17)$

15) choose next training sample to offering Fuzzy ART-BP composite nerve network, continue step 1)-15), up to all training samples to trained.

Said process can be used the learning training flowcharting of figure (2) Fuzzy ART-BP composite nerve network.

Description of drawings

Fig. 1 ink pre-setting technology synoptic diagram.

The learning training process flow diagram of Fig. 2 Fuzzy ART-BP composite nerve network.

The cyan figure that predicts the outcome in the prediction of Fig. 3 China ink key aperture.

The cyan figure that predicts the outcome in the checking of Fig. 4 China ink key aperture.

Fig. 5 trains the visual representation figure of network prediction difference of corresponding each ink-covered area of network of 100 groups of data.Fig. 6 trains the visual representation figure of network prediction difference of corresponding each ink-covered area of network of 150 groups of data.

Embodiment

At first, promptly obtain complete space of a whole page data message,, and the lattice information that produces transformed through software produce layout information-dot area percentage then through producing lattice information after RIP (raster image processor) rasterization process with the digitizing of training sample original copy.The on-the-spot relative humidity of printing is 30%, and the printing machine rotating speed is 4000/hour, and temperature is 25 ℃; Because each data unit of gathering is inconsistent, in order to accelerate the convergence of training network, thereby must carry out [0 to data; 1] normalization is handled; Top three conditions are done the normalization processing to be respectively: 0.3,0.4,0.25.The full lattice of used ink fountain are 100, and the actual black key aperture of standby gets final product divided by 100 when carrying out the normalization processing, and the value of dot area percentage is also handled without normalization and directly got fractional value between [0-100%].Testing ground temperature, testing ground humidity, printing machine rotating speed are the principal elements of the black key aperture of influence; Also have between adjacent black key and influence each other; So can not be single with of the input of certain any dot area percentage as the BP neural network; So confirm that the neuron number of input layer is 23, comprise the dot area percentage of scene temperature, on-the-spot humidity, printing machine rotating speed and 20 ink-covered areas; Confirming of output layer node number: the output layer node is followed successively by 20 black key apertures that the ink-covered area dot area percentage is corresponding; The hidden layer node number is set in 23.Learning rate β=0.4, factor of momentum α=0.9, the error ε of expectation=10e-4; Learning rate η=0.15 of FuzzyART neural network, alarm threshold ρ=0.95 of Fuzzy ART neural network.

Fuzzy ART-BP composite nerve network algorithm and BP neural network algorithm have remarkable advantages can carry out incremental learning to the new samples data exactly, carry out the test of precision of prediction to this function below.Because Fuzzy ART-BP composite nerve network algorithm has very strong analysis ability to more data, it predicts the outcome and BP neural network prediction result is significantly increased.Therefore, on existing data basis, utilize Monte Carlo simulation to simulate more data, offer Fuzzy ART-BP composite nerve network and carry out training study.It is as shown in the table that wherein the back is handled in the existing partial data normalization of cyan:

Network fan-in factor certificate:

The network output data:

Under identical printing environment; Be under printing machine rotating speed, temperature and the air humidity same case, (what CMYK was that the initial of 4 kinds of printing-ink titles: cyan Cyan, carmetta Magenta, yellow Yellow, K get is last letter of black to four looks that the dot area percentage that same printed sheet is identical is corresponding.) black key aperture also inequality, so will set up neural network respectively according to four looks, and carry out the training study and the prediction of neural network separately to existing data and emulated data.

The training method of employing from low capacity to the transition of high capacity training sample has the tutor to learn to carrying out network based on the ink pre-setting technology of Fuzzy ART-BP composite nerve network; Just the detailed step according to Fuzzy ART-BP composite nerve network algorithm in the summary of the invention has tutor's training study, verifies the superiority of the ink pre-setting technology of itself and BP neural network.At first choose 100 groups of samples to data respectively to carrying out e-learning based on BP neural network ink pre-setting technology with based on Fuzzy ART-BP composite nerve network ink pre-setting technology, the convergence error of treating its network reaches the weights of preserving each automatic network behind the convergence error ε=10e-4 of setting and threshold value etc.; Get one group of data from sample to the data then and be input to two kinds of ink pre-setting systems that trained, carry out black key aperture prediction, wherein cyan predicts the outcome as shown in Figure 3.At last, choose other 50 groups of samples once more data are learnt the ink pre-setting technology of two neural networks, get the predictive ability that same group of data are verified network once more, the cyan prediction effect is as shown in Figure 4.

In order more intuitively to find out the precision of prediction of Fuzzy ART-BP neural network algorithm and BP neural network algorithm; Actual value and two kinds of neural network prediction values that this paper prints the black key aperture when qualified with actual printing are subtracted each other the difference that obtains and are called the network prediction difference, the visual representation of the network prediction difference of corresponding each ink-covered area of network that is 100 groups of data of training shown in Figure 5.The visual representation that is the network prediction difference of corresponding each ink-covered area of network of training 150 groups of data shown in Figure 6.

From Fig. 3 to Fig. 6, can find out, be very considerable based on the prediction effect of the ink pre-setting of Fuzzy ART-BP composite nerve network technology, prediction difference all is controlled at basically ± and about 2%.Along with training sample data amount increases; The prediction difference variation range of BP neural network strengthens, partly reach ± 3%, explain that the sample of new study has destroyed the pattern that network has been remembered; The ability that does not possess incremental learning causes the error that predicts the outcome amid a sharp increase.Compare with the BP neural network; Fuzzy ART-BP composite nerve network can carry out adaptive classification study to the input data, carries out the BP neural network learning with regard to sorted pattern then, makes network more targeted to training mode; Reduce as much as possible network memory pattern degree is destroyed in new model study back; Predict the outcome also be controlled at basically ± about 2%, improved the network prediction accuracy, make network possess the ability of incremental learning.

Claims (1)

1. black key aperture Forecasting Methodology with incremental learning ability may further comprise the steps:

1) with field density promptly evenly and not have the printed sheet that the surface color density that prints meets the GB printing standard blankly be training sample; To four look CMYK; Be cyan Cyan, magenta Magenta, yellow Yellow and black Black, set up corresponding Fuzzy ART-BP composite nerve network respectively;

2) with printing condition, i.e. the dot area percentage normalization of scene temperature, on-the-spot humidity, printing machine rotating speed and 20-30 ink-covered area is handled the back as Fuzzy ART-BP composite nerve network input layer input data; The normalization of China ink key aperture is handled the back as Fuzzy ART-BP composite nerve network output layer input data; Input layer input data are through Fuzzy ART self-adaption cluster, and the weights of preserving type set up the BP neural network to certain type of data after the cluster then to database, regulate the hidden layer node number of suitable BP neural network, are set at 21-35; Utilization input layer, hidden layer and 3 layers of BP neural network algorithm of output layer procedural training module have tutor's training study to training sample; When the convergence error of BP neural network during less than 10e-4; The BP neural network finally restrains, and the weights of preservation BP neural network algorithm Nonlinear Mapping and threshold value are to database;

3) for training sample can be with the dot area percentage and the printing condition of printed sheet; Be scene temperature, on-the-spot humidity and printing machine rotating speed; Be defeated by the prediction module of Fuzzy ART-BP composite nerve network algorithm; Weights that the utilization of Fuzzy ART-BP composite nerve network algorithm program has been stored and threshold value are carried out prediction and calculation to scene temperature, on-the-spot humidity, printing machine rotating speed and the dot area percentage of input, thereby dope the not corresponding black key aperture of dot area percentage of training sample; Black key aperture predicted value is sent to the printing machine control desk through network or storage medium, and control desk receives data and controls corresponding black key automatically and is inked to printing machine completion printing.

Images